Show More
@@ -1,284 +1,275 b'' | |||||
1 | ================= |
|
1 | ================= | |
2 | Parallel examples |
|
2 | Parallel examples | |
3 | ================= |
|
3 | ================= | |
4 |
|
4 | |||
5 | .. note:: |
|
|||
6 |
|
||||
7 | Performance numbers from ``IPython.kernel``, not new ``IPython.parallel``. |
|
|||
8 |
|
||||
9 | In this section we describe two more involved examples of using an IPython |
|
5 | In this section we describe two more involved examples of using an IPython | |
10 | cluster to perform a parallel computation. In these examples, we will be using |
|
6 | cluster to perform a parallel computation. In these examples, we will be using | |
11 | IPython's "pylab" mode, which enables interactive plotting using the |
|
7 | IPython's "pylab" mode, which enables interactive plotting using the | |
12 | Matplotlib package. IPython can be started in this mode by typing:: |
|
8 | Matplotlib package. IPython can be started in this mode by typing:: | |
13 |
|
9 | |||
14 | ipython --pylab |
|
10 | ipython --pylab | |
15 |
|
11 | |||
16 | at the system command line. |
|
12 | at the system command line. | |
17 |
|
13 | |||
18 | 150 million digits of pi |
|
14 | 150 million digits of pi | |
19 | ======================== |
|
15 | ======================== | |
20 |
|
16 | |||
21 | In this example we would like to study the distribution of digits in the |
|
17 | In this example we would like to study the distribution of digits in the | |
22 | number pi (in base 10). While it is not known if pi is a normal number (a |
|
18 | number pi (in base 10). While it is not known if pi is a normal number (a | |
23 | number is normal in base 10 if 0-9 occur with equal likelihood) numerical |
|
19 | number is normal in base 10 if 0-9 occur with equal likelihood) numerical | |
24 | investigations suggest that it is. We will begin with a serial calculation on |
|
20 | investigations suggest that it is. We will begin with a serial calculation on | |
25 | 10,000 digits of pi and then perform a parallel calculation involving 150 |
|
21 | 10,000 digits of pi and then perform a parallel calculation involving 150 | |
26 | million digits. |
|
22 | million digits. | |
27 |
|
23 | |||
28 | In both the serial and parallel calculation we will be using functions defined |
|
24 | In both the serial and parallel calculation we will be using functions defined | |
29 | in the :file:`pidigits.py` file, which is available in the |
|
25 | in the :file:`pidigits.py` file, which is available in the | |
30 | :file:`docs/examples/parallel` directory of the IPython source distribution. |
|
26 | :file:`docs/examples/parallel` directory of the IPython source distribution. | |
31 | These functions provide basic facilities for working with the digits of pi and |
|
27 | These functions provide basic facilities for working with the digits of pi and | |
32 | can be loaded into IPython by putting :file:`pidigits.py` in your current |
|
28 | can be loaded into IPython by putting :file:`pidigits.py` in your current | |
33 | working directory and then doing: |
|
29 | working directory and then doing: | |
34 |
|
30 | |||
35 | .. sourcecode:: ipython |
|
31 | .. sourcecode:: ipython | |
36 |
|
32 | |||
37 | In [1]: run pidigits.py |
|
33 | In [1]: run pidigits.py | |
38 |
|
34 | |||
39 | Serial calculation |
|
35 | Serial calculation | |
40 | ------------------ |
|
36 | ------------------ | |
41 |
|
37 | |||
42 | For the serial calculation, we will use `SymPy <http://www.sympy.org>`_ to |
|
38 | For the serial calculation, we will use `SymPy <http://www.sympy.org>`_ to | |
43 | calculate 10,000 digits of pi and then look at the frequencies of the digits |
|
39 | calculate 10,000 digits of pi and then look at the frequencies of the digits | |
44 | 0-9. Out of 10,000 digits, we expect each digit to occur 1,000 times. While |
|
40 | 0-9. Out of 10,000 digits, we expect each digit to occur 1,000 times. While | |
45 | SymPy is capable of calculating many more digits of pi, our purpose here is to |
|
41 | SymPy is capable of calculating many more digits of pi, our purpose here is to | |
46 | set the stage for the much larger parallel calculation. |
|
42 | set the stage for the much larger parallel calculation. | |
47 |
|
43 | |||
48 | In this example, we use two functions from :file:`pidigits.py`: |
|
44 | In this example, we use two functions from :file:`pidigits.py`: | |
49 | :func:`one_digit_freqs` (which calculates how many times each digit occurs) |
|
45 | :func:`one_digit_freqs` (which calculates how many times each digit occurs) | |
50 | and :func:`plot_one_digit_freqs` (which uses Matplotlib to plot the result). |
|
46 | and :func:`plot_one_digit_freqs` (which uses Matplotlib to plot the result). | |
51 | Here is an interactive IPython session that uses these functions with |
|
47 | Here is an interactive IPython session that uses these functions with | |
52 | SymPy: |
|
48 | SymPy: | |
53 |
|
49 | |||
54 | .. sourcecode:: ipython |
|
50 | .. sourcecode:: ipython | |
55 |
|
51 | |||
56 | In [7]: import sympy |
|
52 | In [7]: import sympy | |
57 |
|
53 | |||
58 | In [8]: pi = sympy.pi.evalf(40) |
|
54 | In [8]: pi = sympy.pi.evalf(40) | |
59 |
|
55 | |||
60 | In [9]: pi |
|
56 | In [9]: pi | |
61 | Out[9]: 3.141592653589793238462643383279502884197 |
|
57 | Out[9]: 3.141592653589793238462643383279502884197 | |
62 |
|
58 | |||
63 | In [10]: pi = sympy.pi.evalf(10000) |
|
59 | In [10]: pi = sympy.pi.evalf(10000) | |
64 |
|
60 | |||
65 | In [11]: digits = (d for d in str(pi)[2:]) # create a sequence of digits |
|
61 | In [11]: digits = (d for d in str(pi)[2:]) # create a sequence of digits | |
66 |
|
62 | |||
67 | In [12]: run pidigits.py # load one_digit_freqs/plot_one_digit_freqs |
|
63 | In [12]: run pidigits.py # load one_digit_freqs/plot_one_digit_freqs | |
68 |
|
64 | |||
69 | In [13]: freqs = one_digit_freqs(digits) |
|
65 | In [13]: freqs = one_digit_freqs(digits) | |
70 |
|
66 | |||
71 | In [14]: plot_one_digit_freqs(freqs) |
|
67 | In [14]: plot_one_digit_freqs(freqs) | |
72 | Out[14]: [<matplotlib.lines.Line2D object at 0x18a55290>] |
|
68 | Out[14]: [<matplotlib.lines.Line2D object at 0x18a55290>] | |
73 |
|
69 | |||
74 | The resulting plot of the single digit counts shows that each digit occurs |
|
70 | The resulting plot of the single digit counts shows that each digit occurs | |
75 | approximately 1,000 times, but that with only 10,000 digits the |
|
71 | approximately 1,000 times, but that with only 10,000 digits the | |
76 | statistical fluctuations are still rather large: |
|
72 | statistical fluctuations are still rather large: | |
77 |
|
73 | |||
78 | .. image:: figs/single_digits.* |
|
74 | .. image:: figs/single_digits.* | |
79 |
|
75 | |||
80 | It is clear that to reduce the relative fluctuations in the counts, we need |
|
76 | It is clear that to reduce the relative fluctuations in the counts, we need | |
81 | to look at many more digits of pi. That brings us to the parallel calculation. |
|
77 | to look at many more digits of pi. That brings us to the parallel calculation. | |
82 |
|
78 | |||
83 | Parallel calculation |
|
79 | Parallel calculation | |
84 | -------------------- |
|
80 | -------------------- | |
85 |
|
81 | |||
86 | Calculating many digits of pi is a challenging computational problem in itself. |
|
82 | Calculating many digits of pi is a challenging computational problem in itself. | |
87 | Because we want to focus on the distribution of digits in this example, we |
|
83 | Because we want to focus on the distribution of digits in this example, we | |
88 | will use pre-computed digit of pi from the website of Professor Yasumasa |
|
84 | will use pre-computed digit of pi from the website of Professor Yasumasa | |
89 | Kanada at the University of Tokyo (http://www.super-computing.org). These |
|
85 | Kanada at the University of Tokyo (http://www.super-computing.org). These | |
90 | digits come in a set of text files (ftp://pi.super-computing.org/.2/pi200m/) |
|
86 | digits come in a set of text files (ftp://pi.super-computing.org/.2/pi200m/) | |
91 | that each have 10 million digits of pi. |
|
87 | that each have 10 million digits of pi. | |
92 |
|
88 | |||
93 | For the parallel calculation, we have copied these files to the local hard |
|
89 | For the parallel calculation, we have copied these files to the local hard | |
94 | drives of the compute nodes. A total of 15 of these files will be used, for a |
|
90 | drives of the compute nodes. A total of 15 of these files will be used, for a | |
95 | total of 150 million digits of pi. To make things a little more interesting we |
|
91 | total of 150 million digits of pi. To make things a little more interesting we | |
96 | will calculate the frequencies of all 2 digits sequences (00-99) and then plot |
|
92 | will calculate the frequencies of all 2 digits sequences (00-99) and then plot | |
97 | the result using a 2D matrix in Matplotlib. |
|
93 | the result using a 2D matrix in Matplotlib. | |
98 |
|
94 | |||
99 | The overall idea of the calculation is simple: each IPython engine will |
|
95 | The overall idea of the calculation is simple: each IPython engine will | |
100 | compute the two digit counts for the digits in a single file. Then in a final |
|
96 | compute the two digit counts for the digits in a single file. Then in a final | |
101 | step the counts from each engine will be added up. To perform this |
|
97 | step the counts from each engine will be added up. To perform this | |
102 | calculation, we will need two top-level functions from :file:`pidigits.py`: |
|
98 | calculation, we will need two top-level functions from :file:`pidigits.py`: | |
103 |
|
99 | |||
104 | .. literalinclude:: ../../examples/parallel/pi/pidigits.py |
|
100 | .. literalinclude:: ../../examples/parallel/pi/pidigits.py | |
105 | :language: python |
|
101 | :language: python | |
106 | :lines: 47-62 |
|
102 | :lines: 47-62 | |
107 |
|
103 | |||
108 | We will also use the :func:`plot_two_digit_freqs` function to plot the |
|
104 | We will also use the :func:`plot_two_digit_freqs` function to plot the | |
109 | results. The code to run this calculation in parallel is contained in |
|
105 | results. The code to run this calculation in parallel is contained in | |
110 | :file:`docs/examples/parallel/parallelpi.py`. This code can be run in parallel |
|
106 | :file:`docs/examples/parallel/parallelpi.py`. This code can be run in parallel | |
111 | using IPython by following these steps: |
|
107 | using IPython by following these steps: | |
112 |
|
108 | |||
113 |
1. Use :command:`ipcluster` to start 15 engines. We used |
|
109 | 1. Use :command:`ipcluster` to start 15 engines. We used 16 cores of an SGE linux | |
114 | core CPUs) cluster with hyperthreading enabled which makes the 8 cores |
|
110 | cluster (1 controller + 15 engines). | |
115 | looks like 16 (1 controller + 15 engines) in the OS. However, the maximum |
|
|||
116 | speedup we can observe is still only 8x. |
|
|||
117 | 2. With the file :file:`parallelpi.py` in your current working directory, open |
|
111 | 2. With the file :file:`parallelpi.py` in your current working directory, open | |
118 | up IPython in pylab mode and type ``run parallelpi.py``. This will download |
|
112 | up IPython in pylab mode and type ``run parallelpi.py``. This will download | |
119 | the pi files via ftp the first time you run it, if they are not |
|
113 | the pi files via ftp the first time you run it, if they are not | |
120 | present in the Engines' working directory. |
|
114 | present in the Engines' working directory. | |
121 |
|
115 | |||
122 |
When run on our |
|
116 | When run on our 16 cores, we observe a speedup of 14.2x. This is slightly | |
123 |
less than linear scaling ( |
|
117 | less than linear scaling (16x) because the controller is also running on one of | |
124 | the cores. |
|
118 | the cores. | |
125 |
|
119 | |||
126 | To emphasize the interactive nature of IPython, we now show how the |
|
120 | To emphasize the interactive nature of IPython, we now show how the | |
127 | calculation can also be run by simply typing the commands from |
|
121 | calculation can also be run by simply typing the commands from | |
128 | :file:`parallelpi.py` interactively into IPython: |
|
122 | :file:`parallelpi.py` interactively into IPython: | |
129 |
|
123 | |||
130 | .. sourcecode:: ipython |
|
124 | .. sourcecode:: ipython | |
131 |
|
125 | |||
132 | In [1]: from IPython.parallel import Client |
|
126 | In [1]: from IPython.parallel import Client | |
133 |
|
127 | |||
134 | # The Client allows us to use the engines interactively. |
|
128 | # The Client allows us to use the engines interactively. | |
135 | # We simply pass Client the name of the cluster profile we |
|
129 | # We simply pass Client the name of the cluster profile we | |
136 | # are using. |
|
130 | # are using. | |
137 | In [2]: c = Client(profile='mycluster') |
|
131 | In [2]: c = Client(profile='mycluster') | |
138 | In [3]: view = c.load_balanced_view() |
|
132 | In [3]: v = c[:] | |
139 |
|
133 | |||
140 | In [3]: c.ids |
|
134 | In [3]: c.ids | |
141 | Out[3]: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] |
|
135 | Out[3]: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] | |
142 |
|
136 | |||
143 | In [4]: run pidigits.py |
|
137 | In [4]: run pidigits.py | |
144 |
|
138 | |||
145 | In [5]: filestring = 'pi200m.ascii.%(i)02dof20' |
|
139 | In [5]: filestring = 'pi200m.ascii.%(i)02dof20' | |
146 |
|
140 | |||
147 | # Create the list of files to process. |
|
141 | # Create the list of files to process. | |
148 | In [6]: files = [filestring % {'i':i} for i in range(1,16)] |
|
142 | In [6]: files = [filestring % {'i':i} for i in range(1,16)] | |
149 |
|
143 | |||
150 | In [7]: files |
|
144 | In [7]: files | |
151 | Out[7]: |
|
145 | Out[7]: | |
152 | ['pi200m.ascii.01of20', |
|
146 | ['pi200m.ascii.01of20', | |
153 | 'pi200m.ascii.02of20', |
|
147 | 'pi200m.ascii.02of20', | |
154 | 'pi200m.ascii.03of20', |
|
148 | 'pi200m.ascii.03of20', | |
155 | 'pi200m.ascii.04of20', |
|
149 | 'pi200m.ascii.04of20', | |
156 | 'pi200m.ascii.05of20', |
|
150 | 'pi200m.ascii.05of20', | |
157 | 'pi200m.ascii.06of20', |
|
151 | 'pi200m.ascii.06of20', | |
158 | 'pi200m.ascii.07of20', |
|
152 | 'pi200m.ascii.07of20', | |
159 | 'pi200m.ascii.08of20', |
|
153 | 'pi200m.ascii.08of20', | |
160 | 'pi200m.ascii.09of20', |
|
154 | 'pi200m.ascii.09of20', | |
161 | 'pi200m.ascii.10of20', |
|
155 | 'pi200m.ascii.10of20', | |
162 | 'pi200m.ascii.11of20', |
|
156 | 'pi200m.ascii.11of20', | |
163 | 'pi200m.ascii.12of20', |
|
157 | 'pi200m.ascii.12of20', | |
164 | 'pi200m.ascii.13of20', |
|
158 | 'pi200m.ascii.13of20', | |
165 | 'pi200m.ascii.14of20', |
|
159 | 'pi200m.ascii.14of20', | |
166 | 'pi200m.ascii.15of20'] |
|
160 | 'pi200m.ascii.15of20'] | |
167 |
|
161 | |||
168 | # download the data files if they don't already exist: |
|
162 | # download the data files if they don't already exist: | |
169 | In [8]: v.map(fetch_pi_file, files) |
|
163 | In [8]: v.map(fetch_pi_file, files) | |
170 |
|
164 | |||
171 | # This is the parallel calculation using the Client.map method |
|
165 | # This is the parallel calculation using the Client.map method | |
172 | # which applies compute_two_digit_freqs to each file in files in parallel. |
|
166 | # which applies compute_two_digit_freqs to each file in files in parallel. | |
173 | In [9]: freqs_all = v.map(compute_two_digit_freqs, files) |
|
167 | In [9]: freqs_all = v.map(compute_two_digit_freqs, files) | |
174 |
|
168 | |||
175 | # Add up the frequencies from each engine. |
|
169 | # Add up the frequencies from each engine. | |
176 | In [10]: freqs = reduce_freqs(freqs_all) |
|
170 | In [10]: freqs = reduce_freqs(freqs_all) | |
177 |
|
171 | |||
178 | In [11]: plot_two_digit_freqs(freqs) |
|
172 | In [11]: plot_two_digit_freqs(freqs) | |
179 | Out[11]: <matplotlib.image.AxesImage object at 0x18beb110> |
|
173 | Out[11]: <matplotlib.image.AxesImage object at 0x18beb110> | |
180 |
|
174 | |||
181 | In [12]: plt.title('2 digit counts of 150m digits of pi') |
|
175 | In [12]: plt.title('2 digit counts of 150m digits of pi') | |
182 | Out[12]: <matplotlib.text.Text object at 0x18d1f9b0> |
|
176 | Out[12]: <matplotlib.text.Text object at 0x18d1f9b0> | |
183 |
|
177 | |||
184 | The resulting plot generated by Matplotlib is shown below. The colors indicate |
|
178 | The resulting plot generated by Matplotlib is shown below. The colors indicate | |
185 | which two digit sequences are more (red) or less (blue) likely to occur in the |
|
179 | which two digit sequences are more (red) or less (blue) likely to occur in the | |
186 | first 150 million digits of pi. We clearly see that the sequence "41" is |
|
180 | first 150 million digits of pi. We clearly see that the sequence "41" is | |
187 | most likely and that "06" and "07" are least likely. Further analysis would |
|
181 | most likely and that "06" and "07" are least likely. Further analysis would | |
188 | show that the relative size of the statistical fluctuations have decreased |
|
182 | show that the relative size of the statistical fluctuations have decreased | |
189 | compared to the 10,000 digit calculation. |
|
183 | compared to the 10,000 digit calculation. | |
190 |
|
184 | |||
191 | .. image:: figs/two_digit_counts.* |
|
185 | .. image:: figs/two_digit_counts.* | |
192 |
|
186 | |||
193 |
|
187 | |||
194 | Parallel options pricing |
|
188 | Parallel options pricing | |
195 | ======================== |
|
189 | ======================== | |
196 |
|
190 | |||
197 | An option is a financial contract that gives the buyer of the contract the |
|
191 | An option is a financial contract that gives the buyer of the contract the | |
198 | right to buy (a "call") or sell (a "put") a secondary asset (a stock for |
|
192 | right to buy (a "call") or sell (a "put") a secondary asset (a stock for | |
199 | example) at a particular date in the future (the expiration date) for a |
|
193 | example) at a particular date in the future (the expiration date) for a | |
200 | pre-agreed upon price (the strike price). For this right, the buyer pays the |
|
194 | pre-agreed upon price (the strike price). For this right, the buyer pays the | |
201 | seller a premium (the option price). There are a wide variety of flavors of |
|
195 | seller a premium (the option price). There are a wide variety of flavors of | |
202 | options (American, European, Asian, etc.) that are useful for different |
|
196 | options (American, European, Asian, etc.) that are useful for different | |
203 | purposes: hedging against risk, speculation, etc. |
|
197 | purposes: hedging against risk, speculation, etc. | |
204 |
|
198 | |||
205 | Much of modern finance is driven by the need to price these contracts |
|
199 | Much of modern finance is driven by the need to price these contracts | |
206 | accurately based on what is known about the properties (such as volatility) of |
|
200 | accurately based on what is known about the properties (such as volatility) of | |
207 | the underlying asset. One method of pricing options is to use a Monte Carlo |
|
201 | the underlying asset. One method of pricing options is to use a Monte Carlo | |
208 | simulation of the underlying asset price. In this example we use this approach |
|
202 | simulation of the underlying asset price. In this example we use this approach | |
209 | to price both European and Asian (path dependent) options for various strike |
|
203 | to price both European and Asian (path dependent) options for various strike | |
210 | prices and volatilities. |
|
204 | prices and volatilities. | |
211 |
|
205 | |||
212 | The code for this example can be found in the :file:`docs/examples/parallel` |
|
206 | The code for this example can be found in the :file:`docs/examples/parallel/options` | |
213 | directory of the IPython source. The function :func:`price_options` in |
|
207 | directory of the IPython source. The function :func:`price_options` in | |
214 |
:file:`mc |
|
208 | :file:`mckernel.py` implements the basic Monte Carlo pricing algorithm using | |
215 | the NumPy package and is shown here: |
|
209 | the NumPy package and is shown here: | |
216 |
|
210 | |||
217 |
.. literalinclude:: ../../examples/parallel/options/mc |
|
211 | .. literalinclude:: ../../examples/parallel/options/mckernel.py | |
218 | :language: python |
|
212 | :language: python | |
219 |
|
213 | |||
220 | To run this code in parallel, we will use IPython's :class:`LoadBalancedView` class, |
|
214 | To run this code in parallel, we will use IPython's :class:`LoadBalancedView` class, | |
221 | which distributes work to the engines using dynamic load balancing. This |
|
215 | which distributes work to the engines using dynamic load balancing. This | |
222 | view is a wrapper of the :class:`Client` class shown in |
|
216 | view is a wrapper of the :class:`Client` class shown in | |
223 | the previous example. The parallel calculation using :class:`LoadBalancedView` can |
|
217 | the previous example. The parallel calculation using :class:`LoadBalancedView` can | |
224 | be found in the file :file:`mcpricer.py`. The code in this file creates a |
|
218 | be found in the file :file:`mcpricer.py`. The code in this file creates a | |
225 | :class:`LoadBalancedView` instance and then submits a set of tasks using |
|
219 | :class:`LoadBalancedView` instance and then submits a set of tasks using | |
226 | :meth:`LoadBalancedView.apply` that calculate the option prices for different |
|
220 | :meth:`LoadBalancedView.apply` that calculate the option prices for different | |
227 | volatilities and strike prices. The results are then plotted as a 2D contour |
|
221 | volatilities and strike prices. The results are then plotted as a 2D contour | |
228 | plot using Matplotlib. |
|
222 | plot using Matplotlib. | |
229 |
|
223 | |||
230 |
.. literalinclude:: ../../examples/parallel/options/mc |
|
224 | .. literalinclude:: ../../examples/parallel/options/mcpricer.py | |
231 | :language: python |
|
225 | :language: python | |
232 |
|
226 | |||
233 | To use this code, start an IPython cluster using :command:`ipcluster`, open |
|
227 | To use this code, start an IPython cluster using :command:`ipcluster`, open | |
234 | IPython in the pylab mode with the file :file:`mckernel.py` in your current |
|
228 | IPython in the pylab mode with the file :file:`mckernel.py` in your current | |
235 | working directory and then type: |
|
229 | working directory and then type: | |
236 |
|
230 | |||
237 | .. sourcecode:: ipython |
|
231 | .. sourcecode:: ipython | |
238 |
|
232 | |||
239 |
In [7]: run mc |
|
233 | In [7]: run mcpricer.py | |
240 | Submitted tasks: [0, 1, 2, ...] |
|
234 | ||
|
235 | Submitted tasks: 30 | |||
241 |
|
236 | |||
242 | Once all the tasks have finished, the results can be plotted using the |
|
237 | Once all the tasks have finished, the results can be plotted using the | |
243 | :func:`plot_options` function. Here we make contour plots of the Asian |
|
238 | :func:`plot_options` function. Here we make contour plots of the Asian | |
244 | call and Asian put options as function of the volatility and strike price: |
|
239 | call and Asian put options as function of the volatility and strike price: | |
245 |
|
240 | |||
246 | .. sourcecode:: ipython |
|
241 | .. sourcecode:: ipython | |
247 |
|
242 | |||
248 |
In [8]: plot_options(sigma_vals, |
|
243 | In [8]: plot_options(sigma_vals, strike_vals, prices['acall']) | |
249 |
|
244 | |||
250 | In [9]: plt.figure() |
|
245 | In [9]: plt.figure() | |
251 | Out[9]: <matplotlib.figure.Figure object at 0x18c178d0> |
|
246 | Out[9]: <matplotlib.figure.Figure object at 0x18c178d0> | |
252 |
|
247 | |||
253 |
In [10]: plot_options(sigma_vals, |
|
248 | In [10]: plot_options(sigma_vals, strike_vals, prices['aput']) | |
254 |
|
249 | |||
255 |
These results are shown in the two figures below. On |
|
250 | These results are shown in the two figures below. On our 15 engines, the | |
256 |
entire calculation (1 |
|
251 | entire calculation (15 strike prices, 15 volatilities, 100,000 paths for each) | |
257 |
took 3 |
|
252 | took 37 seconds in parallel, giving a speedup of 14.1x, which is comparable | |
258 | to the speedup observed in our previous example. |
|
253 | to the speedup observed in our previous example. | |
259 |
|
254 | |||
260 | .. image:: figs/asian_call.* |
|
255 | .. image:: figs/asian_call.* | |
261 |
|
256 | |||
262 | .. image:: figs/asian_put.* |
|
257 | .. image:: figs/asian_put.* | |
263 |
|
258 | |||
264 | Conclusion |
|
259 | Conclusion | |
265 | ========== |
|
260 | ========== | |
266 |
|
261 | |||
267 | To conclude these examples, we summarize the key features of IPython's |
|
262 | To conclude these examples, we summarize the key features of IPython's | |
268 | parallel architecture that have been demonstrated: |
|
263 | parallel architecture that have been demonstrated: | |
269 |
|
264 | |||
270 | * Serial code can be parallelized often with only a few extra lines of code. |
|
265 | * Serial code can be parallelized often with only a few extra lines of code. | |
271 | We have used the :class:`DirectView` and :class:`LoadBalancedView` classes |
|
266 | We have used the :class:`DirectView` and :class:`LoadBalancedView` classes | |
272 | for this purpose. |
|
267 | for this purpose. | |
273 | * The resulting parallel code can be run without ever leaving the IPython's |
|
268 | * The resulting parallel code can be run without ever leaving the IPython's | |
274 | interactive shell. |
|
269 | interactive shell. | |
275 | * Any data computed in parallel can be explored interactively through |
|
270 | * Any data computed in parallel can be explored interactively through | |
276 | visualization or further numerical calculations. |
|
271 | visualization or further numerical calculations. | |
277 |
* We have run these examples on a cluster running |
|
272 | * We have run these examples on a cluster running RHEL 5 and Sun GridEngine. | |
278 |
IPython's built in support for |
|
273 | IPython's built in support for SGE (and other batch systems) makes it easy | |
279 |
|
|
274 | to get started with IPython's parallel capabilities. | |
280 |
|
||||
281 | .. note:: |
|
|||
282 |
|
275 | |||
283 | The new parallel code has never been run on Windows HPC Server, so the last |
|
|||
284 | conclusion is untested. |
|
@@ -1,761 +1,826 b'' | |||||
1 | .. _parallel_process: |
|
1 | .. _parallel_process: | |
2 |
|
2 | |||
3 | =========================================== |
|
3 | =========================================== | |
4 | Starting the IPython controller and engines |
|
4 | Starting the IPython controller and engines | |
5 | =========================================== |
|
5 | =========================================== | |
6 |
|
6 | |||
7 | To use IPython for parallel computing, you need to start one instance of |
|
7 | To use IPython for parallel computing, you need to start one instance of | |
8 | the controller and one or more instances of the engine. The controller |
|
8 | the controller and one or more instances of the engine. The controller | |
9 | and each engine can run on different machines or on the same machine. |
|
9 | and each engine can run on different machines or on the same machine. | |
10 | Because of this, there are many different possibilities. |
|
10 | Because of this, there are many different possibilities. | |
11 |
|
11 | |||
12 | Broadly speaking, there are two ways of going about starting a controller and engines: |
|
12 | Broadly speaking, there are two ways of going about starting a controller and engines: | |
13 |
|
13 | |||
14 | * In an automated manner using the :command:`ipcluster` command. |
|
14 | * In an automated manner using the :command:`ipcluster` command. | |
15 | * In a more manual way using the :command:`ipcontroller` and |
|
15 | * In a more manual way using the :command:`ipcontroller` and | |
16 | :command:`ipengine` commands. |
|
16 | :command:`ipengine` commands. | |
17 |
|
17 | |||
18 | This document describes both of these methods. We recommend that new users |
|
18 | This document describes both of these methods. We recommend that new users | |
19 | start with the :command:`ipcluster` command as it simplifies many common usage |
|
19 | start with the :command:`ipcluster` command as it simplifies many common usage | |
20 | cases. |
|
20 | cases. | |
21 |
|
21 | |||
22 | General considerations |
|
22 | General considerations | |
23 | ====================== |
|
23 | ====================== | |
24 |
|
24 | |||
25 | Before delving into the details about how you can start a controller and |
|
25 | Before delving into the details about how you can start a controller and | |
26 | engines using the various methods, we outline some of the general issues that |
|
26 | engines using the various methods, we outline some of the general issues that | |
27 | come up when starting the controller and engines. These things come up no |
|
27 | come up when starting the controller and engines. These things come up no | |
28 | matter which method you use to start your IPython cluster. |
|
28 | matter which method you use to start your IPython cluster. | |
29 |
|
29 | |||
30 | If you are running engines on multiple machines, you will likely need to instruct the |
|
30 | If you are running engines on multiple machines, you will likely need to instruct the | |
31 | controller to listen for connections on an external interface. This can be done by specifying |
|
31 | controller to listen for connections on an external interface. This can be done by specifying | |
32 | the ``ip`` argument on the command-line, or the ``HubFactory.ip`` configurable in |
|
32 | the ``ip`` argument on the command-line, or the ``HubFactory.ip`` configurable in | |
33 | :file:`ipcontroller_config.py`. |
|
33 | :file:`ipcontroller_config.py`. | |
34 |
|
34 | |||
35 | If your machines are on a trusted network, you can safely instruct the controller to listen |
|
35 | If your machines are on a trusted network, you can safely instruct the controller to listen | |
36 | on all public interfaces with:: |
|
36 | on all public interfaces with:: | |
37 |
|
37 | |||
38 | $> ipcontroller --ip=* |
|
38 | $> ipcontroller --ip=* | |
39 |
|
39 | |||
40 | Or you can set the same behavior as the default by adding the following line to your :file:`ipcontroller_config.py`: |
|
40 | Or you can set the same behavior as the default by adding the following line to your :file:`ipcontroller_config.py`: | |
41 |
|
41 | |||
42 | .. sourcecode:: python |
|
42 | .. sourcecode:: python | |
43 |
|
43 | |||
44 | c.HubFactory.ip = '*' |
|
44 | c.HubFactory.ip = '*' | |
45 |
|
45 | |||
46 | .. note:: |
|
46 | .. note:: | |
47 |
|
47 | |||
48 | Due to the lack of security in ZeroMQ, the controller will only listen for connections on |
|
48 | Due to the lack of security in ZeroMQ, the controller will only listen for connections on | |
49 | localhost by default. If you see Timeout errors on engines or clients, then the first |
|
49 | localhost by default. If you see Timeout errors on engines or clients, then the first | |
50 | thing you should check is the ip address the controller is listening on, and make sure |
|
50 | thing you should check is the ip address the controller is listening on, and make sure | |
51 | that it is visible from the timing out machine. |
|
51 | that it is visible from the timing out machine. | |
52 |
|
52 | |||
53 | .. seealso:: |
|
53 | .. seealso:: | |
54 |
|
54 | |||
55 | Our `notes <parallel_security>`_ on security in the new parallel computing code. |
|
55 | Our `notes <parallel_security>`_ on security in the new parallel computing code. | |
56 |
|
56 | |||
57 | Let's say that you want to start the controller on ``host0`` and engines on |
|
57 | Let's say that you want to start the controller on ``host0`` and engines on | |
58 | hosts ``host1``-``hostn``. The following steps are then required: |
|
58 | hosts ``host1``-``hostn``. The following steps are then required: | |
59 |
|
59 | |||
60 | 1. Start the controller on ``host0`` by running :command:`ipcontroller` on |
|
60 | 1. Start the controller on ``host0`` by running :command:`ipcontroller` on | |
61 | ``host0``. The controller must be instructed to listen on an interface visible |
|
61 | ``host0``. The controller must be instructed to listen on an interface visible | |
62 | to the engine machines, via the ``ip`` command-line argument or ``HubFactory.ip`` |
|
62 | to the engine machines, via the ``ip`` command-line argument or ``HubFactory.ip`` | |
63 | in :file:`ipcontroller_config.py`. |
|
63 | in :file:`ipcontroller_config.py`. | |
64 | 2. Move the JSON file (:file:`ipcontroller-engine.json`) created by the |
|
64 | 2. Move the JSON file (:file:`ipcontroller-engine.json`) created by the | |
65 | controller from ``host0`` to hosts ``host1``-``hostn``. |
|
65 | controller from ``host0`` to hosts ``host1``-``hostn``. | |
66 | 3. Start the engines on hosts ``host1``-``hostn`` by running |
|
66 | 3. Start the engines on hosts ``host1``-``hostn`` by running | |
67 | :command:`ipengine`. This command has to be told where the JSON file |
|
67 | :command:`ipengine`. This command has to be told where the JSON file | |
68 | (:file:`ipcontroller-engine.json`) is located. |
|
68 | (:file:`ipcontroller-engine.json`) is located. | |
69 |
|
69 | |||
70 | At this point, the controller and engines will be connected. By default, the JSON files |
|
70 | At this point, the controller and engines will be connected. By default, the JSON files | |
71 | created by the controller are put into the :file:`~/.ipython/profile_default/security` |
|
71 | created by the controller are put into the :file:`~/.ipython/profile_default/security` | |
72 | directory. If the engines share a filesystem with the controller, step 2 can be skipped as |
|
72 | directory. If the engines share a filesystem with the controller, step 2 can be skipped as | |
73 | the engines will automatically look at that location. |
|
73 | the engines will automatically look at that location. | |
74 |
|
74 | |||
75 | The final step required to actually use the running controller from a client is to move |
|
75 | The final step required to actually use the running controller from a client is to move | |
76 | the JSON file :file:`ipcontroller-client.json` from ``host0`` to any host where clients |
|
76 | the JSON file :file:`ipcontroller-client.json` from ``host0`` to any host where clients | |
77 | will be run. If these file are put into the :file:`~/.ipython/profile_default/security` |
|
77 | will be run. If these file are put into the :file:`~/.ipython/profile_default/security` | |
78 | directory of the client's host, they will be found automatically. Otherwise, the full path |
|
78 | directory of the client's host, they will be found automatically. Otherwise, the full path | |
79 | to them has to be passed to the client's constructor. |
|
79 | to them has to be passed to the client's constructor. | |
80 |
|
80 | |||
81 | Using :command:`ipcluster` |
|
81 | Using :command:`ipcluster` | |
82 | =========================== |
|
82 | =========================== | |
83 |
|
83 | |||
84 | The :command:`ipcluster` command provides a simple way of starting a |
|
84 | The :command:`ipcluster` command provides a simple way of starting a | |
85 | controller and engines in the following situations: |
|
85 | controller and engines in the following situations: | |
86 |
|
86 | |||
87 | 1. When the controller and engines are all run on localhost. This is useful |
|
87 | 1. When the controller and engines are all run on localhost. This is useful | |
88 | for testing or running on a multicore computer. |
|
88 | for testing or running on a multicore computer. | |
89 | 2. When engines are started using the :command:`mpiexec` command that comes |
|
89 | 2. When engines are started using the :command:`mpiexec` command that comes | |
90 | with most MPI [MPI]_ implementations |
|
90 | with most MPI [MPI]_ implementations | |
91 | 3. When engines are started using the PBS [PBS]_ batch system |
|
91 | 3. When engines are started using the PBS [PBS]_ batch system | |
92 | (or other `qsub` systems, such as SGE). |
|
92 | (or other `qsub` systems, such as SGE). | |
93 | 4. When the controller is started on localhost and the engines are started on |
|
93 | 4. When the controller is started on localhost and the engines are started on | |
94 | remote nodes using :command:`ssh`. |
|
94 | remote nodes using :command:`ssh`. | |
95 | 5. When engines are started using the Windows HPC Server batch system. |
|
95 | 5. When engines are started using the Windows HPC Server batch system. | |
96 |
|
96 | |||
97 | .. note:: |
|
97 | .. note:: | |
98 |
|
98 | |||
99 | Currently :command:`ipcluster` requires that the |
|
99 | Currently :command:`ipcluster` requires that the | |
100 | :file:`~/.ipython/profile_<name>/security` directory live on a shared filesystem that is |
|
100 | :file:`~/.ipython/profile_<name>/security` directory live on a shared filesystem that is | |
101 | seen by both the controller and engines. If you don't have a shared file |
|
101 | seen by both the controller and engines. If you don't have a shared file | |
102 | system you will need to use :command:`ipcontroller` and |
|
102 | system you will need to use :command:`ipcontroller` and | |
103 | :command:`ipengine` directly. |
|
103 | :command:`ipengine` directly. | |
104 |
|
104 | |||
105 | Under the hood, :command:`ipcluster` just uses :command:`ipcontroller` |
|
105 | Under the hood, :command:`ipcluster` just uses :command:`ipcontroller` | |
106 | and :command:`ipengine` to perform the steps described above. |
|
106 | and :command:`ipengine` to perform the steps described above. | |
107 |
|
107 | |||
108 | The simplest way to use ipcluster requires no configuration, and will |
|
108 | The simplest way to use ipcluster requires no configuration, and will | |
109 | launch a controller and a number of engines on the local machine. For instance, |
|
109 | launch a controller and a number of engines on the local machine. For instance, | |
110 | to start one controller and 4 engines on localhost, just do:: |
|
110 | to start one controller and 4 engines on localhost, just do:: | |
111 |
|
111 | |||
112 | $ ipcluster start -n 4 |
|
112 | $ ipcluster start -n 4 | |
113 |
|
113 | |||
114 | To see other command line options, do:: |
|
114 | To see other command line options, do:: | |
115 |
|
115 | |||
116 | $ ipcluster -h |
|
116 | $ ipcluster -h | |
117 |
|
117 | |||
118 |
|
118 | |||
119 | Configuring an IPython cluster |
|
119 | Configuring an IPython cluster | |
120 | ============================== |
|
120 | ============================== | |
121 |
|
121 | |||
122 | Cluster configurations are stored as `profiles`. You can create a new profile with:: |
|
122 | Cluster configurations are stored as `profiles`. You can create a new profile with:: | |
123 |
|
123 | |||
124 | $ ipython profile create --parallel --profile=myprofile |
|
124 | $ ipython profile create --parallel --profile=myprofile | |
125 |
|
125 | |||
126 | This will create the directory :file:`IPYTHONDIR/profile_myprofile`, and populate it |
|
126 | This will create the directory :file:`IPYTHONDIR/profile_myprofile`, and populate it | |
127 | with the default configuration files for the three IPython cluster commands. Once |
|
127 | with the default configuration files for the three IPython cluster commands. Once | |
128 | you edit those files, you can continue to call ipcluster/ipcontroller/ipengine |
|
128 | you edit those files, you can continue to call ipcluster/ipcontroller/ipengine | |
129 | with no arguments beyond ``profile=myprofile``, and any configuration will be maintained. |
|
129 | with no arguments beyond ``profile=myprofile``, and any configuration will be maintained. | |
130 |
|
130 | |||
131 | There is no limit to the number of profiles you can have, so you can maintain a profile for each |
|
131 | There is no limit to the number of profiles you can have, so you can maintain a profile for each | |
132 | of your common use cases. The default profile will be used whenever the |
|
132 | of your common use cases. The default profile will be used whenever the | |
133 | profile argument is not specified, so edit :file:`IPYTHONDIR/profile_default/*_config.py` to |
|
133 | profile argument is not specified, so edit :file:`IPYTHONDIR/profile_default/*_config.py` to | |
134 | represent your most common use case. |
|
134 | represent your most common use case. | |
135 |
|
135 | |||
136 | The configuration files are loaded with commented-out settings and explanations, |
|
136 | The configuration files are loaded with commented-out settings and explanations, | |
137 | which should cover most of the available possibilities. |
|
137 | which should cover most of the available possibilities. | |
138 |
|
138 | |||
139 | Using various batch systems with :command:`ipcluster` |
|
139 | Using various batch systems with :command:`ipcluster` | |
140 | ----------------------------------------------------- |
|
140 | ----------------------------------------------------- | |
141 |
|
141 | |||
142 | :command:`ipcluster` has a notion of Launchers that can start controllers |
|
142 | :command:`ipcluster` has a notion of Launchers that can start controllers | |
143 | and engines with various remote execution schemes. Currently supported |
|
143 | and engines with various remote execution schemes. Currently supported | |
144 | models include :command:`ssh`, :command:`mpiexec`, PBS-style (Torque, SGE, LSF), |
|
144 | models include :command:`ssh`, :command:`mpiexec`, PBS-style (Torque, SGE, LSF), | |
145 | and Windows HPC Server. |
|
145 | and Windows HPC Server. | |
146 |
|
146 | |||
147 | In general, these are configured by the :attr:`IPClusterEngines.engine_set_launcher_class`, |
|
147 | In general, these are configured by the :attr:`IPClusterEngines.engine_set_launcher_class`, | |
148 | and :attr:`IPClusterStart.controller_launcher_class` configurables, which can be the |
|
148 | and :attr:`IPClusterStart.controller_launcher_class` configurables, which can be the | |
149 | fully specified object name (e.g. ``'IPython.parallel.apps.launcher.LocalControllerLauncher'``), |
|
149 | fully specified object name (e.g. ``'IPython.parallel.apps.launcher.LocalControllerLauncher'``), | |
150 | but if you are using IPython's builtin launchers, you can specify just the class name, |
|
150 | but if you are using IPython's builtin launchers, you can specify just the class name, | |
151 | or even just the prefix e.g: |
|
151 | or even just the prefix e.g: | |
152 |
|
152 | |||
153 | .. sourcecode:: python |
|
153 | .. sourcecode:: python | |
154 |
|
154 | |||
155 | c.IPClusterEngines.engine_launcher_class = 'SSH' |
|
155 | c.IPClusterEngines.engine_launcher_class = 'SSH' | |
156 | # equivalent to |
|
156 | # equivalent to | |
157 | c.IPClusterEngines.engine_launcher_class = 'SSHEngineSetLauncher' |
|
157 | c.IPClusterEngines.engine_launcher_class = 'SSHEngineSetLauncher' | |
158 | # both of which expand to |
|
158 | # both of which expand to | |
159 | c.IPClusterEngines.engine_launcher_class = 'IPython.parallel.apps.launcher.SSHEngineSetLauncher' |
|
159 | c.IPClusterEngines.engine_launcher_class = 'IPython.parallel.apps.launcher.SSHEngineSetLauncher' | |
160 |
|
160 | |||
161 | The shortest form being of particular use on the command line, where all you need to do to |
|
161 | The shortest form being of particular use on the command line, where all you need to do to | |
162 | get an IPython cluster running with engines started with MPI is: |
|
162 | get an IPython cluster running with engines started with MPI is: | |
163 |
|
163 | |||
164 | .. sourcecode:: bash |
|
164 | .. sourcecode:: bash | |
165 |
|
165 | |||
166 | $> ipcluster start --engines=MPIExec |
|
166 | $> ipcluster start --engines=MPIExec | |
167 |
|
167 | |||
168 | Assuming that the default MPI config is sufficient. |
|
168 | Assuming that the default MPI config is sufficient. | |
169 |
|
169 | |||
170 | .. note:: |
|
170 | .. note:: | |
171 |
|
171 | |||
172 | shortcuts for builtin launcher names were added in 0.12, as was the ``_class`` suffix |
|
172 | shortcuts for builtin launcher names were added in 0.12, as was the ``_class`` suffix | |
173 | on the configurable names. If you use the old 0.11 names (e.g. ``engine_set_launcher``), |
|
173 | on the configurable names. If you use the old 0.11 names (e.g. ``engine_set_launcher``), | |
174 | they will still work, but you will get a deprecation warning that the name has changed. |
|
174 | they will still work, but you will get a deprecation warning that the name has changed. | |
175 |
|
175 | |||
176 |
|
176 | |||
177 | .. note:: |
|
177 | .. note:: | |
178 |
|
178 | |||
179 | The Launchers and configuration are designed in such a way that advanced |
|
179 | The Launchers and configuration are designed in such a way that advanced | |
180 | users can subclass and configure them to fit their own system that we |
|
180 | users can subclass and configure them to fit their own system that we | |
181 | have not yet supported (such as Condor) |
|
181 | have not yet supported (such as Condor) | |
182 |
|
182 | |||
183 | Using :command:`ipcluster` in mpiexec/mpirun mode |
|
183 | Using :command:`ipcluster` in mpiexec/mpirun mode | |
184 |
------------------------------------------------- |
|
184 | ------------------------------------------------- | |
185 |
|
185 | |||
186 |
|
186 | |||
187 | The mpiexec/mpirun mode is useful if you: |
|
187 | The mpiexec/mpirun mode is useful if you: | |
188 |
|
188 | |||
189 | 1. Have MPI installed. |
|
189 | 1. Have MPI installed. | |
190 | 2. Your systems are configured to use the :command:`mpiexec` or |
|
190 | 2. Your systems are configured to use the :command:`mpiexec` or | |
191 | :command:`mpirun` commands to start MPI processes. |
|
191 | :command:`mpirun` commands to start MPI processes. | |
192 |
|
192 | |||
193 | If these are satisfied, you can create a new profile:: |
|
193 | If these are satisfied, you can create a new profile:: | |
194 |
|
194 | |||
195 | $ ipython profile create --parallel --profile=mpi |
|
195 | $ ipython profile create --parallel --profile=mpi | |
196 |
|
196 | |||
197 | and edit the file :file:`IPYTHONDIR/profile_mpi/ipcluster_config.py`. |
|
197 | and edit the file :file:`IPYTHONDIR/profile_mpi/ipcluster_config.py`. | |
198 |
|
198 | |||
199 | There, instruct ipcluster to use the MPIExec launchers by adding the lines: |
|
199 | There, instruct ipcluster to use the MPIExec launchers by adding the lines: | |
200 |
|
200 | |||
201 | .. sourcecode:: python |
|
201 | .. sourcecode:: python | |
202 |
|
202 | |||
203 | c.IPClusterEngines.engine_launcher_class = 'MPIExecEngineSetLauncher' |
|
203 | c.IPClusterEngines.engine_launcher_class = 'MPIExecEngineSetLauncher' | |
204 |
|
204 | |||
205 | If the default MPI configuration is correct, then you can now start your cluster, with:: |
|
205 | If the default MPI configuration is correct, then you can now start your cluster, with:: | |
206 |
|
206 | |||
207 | $ ipcluster start -n 4 --profile=mpi |
|
207 | $ ipcluster start -n 4 --profile=mpi | |
208 |
|
208 | |||
209 | This does the following: |
|
209 | This does the following: | |
210 |
|
210 | |||
211 | 1. Starts the IPython controller on current host. |
|
211 | 1. Starts the IPython controller on current host. | |
212 | 2. Uses :command:`mpiexec` to start 4 engines. |
|
212 | 2. Uses :command:`mpiexec` to start 4 engines. | |
213 |
|
213 | |||
214 | If you have a reason to also start the Controller with mpi, you can specify: |
|
214 | If you have a reason to also start the Controller with mpi, you can specify: | |
215 |
|
215 | |||
216 | .. sourcecode:: python |
|
216 | .. sourcecode:: python | |
217 |
|
217 | |||
218 | c.IPClusterStart.controller_launcher_class = 'MPIExecControllerLauncher' |
|
218 | c.IPClusterStart.controller_launcher_class = 'MPIExecControllerLauncher' | |
219 |
|
219 | |||
220 | .. note:: |
|
220 | .. note:: | |
221 |
|
221 | |||
222 | The Controller *will not* be in the same MPI universe as the engines, so there is not |
|
222 | The Controller *will not* be in the same MPI universe as the engines, so there is not | |
223 | much reason to do this unless sysadmins demand it. |
|
223 | much reason to do this unless sysadmins demand it. | |
224 |
|
224 | |||
225 | On newer MPI implementations (such as OpenMPI), this will work even if you |
|
225 | On newer MPI implementations (such as OpenMPI), this will work even if you | |
226 | don't make any calls to MPI or call :func:`MPI_Init`. However, older MPI |
|
226 | don't make any calls to MPI or call :func:`MPI_Init`. However, older MPI | |
227 | implementations actually require each process to call :func:`MPI_Init` upon |
|
227 | implementations actually require each process to call :func:`MPI_Init` upon | |
228 | starting. The easiest way of having this done is to install the mpi4py |
|
228 | starting. The easiest way of having this done is to install the mpi4py | |
229 | [mpi4py]_ package and then specify the ``c.MPI.use`` option in :file:`ipengine_config.py`: |
|
229 | [mpi4py]_ package and then specify the ``c.MPI.use`` option in :file:`ipengine_config.py`: | |
230 |
|
230 | |||
231 | .. sourcecode:: python |
|
231 | .. sourcecode:: python | |
232 |
|
232 | |||
233 | c.MPI.use = 'mpi4py' |
|
233 | c.MPI.use = 'mpi4py' | |
234 |
|
234 | |||
235 | Unfortunately, even this won't work for some MPI implementations. If you are |
|
235 | Unfortunately, even this won't work for some MPI implementations. If you are | |
236 | having problems with this, you will likely have to use a custom Python |
|
236 | having problems with this, you will likely have to use a custom Python | |
237 | executable that itself calls :func:`MPI_Init` at the appropriate time. |
|
237 | executable that itself calls :func:`MPI_Init` at the appropriate time. | |
238 | Fortunately, mpi4py comes with such a custom Python executable that is easy to |
|
238 | Fortunately, mpi4py comes with such a custom Python executable that is easy to | |
239 | install and use. However, this custom Python executable approach will not work |
|
239 | install and use. However, this custom Python executable approach will not work | |
240 | with :command:`ipcluster` currently. |
|
240 | with :command:`ipcluster` currently. | |
241 |
|
241 | |||
242 | More details on using MPI with IPython can be found :ref:`here <parallelmpi>`. |
|
242 | More details on using MPI with IPython can be found :ref:`here <parallelmpi>`. | |
243 |
|
243 | |||
244 |
|
244 | |||
245 | Using :command:`ipcluster` in PBS mode |
|
245 | Using :command:`ipcluster` in PBS mode | |
246 |
-------------------------------------- |
|
246 | -------------------------------------- | |
247 |
|
247 | |||
248 | The PBS mode uses the Portable Batch System (PBS) to start the engines. |
|
248 | The PBS mode uses the Portable Batch System (PBS) to start the engines. | |
249 |
|
249 | |||
250 | As usual, we will start by creating a fresh profile:: |
|
250 | As usual, we will start by creating a fresh profile:: | |
251 |
|
251 | |||
252 | $ ipython profile create --parallel --profile=pbs |
|
252 | $ ipython profile create --parallel --profile=pbs | |
253 |
|
253 | |||
254 | And in :file:`ipcluster_config.py`, we will select the PBS launchers for the controller |
|
254 | And in :file:`ipcluster_config.py`, we will select the PBS launchers for the controller | |
255 | and engines: |
|
255 | and engines: | |
256 |
|
256 | |||
257 | .. sourcecode:: python |
|
257 | .. sourcecode:: python | |
258 |
|
258 | |||
259 | c.IPClusterStart.controller_launcher_class = 'PBSControllerLauncher' |
|
259 | c.IPClusterStart.controller_launcher_class = 'PBSControllerLauncher' | |
260 | c.IPClusterEngines.engine_launcher_class = 'PBSEngineSetLauncher' |
|
260 | c.IPClusterEngines.engine_launcher_class = 'PBSEngineSetLauncher' | |
261 |
|
261 | |||
262 | .. note:: |
|
262 | .. note:: | |
263 |
|
263 | |||
264 | Note that the configurable is IPClusterEngines for the engine launcher, and |
|
264 | Note that the configurable is IPClusterEngines for the engine launcher, and | |
265 | IPClusterStart for the controller launcher. This is because the start command is a |
|
265 | IPClusterStart for the controller launcher. This is because the start command is a | |
266 | subclass of the engine command, adding a controller launcher. Since it is a subclass, |
|
266 | subclass of the engine command, adding a controller launcher. Since it is a subclass, | |
267 | any configuration made in IPClusterEngines is inherited by IPClusterStart unless it is |
|
267 | any configuration made in IPClusterEngines is inherited by IPClusterStart unless it is | |
268 | overridden. |
|
268 | overridden. | |
269 |
|
269 | |||
270 | IPython does provide simple default batch templates for PBS and SGE, but you may need |
|
270 | IPython does provide simple default batch templates for PBS and SGE, but you may need | |
271 | to specify your own. Here is a sample PBS script template: |
|
271 | to specify your own. Here is a sample PBS script template: | |
272 |
|
272 | |||
273 | .. sourcecode:: bash |
|
273 | .. sourcecode:: bash | |
274 |
|
274 | |||
275 | #PBS -N ipython |
|
275 | #PBS -N ipython | |
276 | #PBS -j oe |
|
276 | #PBS -j oe | |
277 | #PBS -l walltime=00:10:00 |
|
277 | #PBS -l walltime=00:10:00 | |
278 | #PBS -l nodes={n/4}:ppn=4 |
|
278 | #PBS -l nodes={n/4}:ppn=4 | |
279 | #PBS -q {queue} |
|
279 | #PBS -q {queue} | |
280 |
|
280 | |||
281 | cd $PBS_O_WORKDIR |
|
281 | cd $PBS_O_WORKDIR | |
282 | export PATH=$HOME/usr/local/bin |
|
282 | export PATH=$HOME/usr/local/bin | |
283 | export PYTHONPATH=$HOME/usr/local/lib/python2.7/site-packages |
|
283 | export PYTHONPATH=$HOME/usr/local/lib/python2.7/site-packages | |
284 | /usr/local/bin/mpiexec -n {n} ipengine --profile-dir={profile_dir} |
|
284 | /usr/local/bin/mpiexec -n {n} ipengine --profile-dir={profile_dir} | |
285 |
|
285 | |||
286 | There are a few important points about this template: |
|
286 | There are a few important points about this template: | |
287 |
|
287 | |||
288 | 1. This template will be rendered at runtime using IPython's :class:`EvalFormatter`. |
|
288 | 1. This template will be rendered at runtime using IPython's :class:`EvalFormatter`. | |
289 | This is simply a subclass of :class:`string.Formatter` that allows simple expressions |
|
289 | This is simply a subclass of :class:`string.Formatter` that allows simple expressions | |
290 | on keys. |
|
290 | on keys. | |
291 |
|
291 | |||
292 | 2. Instead of putting in the actual number of engines, use the notation |
|
292 | 2. Instead of putting in the actual number of engines, use the notation | |
293 | ``{n}`` to indicate the number of engines to be started. You can also use |
|
293 | ``{n}`` to indicate the number of engines to be started. You can also use | |
294 | expressions like ``{n/4}`` in the template to indicate the number of nodes. |
|
294 | expressions like ``{n/4}`` in the template to indicate the number of nodes. | |
295 | There will always be ``{n}`` and ``{profile_dir}`` variables passed to the formatter. |
|
295 | There will always be ``{n}`` and ``{profile_dir}`` variables passed to the formatter. | |
296 | These allow the batch system to know how many engines, and where the configuration |
|
296 | These allow the batch system to know how many engines, and where the configuration | |
297 | files reside. The same is true for the batch queue, with the template variable |
|
297 | files reside. The same is true for the batch queue, with the template variable | |
298 | ``{queue}``. |
|
298 | ``{queue}``. | |
299 |
|
299 | |||
300 | 3. Any options to :command:`ipengine` can be given in the batch script |
|
300 | 3. Any options to :command:`ipengine` can be given in the batch script | |
301 | template, or in :file:`ipengine_config.py`. |
|
301 | template, or in :file:`ipengine_config.py`. | |
302 |
|
302 | |||
303 | 4. Depending on the configuration of you system, you may have to set |
|
303 | 4. Depending on the configuration of you system, you may have to set | |
304 | environment variables in the script template. |
|
304 | environment variables in the script template. | |
305 |
|
305 | |||
306 | The controller template should be similar, but simpler: |
|
306 | The controller template should be similar, but simpler: | |
307 |
|
307 | |||
308 | .. sourcecode:: bash |
|
308 | .. sourcecode:: bash | |
309 |
|
309 | |||
310 | #PBS -N ipython |
|
310 | #PBS -N ipython | |
311 | #PBS -j oe |
|
311 | #PBS -j oe | |
312 | #PBS -l walltime=00:10:00 |
|
312 | #PBS -l walltime=00:10:00 | |
313 | #PBS -l nodes=1:ppn=4 |
|
313 | #PBS -l nodes=1:ppn=4 | |
314 | #PBS -q {queue} |
|
314 | #PBS -q {queue} | |
315 |
|
315 | |||
316 | cd $PBS_O_WORKDIR |
|
316 | cd $PBS_O_WORKDIR | |
317 | export PATH=$HOME/usr/local/bin |
|
317 | export PATH=$HOME/usr/local/bin | |
318 | export PYTHONPATH=$HOME/usr/local/lib/python2.7/site-packages |
|
318 | export PYTHONPATH=$HOME/usr/local/lib/python2.7/site-packages | |
319 | ipcontroller --profile-dir={profile_dir} |
|
319 | ipcontroller --profile-dir={profile_dir} | |
320 |
|
320 | |||
321 |
|
321 | |||
322 | Once you have created these scripts, save them with names like |
|
322 | Once you have created these scripts, save them with names like | |
323 | :file:`pbs.engine.template`. Now you can load them into the :file:`ipcluster_config` with: |
|
323 | :file:`pbs.engine.template`. Now you can load them into the :file:`ipcluster_config` with: | |
324 |
|
324 | |||
325 | .. sourcecode:: python |
|
325 | .. sourcecode:: python | |
326 |
|
326 | |||
327 | c.PBSEngineSetLauncher.batch_template_file = "pbs.engine.template" |
|
327 | c.PBSEngineSetLauncher.batch_template_file = "pbs.engine.template" | |
328 |
|
328 | |||
329 | c.PBSControllerLauncher.batch_template_file = "pbs.controller.template" |
|
329 | c.PBSControllerLauncher.batch_template_file = "pbs.controller.template" | |
330 |
|
330 | |||
331 |
|
331 | |||
332 | Alternately, you can just define the templates as strings inside :file:`ipcluster_config`. |
|
332 | Alternately, you can just define the templates as strings inside :file:`ipcluster_config`. | |
333 |
|
333 | |||
334 | Whether you are using your own templates or our defaults, the extra configurables available are |
|
334 | Whether you are using your own templates or our defaults, the extra configurables available are | |
335 | the number of engines to launch (``{n}``, and the batch system queue to which the jobs are to be |
|
335 | the number of engines to launch (``{n}``, and the batch system queue to which the jobs are to be | |
336 | submitted (``{queue}``)). These are configurables, and can be specified in |
|
336 | submitted (``{queue}``)). These are configurables, and can be specified in | |
337 | :file:`ipcluster_config`: |
|
337 | :file:`ipcluster_config`: | |
338 |
|
338 | |||
339 | .. sourcecode:: python |
|
339 | .. sourcecode:: python | |
340 |
|
340 | |||
341 | c.PBSLauncher.queue = 'veryshort.q' |
|
341 | c.PBSLauncher.queue = 'veryshort.q' | |
342 | c.IPClusterEngines.n = 64 |
|
342 | c.IPClusterEngines.n = 64 | |
343 |
|
343 | |||
344 | Note that assuming you are running PBS on a multi-node cluster, the Controller's default behavior |
|
344 | Note that assuming you are running PBS on a multi-node cluster, the Controller's default behavior | |
345 | of listening only on localhost is likely too restrictive. In this case, also assuming the |
|
345 | of listening only on localhost is likely too restrictive. In this case, also assuming the | |
346 | nodes are safely behind a firewall, you can simply instruct the Controller to listen for |
|
346 | nodes are safely behind a firewall, you can simply instruct the Controller to listen for | |
347 | connections on all its interfaces, by adding in :file:`ipcontroller_config`: |
|
347 | connections on all its interfaces, by adding in :file:`ipcontroller_config`: | |
348 |
|
348 | |||
349 | .. sourcecode:: python |
|
349 | .. sourcecode:: python | |
350 |
|
350 | |||
351 | c.HubFactory.ip = '*' |
|
351 | c.HubFactory.ip = '*' | |
352 |
|
352 | |||
353 | You can now run the cluster with:: |
|
353 | You can now run the cluster with:: | |
354 |
|
354 | |||
355 | $ ipcluster start --profile=pbs -n 128 |
|
355 | $ ipcluster start --profile=pbs -n 128 | |
356 |
|
356 | |||
357 | Additional configuration options can be found in the PBS section of :file:`ipcluster_config`. |
|
357 | Additional configuration options can be found in the PBS section of :file:`ipcluster_config`. | |
358 |
|
358 | |||
359 | .. note:: |
|
359 | .. note:: | |
360 |
|
360 | |||
361 | Due to the flexibility of configuration, the PBS launchers work with simple changes |
|
361 | Due to the flexibility of configuration, the PBS launchers work with simple changes | |
362 | to the template for other :command:`qsub`-using systems, such as Sun Grid Engine, |
|
362 | to the template for other :command:`qsub`-using systems, such as Sun Grid Engine, | |
363 | and with further configuration in similar batch systems like Condor. |
|
363 | and with further configuration in similar batch systems like Condor. | |
364 |
|
364 | |||
365 |
|
365 | |||
366 | Using :command:`ipcluster` in SSH mode |
|
366 | Using :command:`ipcluster` in SSH mode | |
367 |
-------------------------------------- |
|
367 | -------------------------------------- | |
368 |
|
368 | |||
369 |
|
369 | |||
370 | The SSH mode uses :command:`ssh` to execute :command:`ipengine` on remote |
|
370 | The SSH mode uses :command:`ssh` to execute :command:`ipengine` on remote | |
371 | nodes and :command:`ipcontroller` can be run remotely as well, or on localhost. |
|
371 | nodes and :command:`ipcontroller` can be run remotely as well, or on localhost. | |
372 |
|
372 | |||
373 | .. note:: |
|
373 | .. note:: | |
374 |
|
374 | |||
375 | When using this mode it highly recommended that you have set up SSH keys |
|
375 | When using this mode it highly recommended that you have set up SSH keys | |
376 | and are using ssh-agent [SSH]_ for password-less logins. |
|
376 | and are using ssh-agent [SSH]_ for password-less logins. | |
377 |
|
377 | |||
378 | As usual, we start by creating a clean profile:: |
|
378 | As usual, we start by creating a clean profile:: | |
379 |
|
379 | |||
380 | $ ipython profile create --parallel --profile=ssh |
|
380 | $ ipython profile create --parallel --profile=ssh | |
381 |
|
381 | |||
382 | To use this mode, select the SSH launchers in :file:`ipcluster_config.py`: |
|
382 | To use this mode, select the SSH launchers in :file:`ipcluster_config.py`: | |
383 |
|
383 | |||
384 | .. sourcecode:: python |
|
384 | .. sourcecode:: python | |
385 |
|
385 | |||
386 | c.IPClusterEngines.engine_launcher_class = 'SSHEngineSetLauncher' |
|
386 | c.IPClusterEngines.engine_launcher_class = 'SSHEngineSetLauncher' | |
387 | # and if the Controller is also to be remote: |
|
387 | # and if the Controller is also to be remote: | |
388 | c.IPClusterStart.controller_launcher_class = 'SSHControllerLauncher' |
|
388 | c.IPClusterStart.controller_launcher_class = 'SSHControllerLauncher' | |
389 |
|
389 | |||
390 |
|
390 | |||
391 |
|
391 | |||
392 | The controller's remote location and configuration can be specified: |
|
392 | The controller's remote location and configuration can be specified: | |
393 |
|
393 | |||
394 | .. sourcecode:: python |
|
394 | .. sourcecode:: python | |
395 |
|
395 | |||
396 | # Set the user and hostname for the controller |
|
396 | # Set the user and hostname for the controller | |
397 | # c.SSHControllerLauncher.hostname = 'controller.example.com' |
|
397 | # c.SSHControllerLauncher.hostname = 'controller.example.com' | |
398 | # c.SSHControllerLauncher.user = os.environ.get('USER','username') |
|
398 | # c.SSHControllerLauncher.user = os.environ.get('USER','username') | |
399 |
|
399 | |||
400 | # Set the arguments to be passed to ipcontroller |
|
400 | # Set the arguments to be passed to ipcontroller | |
401 | # note that remotely launched ipcontroller will not get the contents of |
|
401 | # note that remotely launched ipcontroller will not get the contents of | |
402 | # the local ipcontroller_config.py unless it resides on the *remote host* |
|
402 | # the local ipcontroller_config.py unless it resides on the *remote host* | |
403 | # in the location specified by the `profile-dir` argument. |
|
403 | # in the location specified by the `profile-dir` argument. | |
404 |
# c.SSHControllerLauncher. |
|
404 | # c.SSHControllerLauncher.controller_args = ['--reuse', '--ip=*', '--profile-dir=/path/to/cd'] | |
405 |
|
405 | |||
406 | .. note:: |
|
406 | .. note:: | |
407 |
|
407 | |||
408 | SSH mode does not do any file movement, so you will need to distribute configuration |
|
408 | SSH mode does not do any file movement, so you will need to distribute configuration | |
409 | files manually. To aid in this, the `reuse_files` flag defaults to True for ssh-launched |
|
409 | files manually. To aid in this, the `reuse_files` flag defaults to True for ssh-launched | |
410 | Controllers, so you will only need to do this once, unless you override this flag back |
|
410 | Controllers, so you will only need to do this once, unless you override this flag back | |
411 | to False. |
|
411 | to False. | |
412 |
|
412 | |||
413 | Engines are specified in a dictionary, by hostname and the number of engines to be run |
|
413 | Engines are specified in a dictionary, by hostname and the number of engines to be run | |
414 | on that host. |
|
414 | on that host. | |
415 |
|
415 | |||
416 | .. sourcecode:: python |
|
416 | .. sourcecode:: python | |
417 |
|
417 | |||
418 | c.SSHEngineSetLauncher.engines = { 'host1.example.com' : 2, |
|
418 | c.SSHEngineSetLauncher.engines = { 'host1.example.com' : 2, | |
419 | 'host2.example.com' : 5, |
|
419 | 'host2.example.com' : 5, | |
420 | 'host3.example.com' : (1, ['--profile-dir=/home/different/location']), |
|
420 | 'host3.example.com' : (1, ['--profile-dir=/home/different/location']), | |
421 | 'host4.example.com' : 8 } |
|
421 | 'host4.example.com' : 8 } | |
422 |
|
422 | |||
423 | * The `engines` dict, where the keys are the host we want to run engines on and |
|
423 | * The `engines` dict, where the keys are the host we want to run engines on and | |
424 | the value is the number of engines to run on that host. |
|
424 | the value is the number of engines to run on that host. | |
425 | * on host3, the value is a tuple, where the number of engines is first, and the arguments |
|
425 | * on host3, the value is a tuple, where the number of engines is first, and the arguments | |
426 | to be passed to :command:`ipengine` are the second element. |
|
426 | to be passed to :command:`ipengine` are the second element. | |
427 |
|
427 | |||
428 | For engines without explicitly specified arguments, the default arguments are set in |
|
428 | For engines without explicitly specified arguments, the default arguments are set in | |
429 | a single location: |
|
429 | a single location: | |
430 |
|
430 | |||
431 | .. sourcecode:: python |
|
431 | .. sourcecode:: python | |
432 |
|
432 | |||
433 | c.SSHEngineSetLauncher.engine_args = ['--profile-dir=/path/to/profile_ssh'] |
|
433 | c.SSHEngineSetLauncher.engine_args = ['--profile-dir=/path/to/profile_ssh'] | |
434 |
|
434 | |||
435 | Current limitations of the SSH mode of :command:`ipcluster` are: |
|
435 | Current limitations of the SSH mode of :command:`ipcluster` are: | |
436 |
|
436 | |||
437 | * Untested on Windows. Would require a working :command:`ssh` on Windows. |
|
437 | * Untested on Windows. Would require a working :command:`ssh` on Windows. | |
438 | Also, we are using shell scripts to setup and execute commands on remote |
|
438 | Also, we are using shell scripts to setup and execute commands on remote | |
439 | hosts. |
|
439 | hosts. | |
440 | * No file movement - This is a regression from 0.10, which moved connection files |
|
440 | * No file movement - This is a regression from 0.10, which moved connection files | |
441 |
around with scp. This will be improved, |
|
441 | around with scp. This will be improved, Pull Requests are welcome. | |
|
442 | ||||
442 |
|
443 | |||
443 | Using the :command:`ipcontroller` and :command:`ipengine` commands |
|
444 | Using the :command:`ipcontroller` and :command:`ipengine` commands | |
444 |
================================================================== |
|
445 | ================================================================== | |
445 |
|
446 | |||
446 | It is also possible to use the :command:`ipcontroller` and :command:`ipengine` |
|
447 | It is also possible to use the :command:`ipcontroller` and :command:`ipengine` | |
447 | commands to start your controller and engines. This approach gives you full |
|
448 | commands to start your controller and engines. This approach gives you full | |
448 | control over all aspects of the startup process. |
|
449 | control over all aspects of the startup process. | |
449 |
|
450 | |||
450 | Starting the controller and engine on your local machine |
|
451 | Starting the controller and engine on your local machine | |
451 | -------------------------------------------------------- |
|
452 | -------------------------------------------------------- | |
452 |
|
453 | |||
453 | To use :command:`ipcontroller` and :command:`ipengine` to start things on your |
|
454 | To use :command:`ipcontroller` and :command:`ipengine` to start things on your | |
454 | local machine, do the following. |
|
455 | local machine, do the following. | |
455 |
|
456 | |||
456 | First start the controller:: |
|
457 | First start the controller:: | |
457 |
|
458 | |||
458 | $ ipcontroller |
|
459 | $ ipcontroller | |
459 |
|
460 | |||
460 | Next, start however many instances of the engine you want using (repeatedly) |
|
461 | Next, start however many instances of the engine you want using (repeatedly) | |
461 | the command:: |
|
462 | the command:: | |
462 |
|
463 | |||
463 | $ ipengine |
|
464 | $ ipengine | |
464 |
|
465 | |||
465 | The engines should start and automatically connect to the controller using the |
|
466 | The engines should start and automatically connect to the controller using the | |
466 | JSON files in :file:`~/.ipython/profile_default/security`. You are now ready to use the |
|
467 | JSON files in :file:`~/.ipython/profile_default/security`. You are now ready to use the | |
467 | controller and engines from IPython. |
|
468 | controller and engines from IPython. | |
468 |
|
469 | |||
469 | .. warning:: |
|
470 | .. warning:: | |
470 |
|
471 | |||
471 | The order of the above operations may be important. You *must* |
|
472 | The order of the above operations may be important. You *must* | |
472 | start the controller before the engines, unless you are reusing connection |
|
473 | start the controller before the engines, unless you are reusing connection | |
473 | information (via ``--reuse``), in which case ordering is not important. |
|
474 | information (via ``--reuse``), in which case ordering is not important. | |
474 |
|
475 | |||
475 | .. note:: |
|
476 | .. note:: | |
476 |
|
477 | |||
477 | On some platforms (OS X), to put the controller and engine into the |
|
478 | On some platforms (OS X), to put the controller and engine into the | |
478 | background you may need to give these commands in the form ``(ipcontroller |
|
479 | background you may need to give these commands in the form ``(ipcontroller | |
479 | &)`` and ``(ipengine &)`` (with the parentheses) for them to work |
|
480 | &)`` and ``(ipengine &)`` (with the parentheses) for them to work | |
480 | properly. |
|
481 | properly. | |
481 |
|
482 | |||
482 | Starting the controller and engines on different hosts |
|
483 | Starting the controller and engines on different hosts | |
483 | ------------------------------------------------------ |
|
484 | ------------------------------------------------------ | |
484 |
|
485 | |||
485 | When the controller and engines are running on different hosts, things are |
|
486 | When the controller and engines are running on different hosts, things are | |
486 | slightly more complicated, but the underlying ideas are the same: |
|
487 | slightly more complicated, but the underlying ideas are the same: | |
487 |
|
488 | |||
488 | 1. Start the controller on a host using :command:`ipcontroller`. The controller must be |
|
489 | 1. Start the controller on a host using :command:`ipcontroller`. The controller must be | |
489 | instructed to listen on an interface visible to the engine machines, via the ``ip`` |
|
490 | instructed to listen on an interface visible to the engine machines, via the ``ip`` | |
490 |
command-line argument or ``HubFactory.ip`` in :file:`ipcontroller_config.py` |
|
491 | command-line argument or ``HubFactory.ip`` in :file:`ipcontroller_config.py`:: | |
|
492 | ||||
|
493 | $ ipcontroller --ip=192.168.1.16 | |||
|
494 | ||||
|
495 | .. sourcecode:: python | |||
|
496 | ||||
|
497 | # in ipcontroller_config.py | |||
|
498 | HubFactory.ip = '192.168.1.16' | |||
|
499 | ||||
491 | 2. Copy :file:`ipcontroller-engine.json` from :file:`~/.ipython/profile_<name>/security` on |
|
500 | 2. Copy :file:`ipcontroller-engine.json` from :file:`~/.ipython/profile_<name>/security` on | |
492 | the controller's host to the host where the engines will run. |
|
501 | the controller's host to the host where the engines will run. | |
493 | 3. Use :command:`ipengine` on the engine's hosts to start the engines. |
|
502 | 3. Use :command:`ipengine` on the engine's hosts to start the engines. | |
494 |
|
503 | |||
495 | The only thing you have to be careful of is to tell :command:`ipengine` where |
|
504 | The only thing you have to be careful of is to tell :command:`ipengine` where | |
496 | the :file:`ipcontroller-engine.json` file is located. There are two ways you |
|
505 | the :file:`ipcontroller-engine.json` file is located. There are two ways you | |
497 | can do this: |
|
506 | can do this: | |
498 |
|
507 | |||
499 | * Put :file:`ipcontroller-engine.json` in the :file:`~/.ipython/profile_<name>/security` |
|
508 | * Put :file:`ipcontroller-engine.json` in the :file:`~/.ipython/profile_<name>/security` | |
500 | directory on the engine's host, where it will be found automatically. |
|
509 | directory on the engine's host, where it will be found automatically. | |
501 | * Call :command:`ipengine` with the ``--file=full_path_to_the_file`` |
|
510 | * Call :command:`ipengine` with the ``--file=full_path_to_the_file`` | |
502 | flag. |
|
511 | flag. | |
503 |
|
512 | |||
504 | The ``file`` flag works like this:: |
|
513 | The ``file`` flag works like this:: | |
505 |
|
514 | |||
506 | $ ipengine --file=/path/to/my/ipcontroller-engine.json |
|
515 | $ ipengine --file=/path/to/my/ipcontroller-engine.json | |
507 |
|
516 | |||
508 | .. note:: |
|
517 | .. note:: | |
509 |
|
518 | |||
510 | If the controller's and engine's hosts all have a shared file system |
|
519 | If the controller's and engine's hosts all have a shared file system | |
511 | (:file:`~/.ipython/profile_<name>/security` is the same on all of them), then things |
|
520 | (:file:`~/.ipython/profile_<name>/security` is the same on all of them), then things | |
512 | will just work! |
|
521 | will just work! | |
513 |
|
522 | |||
514 | SSH Tunnels |
|
523 | SSH Tunnels | |
515 | *********** |
|
524 | *********** | |
516 |
|
525 | |||
517 | If your engines are not on the same LAN as the controller, or you are on a highly |
|
526 | If your engines are not on the same LAN as the controller, or you are on a highly | |
518 | restricted network where your nodes cannot see each others ports, then you can |
|
527 | restricted network where your nodes cannot see each others ports, then you can | |
519 | use SSH tunnels to connect engines to the controller. |
|
528 | use SSH tunnels to connect engines to the controller. | |
520 |
|
529 | |||
521 | .. note:: |
|
530 | .. note:: | |
522 |
|
531 | |||
523 | This does not work in all cases. Manual tunnels may be an option, but are |
|
532 | This does not work in all cases. Manual tunnels may be an option, but are | |
524 | highly inconvenient. Support for manual tunnels will be improved. |
|
533 | highly inconvenient. Support for manual tunnels will be improved. | |
525 |
|
534 | |||
526 | You can instruct all engines to use ssh, by specifying the ssh server in |
|
535 | You can instruct all engines to use ssh, by specifying the ssh server in | |
527 | :file:`ipcontroller-engine.json`: |
|
536 | :file:`ipcontroller-engine.json`: | |
528 |
|
537 | |||
529 | .. I know this is really JSON, but the example is a subset of Python: |
|
538 | .. I know this is really JSON, but the example is a subset of Python: | |
530 | .. sourcecode:: python |
|
539 | .. sourcecode:: python | |
531 |
|
540 | |||
532 | { |
|
541 | { | |
533 | "url":"tcp://192.168.1.123:56951", |
|
542 | "url":"tcp://192.168.1.123:56951", | |
534 | "exec_key":"26f4c040-587d-4a4e-b58b-030b96399584", |
|
543 | "exec_key":"26f4c040-587d-4a4e-b58b-030b96399584", | |
535 | "ssh":"user@example.com", |
|
544 | "ssh":"user@example.com", | |
536 | "location":"192.168.1.123" |
|
545 | "location":"192.168.1.123" | |
537 | } |
|
546 | } | |
538 |
|
547 | |||
539 | This will be specified if you give the ``--enginessh=use@example.com`` argument when |
|
548 | This will be specified if you give the ``--enginessh=use@example.com`` argument when | |
540 | starting :command:`ipcontroller`. |
|
549 | starting :command:`ipcontroller`. | |
541 |
|
550 | |||
542 | Or you can specify an ssh server on the command-line when starting an engine:: |
|
551 | Or you can specify an ssh server on the command-line when starting an engine:: | |
543 |
|
552 | |||
544 | $> ipengine --profile=foo --ssh=my.login.node |
|
553 | $> ipengine --profile=foo --ssh=my.login.node | |
545 |
|
554 | |||
546 | For example, if your system is totally restricted, then all connections will actually be |
|
555 | For example, if your system is totally restricted, then all connections will actually be | |
547 | loopback, and ssh tunnels will be used to connect engines to the controller:: |
|
556 | loopback, and ssh tunnels will be used to connect engines to the controller:: | |
548 |
|
557 | |||
549 | [node1] $> ipcontroller --enginessh=node1 |
|
558 | [node1] $> ipcontroller --enginessh=node1 | |
550 | [node2] $> ipengine |
|
559 | [node2] $> ipengine | |
551 | [node3] $> ipcluster engines --n=4 |
|
560 | [node3] $> ipcluster engines --n=4 | |
552 |
|
561 | |||
553 | Or if you want to start many engines on each node, the command `ipcluster engines --n=4` |
|
562 | Or if you want to start many engines on each node, the command `ipcluster engines --n=4` | |
554 | without any configuration is equivalent to running ipengine 4 times. |
|
563 | without any configuration is equivalent to running ipengine 4 times. | |
555 |
|
564 | |||
|
565 | An example using ipcontroller/engine with ssh | |||
|
566 | --------------------------------------------- | |||
|
567 | ||||
|
568 | No configuration files are necessary to use ipcontroller/engine in an SSH environment | |||
|
569 | without a shared filesystem. You simply need to make sure that the controller is listening | |||
|
570 | on an interface visible to the engines, and move the connection file from the controller to | |||
|
571 | the engines. | |||
|
572 | ||||
|
573 | 1. start the controller, listening on an ip-address visible to the engine machines:: | |||
|
574 | ||||
|
575 | [controller.host] $ ipcontroller --ip=192.168.1.16 | |||
|
576 | ||||
|
577 | [IPControllerApp] Using existing profile dir: u'/Users/me/.ipython/profile_default' | |||
|
578 | [IPControllerApp] Hub listening on tcp://192.168.1.16:63320 for registration. | |||
|
579 | [IPControllerApp] Hub using DB backend: 'IPython.parallel.controller.dictdb.DictDB' | |||
|
580 | [IPControllerApp] hub::created hub | |||
|
581 | [IPControllerApp] writing connection info to /Users/me/.ipython/profile_default/security/ipcontroller-client.json | |||
|
582 | [IPControllerApp] writing connection info to /Users/me/.ipython/profile_default/security/ipcontroller-engine.json | |||
|
583 | [IPControllerApp] task::using Python leastload Task scheduler | |||
|
584 | [IPControllerApp] Heartmonitor started | |||
|
585 | [IPControllerApp] Creating pid file: /Users/me/.ipython/profile_default/pid/ipcontroller.pid | |||
|
586 | Scheduler started [leastload] | |||
|
587 | ||||
|
588 | 2. on each engine, fetch the connection file with scp:: | |||
|
589 | ||||
|
590 | [engine.host.n] $ scp controller.host:.ipython/profile_default/security/ipcontroller-engine.json ./ | |||
|
591 | ||||
|
592 | .. note:: | |||
|
593 | ||||
|
594 | The log output of ipcontroller above shows you where the json files were written. | |||
|
595 | They will be in :file:`~/.ipython` (or :file:`~/.config/ipython`) under | |||
|
596 | :file:`profile_default/security/ipcontroller-engine.json` | |||
|
597 | ||||
|
598 | 3. start the engines, using the connection file:: | |||
|
599 | ||||
|
600 | [engine.host.n] $ ipengine --file=./ipcontroller-engine.json | |||
|
601 | ||||
|
602 | A couple of notes: | |||
|
603 | ||||
|
604 | * You can avoid having to fetch the connection file every time by adding ``--reuse`` flag | |||
|
605 | to ipcontroller, which instructs the controller to read the previous connection file for | |||
|
606 | connection info, rather than generate a new one with randomized ports. | |||
|
607 | ||||
|
608 | * In step 2, if you fetch the connection file directly into the security dir of a profile, | |||
|
609 | then you need not specify its path directly, only the profile (assumes the path exists, | |||
|
610 | otherwise you must create it first):: | |||
|
611 | ||||
|
612 | [engine.host.n] $ scp controller.host:.ipython/profile_default/security/ipcontroller-engine.json ~/.ipython/profile_ssh/security/ | |||
|
613 | [engine.host.n] $ ipengine --profile=ssh | |||
|
614 | ||||
|
615 | Of course, if you fetch the file into the default profile, no arguments must be passed to | |||
|
616 | ipengine at all. | |||
|
617 | ||||
|
618 | * Note that ipengine *did not* specify the ip argument. In general, it is unlikely for any | |||
|
619 | connection information to be specified at the command-line to ipengine, as all of this | |||
|
620 | information should be contained in the connection file written by ipcontroller. | |||
556 |
|
621 | |||
557 | Make JSON files persistent |
|
622 | Make JSON files persistent | |
558 | -------------------------- |
|
623 | -------------------------- | |
559 |
|
624 | |||
560 | At fist glance it may seem that that managing the JSON files is a bit |
|
625 | At fist glance it may seem that that managing the JSON files is a bit | |
561 | annoying. Going back to the house and key analogy, copying the JSON around |
|
626 | annoying. Going back to the house and key analogy, copying the JSON around | |
562 | each time you start the controller is like having to make a new key every time |
|
627 | each time you start the controller is like having to make a new key every time | |
563 | you want to unlock the door and enter your house. As with your house, you want |
|
628 | you want to unlock the door and enter your house. As with your house, you want | |
564 | to be able to create the key (or JSON file) once, and then simply use it at |
|
629 | to be able to create the key (or JSON file) once, and then simply use it at | |
565 | any point in the future. |
|
630 | any point in the future. | |
566 |
|
631 | |||
567 | To do this, the only thing you have to do is specify the `--reuse` flag, so that |
|
632 | To do this, the only thing you have to do is specify the `--reuse` flag, so that | |
568 | the connection information in the JSON files remains accurate:: |
|
633 | the connection information in the JSON files remains accurate:: | |
569 |
|
634 | |||
570 | $ ipcontroller --reuse |
|
635 | $ ipcontroller --reuse | |
571 |
|
636 | |||
572 | Then, just copy the JSON files over the first time and you are set. You can |
|
637 | Then, just copy the JSON files over the first time and you are set. You can | |
573 | start and stop the controller and engines any many times as you want in the |
|
638 | start and stop the controller and engines any many times as you want in the | |
574 | future, just make sure to tell the controller to reuse the file. |
|
639 | future, just make sure to tell the controller to reuse the file. | |
575 |
|
640 | |||
576 | .. note:: |
|
641 | .. note:: | |
577 |
|
642 | |||
578 | You may ask the question: what ports does the controller listen on if you |
|
643 | You may ask the question: what ports does the controller listen on if you | |
579 | don't tell is to use specific ones? The default is to use high random port |
|
644 | don't tell is to use specific ones? The default is to use high random port | |
580 | numbers. We do this for two reasons: i) to increase security through |
|
645 | numbers. We do this for two reasons: i) to increase security through | |
581 | obscurity and ii) to multiple controllers on a given host to start and |
|
646 | obscurity and ii) to multiple controllers on a given host to start and | |
582 | automatically use different ports. |
|
647 | automatically use different ports. | |
583 |
|
648 | |||
584 | Log files |
|
649 | Log files | |
585 | --------- |
|
650 | --------- | |
586 |
|
651 | |||
587 | All of the components of IPython have log files associated with them. |
|
652 | All of the components of IPython have log files associated with them. | |
588 | These log files can be extremely useful in debugging problems with |
|
653 | These log files can be extremely useful in debugging problems with | |
589 | IPython and can be found in the directory :file:`~/.ipython/profile_<name>/log`. |
|
654 | IPython and can be found in the directory :file:`~/.ipython/profile_<name>/log`. | |
590 | Sending the log files to us will often help us to debug any problems. |
|
655 | Sending the log files to us will often help us to debug any problems. | |
591 |
|
656 | |||
592 |
|
657 | |||
593 | Configuring `ipcontroller` |
|
658 | Configuring `ipcontroller` | |
594 | --------------------------- |
|
659 | --------------------------- | |
595 |
|
660 | |||
596 | The IPython Controller takes its configuration from the file :file:`ipcontroller_config.py` |
|
661 | The IPython Controller takes its configuration from the file :file:`ipcontroller_config.py` | |
597 | in the active profile directory. |
|
662 | in the active profile directory. | |
598 |
|
663 | |||
599 | Ports and addresses |
|
664 | Ports and addresses | |
600 | ******************* |
|
665 | ******************* | |
601 |
|
666 | |||
602 | In many cases, you will want to configure the Controller's network identity. By default, |
|
667 | In many cases, you will want to configure the Controller's network identity. By default, | |
603 | the Controller listens only on loopback, which is the most secure but often impractical. |
|
668 | the Controller listens only on loopback, which is the most secure but often impractical. | |
604 | To instruct the controller to listen on a specific interface, you can set the |
|
669 | To instruct the controller to listen on a specific interface, you can set the | |
605 | :attr:`HubFactory.ip` trait. To listen on all interfaces, simply specify: |
|
670 | :attr:`HubFactory.ip` trait. To listen on all interfaces, simply specify: | |
606 |
|
671 | |||
607 | .. sourcecode:: python |
|
672 | .. sourcecode:: python | |
608 |
|
673 | |||
609 | c.HubFactory.ip = '*' |
|
674 | c.HubFactory.ip = '*' | |
610 |
|
675 | |||
611 | When connecting to a Controller that is listening on loopback or behind a firewall, it may |
|
676 | When connecting to a Controller that is listening on loopback or behind a firewall, it may | |
612 | be necessary to specify an SSH server to use for tunnels, and the external IP of the |
|
677 | be necessary to specify an SSH server to use for tunnels, and the external IP of the | |
613 | Controller. If you specified that the HubFactory listen on loopback, or all interfaces, |
|
678 | Controller. If you specified that the HubFactory listen on loopback, or all interfaces, | |
614 | then IPython will try to guess the external IP. If you are on a system with VM network |
|
679 | then IPython will try to guess the external IP. If you are on a system with VM network | |
615 | devices, or many interfaces, this guess may be incorrect. In these cases, you will want |
|
680 | devices, or many interfaces, this guess may be incorrect. In these cases, you will want | |
616 | to specify the 'location' of the Controller. This is the IP of the machine the Controller |
|
681 | to specify the 'location' of the Controller. This is the IP of the machine the Controller | |
617 | is on, as seen by the clients, engines, or the SSH server used to tunnel connections. |
|
682 | is on, as seen by the clients, engines, or the SSH server used to tunnel connections. | |
618 |
|
683 | |||
619 | For example, to set up a cluster with a Controller on a work node, using ssh tunnels |
|
684 | For example, to set up a cluster with a Controller on a work node, using ssh tunnels | |
620 | through the login node, an example :file:`ipcontroller_config.py` might contain: |
|
685 | through the login node, an example :file:`ipcontroller_config.py` might contain: | |
621 |
|
686 | |||
622 | .. sourcecode:: python |
|
687 | .. sourcecode:: python | |
623 |
|
688 | |||
624 | # allow connections on all interfaces from engines |
|
689 | # allow connections on all interfaces from engines | |
625 | # engines on the same node will use loopback, while engines |
|
690 | # engines on the same node will use loopback, while engines | |
626 | # from other nodes will use an external IP |
|
691 | # from other nodes will use an external IP | |
627 | c.HubFactory.ip = '*' |
|
692 | c.HubFactory.ip = '*' | |
628 |
|
693 | |||
629 | # you typically only need to specify the location when there are extra |
|
694 | # you typically only need to specify the location when there are extra | |
630 | # interfaces that may not be visible to peer nodes (e.g. VM interfaces) |
|
695 | # interfaces that may not be visible to peer nodes (e.g. VM interfaces) | |
631 | c.HubFactory.location = '10.0.1.5' |
|
696 | c.HubFactory.location = '10.0.1.5' | |
632 | # or to get an automatic value, try this: |
|
697 | # or to get an automatic value, try this: | |
633 | import socket |
|
698 | import socket | |
634 | ex_ip = socket.gethostbyname_ex(socket.gethostname())[-1][0] |
|
699 | ex_ip = socket.gethostbyname_ex(socket.gethostname())[-1][0] | |
635 | c.HubFactory.location = ex_ip |
|
700 | c.HubFactory.location = ex_ip | |
636 |
|
701 | |||
637 | # now instruct clients to use the login node for SSH tunnels: |
|
702 | # now instruct clients to use the login node for SSH tunnels: | |
638 | c.HubFactory.ssh_server = 'login.mycluster.net' |
|
703 | c.HubFactory.ssh_server = 'login.mycluster.net' | |
639 |
|
704 | |||
640 | After doing this, your :file:`ipcontroller-client.json` file will look something like this: |
|
705 | After doing this, your :file:`ipcontroller-client.json` file will look something like this: | |
641 |
|
706 | |||
642 | .. this can be Python, despite the fact that it's actually JSON, because it's |
|
707 | .. this can be Python, despite the fact that it's actually JSON, because it's | |
643 | .. still valid Python |
|
708 | .. still valid Python | |
644 |
|
709 | |||
645 | .. sourcecode:: python |
|
710 | .. sourcecode:: python | |
646 |
|
711 | |||
647 | { |
|
712 | { | |
648 | "url":"tcp:\/\/*:43447", |
|
713 | "url":"tcp:\/\/*:43447", | |
649 | "exec_key":"9c7779e4-d08a-4c3b-ba8e-db1f80b562c1", |
|
714 | "exec_key":"9c7779e4-d08a-4c3b-ba8e-db1f80b562c1", | |
650 | "ssh":"login.mycluster.net", |
|
715 | "ssh":"login.mycluster.net", | |
651 | "location":"10.0.1.5" |
|
716 | "location":"10.0.1.5" | |
652 | } |
|
717 | } | |
653 |
|
718 | |||
654 | Then this file will be all you need for a client to connect to the controller, tunneling |
|
719 | Then this file will be all you need for a client to connect to the controller, tunneling | |
655 | SSH connections through login.mycluster.net. |
|
720 | SSH connections through login.mycluster.net. | |
656 |
|
721 | |||
657 | Database Backend |
|
722 | Database Backend | |
658 | **************** |
|
723 | **************** | |
659 |
|
724 | |||
660 | The Hub stores all messages and results passed between Clients and Engines. |
|
725 | The Hub stores all messages and results passed between Clients and Engines. | |
661 | For large and/or long-running clusters, it would be unreasonable to keep all |
|
726 | For large and/or long-running clusters, it would be unreasonable to keep all | |
662 | of this information in memory. For this reason, we have two database backends: |
|
727 | of this information in memory. For this reason, we have two database backends: | |
663 | [MongoDB]_ via PyMongo_, and SQLite with the stdlib :py:mod:`sqlite`. |
|
728 | [MongoDB]_ via PyMongo_, and SQLite with the stdlib :py:mod:`sqlite`. | |
664 |
|
729 | |||
665 | MongoDB is our design target, and the dict-like model it uses has driven our design. As far |
|
730 | MongoDB is our design target, and the dict-like model it uses has driven our design. As far | |
666 | as we are concerned, BSON can be considered essentially the same as JSON, adding support |
|
731 | as we are concerned, BSON can be considered essentially the same as JSON, adding support | |
667 | for binary data and datetime objects, and any new database backend must support the same |
|
732 | for binary data and datetime objects, and any new database backend must support the same | |
668 | data types. |
|
733 | data types. | |
669 |
|
734 | |||
670 | .. seealso:: |
|
735 | .. seealso:: | |
671 |
|
736 | |||
672 | MongoDB `BSON doc <http://www.mongodb.org/display/DOCS/BSON>`_ |
|
737 | MongoDB `BSON doc <http://www.mongodb.org/display/DOCS/BSON>`_ | |
673 |
|
738 | |||
674 | To use one of these backends, you must set the :attr:`HubFactory.db_class` trait: |
|
739 | To use one of these backends, you must set the :attr:`HubFactory.db_class` trait: | |
675 |
|
740 | |||
676 | .. sourcecode:: python |
|
741 | .. sourcecode:: python | |
677 |
|
742 | |||
678 | # for a simple dict-based in-memory implementation, use dictdb |
|
743 | # for a simple dict-based in-memory implementation, use dictdb | |
679 | # This is the default and the fastest, since it doesn't involve the filesystem |
|
744 | # This is the default and the fastest, since it doesn't involve the filesystem | |
680 | c.HubFactory.db_class = 'IPython.parallel.controller.dictdb.DictDB' |
|
745 | c.HubFactory.db_class = 'IPython.parallel.controller.dictdb.DictDB' | |
681 |
|
746 | |||
682 | # To use MongoDB: |
|
747 | # To use MongoDB: | |
683 | c.HubFactory.db_class = 'IPython.parallel.controller.mongodb.MongoDB' |
|
748 | c.HubFactory.db_class = 'IPython.parallel.controller.mongodb.MongoDB' | |
684 |
|
749 | |||
685 | # and SQLite: |
|
750 | # and SQLite: | |
686 | c.HubFactory.db_class = 'IPython.parallel.controller.sqlitedb.SQLiteDB' |
|
751 | c.HubFactory.db_class = 'IPython.parallel.controller.sqlitedb.SQLiteDB' | |
687 |
|
752 | |||
688 | When using the proper databases, you can actually allow for tasks to persist from |
|
753 | When using the proper databases, you can actually allow for tasks to persist from | |
689 | one session to the next by specifying the MongoDB database or SQLite table in |
|
754 | one session to the next by specifying the MongoDB database or SQLite table in | |
690 | which tasks are to be stored. The default is to use a table named for the Hub's Session, |
|
755 | which tasks are to be stored. The default is to use a table named for the Hub's Session, | |
691 | which is a UUID, and thus different every time. |
|
756 | which is a UUID, and thus different every time. | |
692 |
|
757 | |||
693 | .. sourcecode:: python |
|
758 | .. sourcecode:: python | |
694 |
|
759 | |||
695 | # To keep persistant task history in MongoDB: |
|
760 | # To keep persistant task history in MongoDB: | |
696 | c.MongoDB.database = 'tasks' |
|
761 | c.MongoDB.database = 'tasks' | |
697 |
|
762 | |||
698 | # and in SQLite: |
|
763 | # and in SQLite: | |
699 | c.SQLiteDB.table = 'tasks' |
|
764 | c.SQLiteDB.table = 'tasks' | |
700 |
|
765 | |||
701 |
|
766 | |||
702 | Since MongoDB servers can be running remotely or configured to listen on a particular port, |
|
767 | Since MongoDB servers can be running remotely or configured to listen on a particular port, | |
703 | you can specify any arguments you may need to the PyMongo `Connection |
|
768 | you can specify any arguments you may need to the PyMongo `Connection | |
704 | <http://api.mongodb.org/python/1.9/api/pymongo/connection.html#pymongo.connection.Connection>`_: |
|
769 | <http://api.mongodb.org/python/1.9/api/pymongo/connection.html#pymongo.connection.Connection>`_: | |
705 |
|
770 | |||
706 | .. sourcecode:: python |
|
771 | .. sourcecode:: python | |
707 |
|
772 | |||
708 | # positional args to pymongo.Connection |
|
773 | # positional args to pymongo.Connection | |
709 | c.MongoDB.connection_args = [] |
|
774 | c.MongoDB.connection_args = [] | |
710 |
|
775 | |||
711 | # keyword args to pymongo.Connection |
|
776 | # keyword args to pymongo.Connection | |
712 | c.MongoDB.connection_kwargs = {} |
|
777 | c.MongoDB.connection_kwargs = {} | |
713 |
|
778 | |||
714 | .. _MongoDB: http://www.mongodb.org |
|
779 | .. _MongoDB: http://www.mongodb.org | |
715 | .. _PyMongo: http://api.mongodb.org/python/1.9/ |
|
780 | .. _PyMongo: http://api.mongodb.org/python/1.9/ | |
716 |
|
781 | |||
717 | Configuring `ipengine` |
|
782 | Configuring `ipengine` | |
718 | ----------------------- |
|
783 | ----------------------- | |
719 |
|
784 | |||
720 | The IPython Engine takes its configuration from the file :file:`ipengine_config.py` |
|
785 | The IPython Engine takes its configuration from the file :file:`ipengine_config.py` | |
721 |
|
786 | |||
722 | The Engine itself also has some amount of configuration. Most of this |
|
787 | The Engine itself also has some amount of configuration. Most of this | |
723 | has to do with initializing MPI or connecting to the controller. |
|
788 | has to do with initializing MPI or connecting to the controller. | |
724 |
|
789 | |||
725 | To instruct the Engine to initialize with an MPI environment set up by |
|
790 | To instruct the Engine to initialize with an MPI environment set up by | |
726 | mpi4py, add: |
|
791 | mpi4py, add: | |
727 |
|
792 | |||
728 | .. sourcecode:: python |
|
793 | .. sourcecode:: python | |
729 |
|
794 | |||
730 | c.MPI.use = 'mpi4py' |
|
795 | c.MPI.use = 'mpi4py' | |
731 |
|
796 | |||
732 | In this case, the Engine will use our default mpi4py init script to set up |
|
797 | In this case, the Engine will use our default mpi4py init script to set up | |
733 | the MPI environment prior to exection. We have default init scripts for |
|
798 | the MPI environment prior to exection. We have default init scripts for | |
734 | mpi4py and pytrilinos. If you want to specify your own code to be run |
|
799 | mpi4py and pytrilinos. If you want to specify your own code to be run | |
735 | at the beginning, specify `c.MPI.init_script`. |
|
800 | at the beginning, specify `c.MPI.init_script`. | |
736 |
|
801 | |||
737 | You can also specify a file or python command to be run at startup of the |
|
802 | You can also specify a file or python command to be run at startup of the | |
738 | Engine: |
|
803 | Engine: | |
739 |
|
804 | |||
740 | .. sourcecode:: python |
|
805 | .. sourcecode:: python | |
741 |
|
806 | |||
742 | c.IPEngineApp.startup_script = u'/path/to/my/startup.py' |
|
807 | c.IPEngineApp.startup_script = u'/path/to/my/startup.py' | |
743 |
|
808 | |||
744 | c.IPEngineApp.startup_command = 'import numpy, scipy, mpi4py' |
|
809 | c.IPEngineApp.startup_command = 'import numpy, scipy, mpi4py' | |
745 |
|
810 | |||
746 | These commands/files will be run again, after each |
|
811 | These commands/files will be run again, after each | |
747 |
|
812 | |||
748 | It's also useful on systems with shared filesystems to run the engines |
|
813 | It's also useful on systems with shared filesystems to run the engines | |
749 | in some scratch directory. This can be set with: |
|
814 | in some scratch directory. This can be set with: | |
750 |
|
815 | |||
751 | .. sourcecode:: python |
|
816 | .. sourcecode:: python | |
752 |
|
817 | |||
753 | c.IPEngineApp.work_dir = u'/path/to/scratch/' |
|
818 | c.IPEngineApp.work_dir = u'/path/to/scratch/' | |
754 |
|
819 | |||
755 |
|
820 | |||
756 |
|
821 | |||
757 | .. [MongoDB] MongoDB database http://www.mongodb.org |
|
822 | .. [MongoDB] MongoDB database http://www.mongodb.org | |
758 |
|
823 | |||
759 | .. [PBS] Portable Batch System http://www.openpbs.org |
|
824 | .. [PBS] Portable Batch System http://www.openpbs.org | |
760 |
|
825 | |||
761 | .. [SSH] SSH-Agent http://en.wikipedia.org/wiki/ssh-agent |
|
826 | .. [SSH] SSH-Agent http://en.wikipedia.org/wiki/ssh-agent |
@@ -1,332 +1,360 b'' | |||||
1 | ============================================ |
|
1 | ============================================ | |
2 | Getting started with Windows HPC Server 2008 |
|
2 | Getting started with Windows HPC Server 2008 | |
3 | ============================================ |
|
3 | ============================================ | |
4 |
|
4 | |||
5 | .. note:: |
|
|||
6 |
|
||||
7 | Not adapted to zmq yet |
|
|||
8 |
|
||||
9 | Introduction |
|
5 | Introduction | |
10 | ============ |
|
6 | ============ | |
11 |
|
7 | |||
12 | The Python programming language is an increasingly popular language for |
|
8 | The Python programming language is an increasingly popular language for | |
13 | numerical computing. This is due to a unique combination of factors. First, |
|
9 | numerical computing. This is due to a unique combination of factors. First, | |
14 | Python is a high-level and *interactive* language that is well matched to |
|
10 | Python is a high-level and *interactive* language that is well matched to | |
15 | interactive numerical work. Second, it is easy (often times trivial) to |
|
11 | interactive numerical work. Second, it is easy (often times trivial) to | |
16 | integrate legacy C/C++/Fortran code into Python. Third, a large number of |
|
12 | integrate legacy C/C++/Fortran code into Python. Third, a large number of | |
17 | high-quality open source projects provide all the needed building blocks for |
|
13 | high-quality open source projects provide all the needed building blocks for | |
18 | numerical computing: numerical arrays (NumPy), algorithms (SciPy), 2D/3D |
|
14 | numerical computing: numerical arrays (NumPy), algorithms (SciPy), 2D/3D | |
19 | Visualization (Matplotlib, Mayavi, Chaco), Symbolic Mathematics (Sage, Sympy) |
|
15 | Visualization (Matplotlib, Mayavi, Chaco), Symbolic Mathematics (Sage, Sympy) | |
20 | and others. |
|
16 | and others. | |
21 |
|
17 | |||
22 | The IPython project is a core part of this open-source toolchain and is |
|
18 | The IPython project is a core part of this open-source toolchain and is | |
23 | focused on creating a comprehensive environment for interactive and |
|
19 | focused on creating a comprehensive environment for interactive and | |
24 | exploratory computing in the Python programming language. It enables all of |
|
20 | exploratory computing in the Python programming language. It enables all of | |
25 | the above tools to be used interactively and consists of two main components: |
|
21 | the above tools to be used interactively and consists of two main components: | |
26 |
|
22 | |||
27 | * An enhanced interactive Python shell with support for interactive plotting |
|
23 | * An enhanced interactive Python shell with support for interactive plotting | |
28 | and visualization. |
|
24 | and visualization. | |
29 | * An architecture for interactive parallel computing. |
|
25 | * An architecture for interactive parallel computing. | |
30 |
|
26 | |||
31 | With these components, it is possible to perform all aspects of a parallel |
|
27 | With these components, it is possible to perform all aspects of a parallel | |
32 | computation interactively. This type of workflow is particularly relevant in |
|
28 | computation interactively. This type of workflow is particularly relevant in | |
33 | scientific and numerical computing where algorithms, code and data are |
|
29 | scientific and numerical computing where algorithms, code and data are | |
34 | continually evolving as the user/developer explores a problem. The broad |
|
30 | continually evolving as the user/developer explores a problem. The broad | |
35 | treads in computing (commodity clusters, multicore, cloud computing, etc.) |
|
31 | treads in computing (commodity clusters, multicore, cloud computing, etc.) | |
36 | make these capabilities of IPython particularly relevant. |
|
32 | make these capabilities of IPython particularly relevant. | |
37 |
|
33 | |||
38 | While IPython is a cross platform tool, it has particularly strong support for |
|
34 | While IPython is a cross platform tool, it has particularly strong support for | |
39 | Windows based compute clusters running Windows HPC Server 2008. This document |
|
35 | Windows based compute clusters running Windows HPC Server 2008. This document | |
40 | describes how to get started with IPython on Windows HPC Server 2008. The |
|
36 | describes how to get started with IPython on Windows HPC Server 2008. The | |
41 | content and emphasis here is practical: installing IPython, configuring |
|
37 | content and emphasis here is practical: installing IPython, configuring | |
42 | IPython to use the Windows job scheduler and running example parallel programs |
|
38 | IPython to use the Windows job scheduler and running example parallel programs | |
43 | interactively. A more complete description of IPython's parallel computing |
|
39 | interactively. A more complete description of IPython's parallel computing | |
44 | capabilities can be found in IPython's online documentation |
|
40 | capabilities can be found in IPython's online documentation | |
45 | (http://ipython.org/documentation.html). |
|
41 | (http://ipython.org/documentation.html). | |
46 |
|
42 | |||
47 | Setting up your Windows cluster |
|
43 | Setting up your Windows cluster | |
48 | =============================== |
|
44 | =============================== | |
49 |
|
45 | |||
50 | This document assumes that you already have a cluster running Windows |
|
46 | This document assumes that you already have a cluster running Windows | |
51 | HPC Server 2008. Here is a broad overview of what is involved with setting up |
|
47 | HPC Server 2008. Here is a broad overview of what is involved with setting up | |
52 | such a cluster: |
|
48 | such a cluster: | |
53 |
|
49 | |||
54 | 1. Install Windows Server 2008 on the head and compute nodes in the cluster. |
|
50 | 1. Install Windows Server 2008 on the head and compute nodes in the cluster. | |
55 | 2. Setup the network configuration on each host. Each host should have a |
|
51 | 2. Setup the network configuration on each host. Each host should have a | |
56 | static IP address. |
|
52 | static IP address. | |
57 | 3. On the head node, activate the "Active Directory Domain Services" role |
|
53 | 3. On the head node, activate the "Active Directory Domain Services" role | |
58 | and make the head node the domain controller. |
|
54 | and make the head node the domain controller. | |
59 | 4. Join the compute nodes to the newly created Active Directory (AD) domain. |
|
55 | 4. Join the compute nodes to the newly created Active Directory (AD) domain. | |
60 | 5. Setup user accounts in the domain with shared home directories. |
|
56 | 5. Setup user accounts in the domain with shared home directories. | |
61 | 6. Install the HPC Pack 2008 on the head node to create a cluster. |
|
57 | 6. Install the HPC Pack 2008 on the head node to create a cluster. | |
62 | 7. Install the HPC Pack 2008 on the compute nodes. |
|
58 | 7. Install the HPC Pack 2008 on the compute nodes. | |
63 |
|
59 | |||
64 | More details about installing and configuring Windows HPC Server 2008 can be |
|
60 | More details about installing and configuring Windows HPC Server 2008 can be | |
65 | found on the Windows HPC Home Page (http://www.microsoft.com/hpc). Regardless |
|
61 | found on the Windows HPC Home Page (http://www.microsoft.com/hpc). Regardless | |
66 | of what steps you follow to set up your cluster, the remainder of this |
|
62 | of what steps you follow to set up your cluster, the remainder of this | |
67 | document will assume that: |
|
63 | document will assume that: | |
68 |
|
64 | |||
69 | * There are domain users that can log on to the AD domain and submit jobs |
|
65 | * There are domain users that can log on to the AD domain and submit jobs | |
70 | to the cluster scheduler. |
|
66 | to the cluster scheduler. | |
71 | * These domain users have shared home directories. While shared home |
|
67 | * These domain users have shared home directories. While shared home | |
72 | directories are not required to use IPython, they make it much easier to |
|
68 | directories are not required to use IPython, they make it much easier to | |
73 | use IPython. |
|
69 | use IPython. | |
74 |
|
70 | |||
75 | Installation of IPython and its dependencies |
|
71 | Installation of IPython and its dependencies | |
76 | ============================================ |
|
72 | ============================================ | |
77 |
|
73 | |||
78 | IPython and all of its dependencies are freely available and open source. |
|
74 | IPython and all of its dependencies are freely available and open source. | |
79 | These packages provide a powerful and cost-effective approach to numerical and |
|
75 | These packages provide a powerful and cost-effective approach to numerical and | |
80 | scientific computing on Windows. The following dependencies are needed to run |
|
76 | scientific computing on Windows. The following dependencies are needed to run | |
81 | IPython on Windows: |
|
77 | IPython on Windows: | |
82 |
|
78 | |||
83 | * Python 2.6 or 2.7 (http://www.python.org) |
|
79 | * Python 2.6 or 2.7 (http://www.python.org) | |
84 | * pywin32 (http://sourceforge.net/projects/pywin32/) |
|
80 | * pywin32 (http://sourceforge.net/projects/pywin32/) | |
85 | * PyReadline (https://launchpad.net/pyreadline) |
|
81 | * PyReadline (https://launchpad.net/pyreadline) | |
86 | * pyzmq (http://github.com/zeromq/pyzmq/downloads) |
|
82 | * pyzmq (http://github.com/zeromq/pyzmq/downloads) | |
87 | * IPython (http://ipython.org) |
|
83 | * IPython (http://ipython.org) | |
88 |
|
84 | |||
89 | In addition, the following dependencies are needed to run the demos described |
|
85 | In addition, the following dependencies are needed to run the demos described | |
90 | in this document. |
|
86 | in this document. | |
91 |
|
87 | |||
92 | * NumPy and SciPy (http://www.scipy.org) |
|
88 | * NumPy and SciPy (http://www.scipy.org) | |
93 | * Matplotlib (http://matplotlib.sourceforge.net/) |
|
89 | * Matplotlib (http://matplotlib.sourceforge.net/) | |
94 |
|
90 | |||
95 | The easiest way of obtaining these dependencies is through the Enthought |
|
91 | The easiest way of obtaining these dependencies is through the Enthought | |
96 | Python Distribution (EPD) (http://www.enthought.com/products/epd.php). EPD is |
|
92 | Python Distribution (EPD) (http://www.enthought.com/products/epd.php). EPD is | |
97 | produced by Enthought, Inc. and contains all of these packages and others in a |
|
93 | produced by Enthought, Inc. and contains all of these packages and others in a | |
98 | single installer and is available free for academic users. While it is also |
|
94 | single installer and is available free for academic users. While it is also | |
99 | possible to download and install each package individually, this is a tedious |
|
95 | possible to download and install each package individually, this is a tedious | |
100 | process. Thus, we highly recommend using EPD to install these packages on |
|
96 | process. Thus, we highly recommend using EPD to install these packages on | |
101 | Windows. |
|
97 | Windows. | |
102 |
|
98 | |||
103 | Regardless of how you install the dependencies, here are the steps you will |
|
99 | Regardless of how you install the dependencies, here are the steps you will | |
104 | need to follow: |
|
100 | need to follow: | |
105 |
|
101 | |||
106 | 1. Install all of the packages listed above, either individually or using EPD |
|
102 | 1. Install all of the packages listed above, either individually or using EPD | |
107 | on the head node, compute nodes and user workstations. |
|
103 | on the head node, compute nodes and user workstations. | |
108 |
|
104 | |||
109 | 2. Make sure that :file:`C:\\Python27` and :file:`C:\\Python27\\Scripts` are |
|
105 | 2. Make sure that :file:`C:\\Python27` and :file:`C:\\Python27\\Scripts` are | |
110 | in the system :envvar:`%PATH%` variable on each node. |
|
106 | in the system :envvar:`%PATH%` variable on each node. | |
111 |
|
107 | |||
112 | 3. Install the latest development version of IPython. This can be done by |
|
108 | 3. Install the latest development version of IPython. This can be done by | |
113 | downloading the the development version from the IPython website |
|
109 | downloading the the development version from the IPython website | |
114 | (http://ipython.org) and following the installation instructions. |
|
110 | (http://ipython.org) and following the installation instructions. | |
115 |
|
111 | |||
116 | Further details about installing IPython or its dependencies can be found in |
|
112 | Further details about installing IPython or its dependencies can be found in | |
117 | the online IPython documentation (http://ipython.org/documentation.html) |
|
113 | the online IPython documentation (http://ipython.org/documentation.html) | |
118 | Once you are finished with the installation, you can try IPython out by |
|
114 | Once you are finished with the installation, you can try IPython out by | |
119 | opening a Windows Command Prompt and typing ``ipython``. This will |
|
115 | opening a Windows Command Prompt and typing ``ipython``. This will | |
120 | start IPython's interactive shell and you should see something like the |
|
116 | start IPython's interactive shell and you should see something like the | |
121 |
following |
|
117 | following:: | |
|
118 | ||||
|
119 | Microsoft Windows [Version 6.0.6001] | |||
|
120 | Copyright (c) 2006 Microsoft Corporation. All rights reserved. | |||
|
121 | ||||
|
122 | Z:\>ipython | |||
|
123 | Python 2.7.2 (default, Jun 12 2011, 15:08:59) [MSC v.1500 32 bit (Intel)] | |||
|
124 | Type "copyright", "credits" or "license" for more information. | |||
|
125 | ||||
|
126 | IPython 0.12.dev -- An enhanced Interactive Python. | |||
|
127 | ? -> Introduction and overview of IPython's features. | |||
|
128 | %quickref -> Quick reference. | |||
|
129 | help -> Python's own help system. | |||
|
130 | object? -> Details about 'object', use 'object??' for extra details. | |||
|
131 | ||||
|
132 | In [1]: | |||
122 |
|
133 | |||
123 | .. image:: figs/ipython_shell.* |
|
|||
124 |
|
134 | |||
125 | Starting an IPython cluster |
|
135 | Starting an IPython cluster | |
126 | =========================== |
|
136 | =========================== | |
127 |
|
137 | |||
128 | To use IPython's parallel computing capabilities, you will need to start an |
|
138 | To use IPython's parallel computing capabilities, you will need to start an | |
129 | IPython cluster. An IPython cluster consists of one controller and multiple |
|
139 | IPython cluster. An IPython cluster consists of one controller and multiple | |
130 | engines: |
|
140 | engines: | |
131 |
|
141 | |||
132 | IPython controller |
|
142 | IPython controller | |
133 | The IPython controller manages the engines and acts as a gateway between |
|
143 | The IPython controller manages the engines and acts as a gateway between | |
134 | the engines and the client, which runs in the user's interactive IPython |
|
144 | the engines and the client, which runs in the user's interactive IPython | |
135 | session. The controller is started using the :command:`ipcontroller` |
|
145 | session. The controller is started using the :command:`ipcontroller` | |
136 | command. |
|
146 | command. | |
137 |
|
147 | |||
138 | IPython engine |
|
148 | IPython engine | |
139 | IPython engines run a user's Python code in parallel on the compute nodes. |
|
149 | IPython engines run a user's Python code in parallel on the compute nodes. | |
140 | Engines are starting using the :command:`ipengine` command. |
|
150 | Engines are starting using the :command:`ipengine` command. | |
141 |
|
151 | |||
142 | Once these processes are started, a user can run Python code interactively and |
|
152 | Once these processes are started, a user can run Python code interactively and | |
143 | in parallel on the engines from within the IPython shell using an appropriate |
|
153 | in parallel on the engines from within the IPython shell using an appropriate | |
144 | client. This includes the ability to interact with, plot and visualize data |
|
154 | client. This includes the ability to interact with, plot and visualize data | |
145 | from the engines. |
|
155 | from the engines. | |
146 |
|
156 | |||
147 | IPython has a command line program called :command:`ipcluster` that automates |
|
157 | IPython has a command line program called :command:`ipcluster` that automates | |
148 | all aspects of starting the controller and engines on the compute nodes. |
|
158 | all aspects of starting the controller and engines on the compute nodes. | |
149 | :command:`ipcluster` has full support for the Windows HPC job scheduler, |
|
159 | :command:`ipcluster` has full support for the Windows HPC job scheduler, | |
150 | meaning that :command:`ipcluster` can use this job scheduler to start the |
|
160 | meaning that :command:`ipcluster` can use this job scheduler to start the | |
151 | controller and engines. In our experience, the Windows HPC job scheduler is |
|
161 | controller and engines. In our experience, the Windows HPC job scheduler is | |
152 | particularly well suited for interactive applications, such as IPython. Once |
|
162 | particularly well suited for interactive applications, such as IPython. Once | |
153 | :command:`ipcluster` is configured properly, a user can start an IPython |
|
163 | :command:`ipcluster` is configured properly, a user can start an IPython | |
154 | cluster from their local workstation almost instantly, without having to log |
|
164 | cluster from their local workstation almost instantly, without having to log | |
155 | on to the head node (as is typically required by Unix based job schedulers). |
|
165 | on to the head node (as is typically required by Unix based job schedulers). | |
156 | This enables a user to move seamlessly between serial and parallel |
|
166 | This enables a user to move seamlessly between serial and parallel | |
157 | computations. |
|
167 | computations. | |
158 |
|
168 | |||
159 | In this section we show how to use :command:`ipcluster` to start an IPython |
|
169 | In this section we show how to use :command:`ipcluster` to start an IPython | |
160 | cluster using the Windows HPC Server 2008 job scheduler. To make sure that |
|
170 | cluster using the Windows HPC Server 2008 job scheduler. To make sure that | |
161 | :command:`ipcluster` is installed and working properly, you should first try |
|
171 | :command:`ipcluster` is installed and working properly, you should first try | |
162 | to start an IPython cluster on your local host. To do this, open a Windows |
|
172 | to start an IPython cluster on your local host. To do this, open a Windows | |
163 | Command Prompt and type the following command:: |
|
173 | Command Prompt and type the following command:: | |
164 |
|
174 | |||
165 |
ipcluster start |
|
175 | ipcluster start -n 2 | |
166 |
|
176 | |||
167 |
You should see a number of messages printed to the screen |
|
177 | You should see a number of messages printed to the screen. | |
168 |
|
|
178 | The result should look something like this:: | |
169 | screenshot: |
|
179 | ||
|
180 | Microsoft Windows [Version 6.1.7600] | |||
|
181 | Copyright (c) 2009 Microsoft Corporation. All rights reserved. | |||
|
182 | ||||
|
183 | Z:\>ipcluster start --profile=mycluster | |||
|
184 | [IPClusterStart] Using existing profile dir: u'\\\\blue\\domainusers$\\bgranger\\.ipython\\profile_mycluster' | |||
|
185 | [IPClusterStart] Starting ipcluster with [daemon=False] | |||
|
186 | [IPClusterStart] Creating pid file: \\blue\domainusers$\bgranger\.ipython\profile_mycluster\pid\ipcluster.pid | |||
|
187 | [IPClusterStart] Writing job description file: \\blue\domainusers$\bgranger\.ipython\profile_mycluster\ipcontroller_job.xml | |||
|
188 | [IPClusterStart] Starting Win HPC Job: job submit /jobfile:\\blue\domainusers$\bgranger\.ipython\profile_mycluster\ipcontroller_job.xml /scheduler:HEADNODE | |||
|
189 | [IPClusterStart] Starting 15 engines | |||
|
190 | [IPClusterStart] Writing job description file: \\blue\domainusers$\bgranger\.ipython\profile_mycluster\ipcontroller_job.xml | |||
|
191 | [IPClusterStart] Starting Win HPC Job: job submit /jobfile:\\blue\domainusers$\bgranger\.ipython\profile_mycluster\ipengineset_job.xml /scheduler:HEADNODE | |||
170 |
|
192 | |||
171 | .. image:: figs/ipcluster_start.* |
|
|||
172 |
|
193 | |||
173 | At this point, the controller and two engines are running on your local host. |
|
194 | At this point, the controller and two engines are running on your local host. | |
174 | This configuration is useful for testing and for situations where you want to |
|
195 | This configuration is useful for testing and for situations where you want to | |
175 | take advantage of multiple cores on your local computer. |
|
196 | take advantage of multiple cores on your local computer. | |
176 |
|
197 | |||
177 | Now that we have confirmed that :command:`ipcluster` is working properly, we |
|
198 | Now that we have confirmed that :command:`ipcluster` is working properly, we | |
178 | describe how to configure and run an IPython cluster on an actual compute |
|
199 | describe how to configure and run an IPython cluster on an actual compute | |
179 | cluster running Windows HPC Server 2008. Here is an outline of the needed |
|
200 | cluster running Windows HPC Server 2008. Here is an outline of the needed | |
180 | steps: |
|
201 | steps: | |
181 |
|
202 | |||
182 |
1. Create a cluster profile using: ``ipython profile create --parallel |
|
203 | 1. Create a cluster profile using: ``ipython profile create mycluster --parallel`` | |
183 |
|
204 | |||
184 | 2. Edit configuration files in the directory :file:`.ipython\\cluster_mycluster` |
|
205 | 2. Edit configuration files in the directory :file:`.ipython\\cluster_mycluster` | |
185 |
|
206 | |||
186 |
3. Start the cluster using: ``ipcluser start profile=mycluster |
|
207 | 3. Start the cluster using: ``ipcluster start --profile=mycluster -n 32`` | |
187 |
|
208 | |||
188 | Creating a cluster profile |
|
209 | Creating a cluster profile | |
189 | -------------------------- |
|
210 | -------------------------- | |
190 |
|
211 | |||
191 | In most cases, you will have to create a cluster profile to use IPython on a |
|
212 | In most cases, you will have to create a cluster profile to use IPython on a | |
192 | cluster. A cluster profile is a name (like "mycluster") that is associated |
|
213 | cluster. A cluster profile is a name (like "mycluster") that is associated | |
193 | with a particular cluster configuration. The profile name is used by |
|
214 | with a particular cluster configuration. The profile name is used by | |
194 | :command:`ipcluster` when working with the cluster. |
|
215 | :command:`ipcluster` when working with the cluster. | |
195 |
|
216 | |||
196 | Associated with each cluster profile is a cluster directory. This cluster |
|
217 | Associated with each cluster profile is a cluster directory. This cluster | |
197 | directory is a specially named directory (typically located in the |
|
218 | directory is a specially named directory (typically located in the | |
198 | :file:`.ipython` subdirectory of your home directory) that contains the |
|
219 | :file:`.ipython` subdirectory of your home directory) that contains the | |
199 | configuration files for a particular cluster profile, as well as log files and |
|
220 | configuration files for a particular cluster profile, as well as log files and | |
200 | security keys. The naming convention for cluster directories is: |
|
221 | security keys. The naming convention for cluster directories is: | |
201 | :file:`profile_<profile name>`. Thus, the cluster directory for a profile named |
|
222 | :file:`profile_<profile name>`. Thus, the cluster directory for a profile named | |
202 | "foo" would be :file:`.ipython\\cluster_foo`. |
|
223 | "foo" would be :file:`.ipython\\cluster_foo`. | |
203 |
|
224 | |||
204 | To create a new cluster profile (named "mycluster") and the associated cluster |
|
225 | To create a new cluster profile (named "mycluster") and the associated cluster | |
205 | directory, type the following command at the Windows Command Prompt:: |
|
226 | directory, type the following command at the Windows Command Prompt:: | |
206 |
|
227 | |||
207 | ipython profile create --parallel --profile=mycluster |
|
228 | ipython profile create --parallel --profile=mycluster | |
208 |
|
229 | |||
209 | The output of this command is shown in the screenshot below. Notice how |
|
230 | The output of this command is shown in the screenshot below. Notice how | |
210 |
:command:`ipcluster` prints out the location of the newly created |
|
231 | :command:`ipcluster` prints out the location of the newly created profile | |
211 |
directory |
|
232 | directory:: | |
|
233 | ||||
|
234 | Z:\>ipython profile create mycluster --parallel | |||
|
235 | [ProfileCreate] Generating default config file: u'\\\\blue\\domainusers$\\bgranger\\.ipython\\profile_mycluster\\ipython_config.py' | |||
|
236 | [ProfileCreate] Generating default config file: u'\\\\blue\\domainusers$\\bgranger\\.ipython\\profile_mycluster\\ipcontroller_config.py' | |||
|
237 | [ProfileCreate] Generating default config file: u'\\\\blue\\domainusers$\\bgranger\\.ipython\\profile_mycluster\\ipengine_config.py' | |||
|
238 | [ProfileCreate] Generating default config file: u'\\\\blue\\domainusers$\\bgranger\\.ipython\\profile_mycluster\\ipcluster_config.py' | |||
|
239 | [ProfileCreate] Generating default config file: u'\\\\blue\\domainusers$\\bgranger\\.ipython\\profile_mycluster\\iplogger_config.py' | |||
212 |
|
240 | |||
213 | .. image:: figs/ipcluster_create.* |
|
241 | Z:\> | |
214 |
|
242 | |||
215 | Configuring a cluster profile |
|
243 | Configuring a cluster profile | |
216 | ----------------------------- |
|
244 | ----------------------------- | |
217 |
|
245 | |||
218 | Next, you will need to configure the newly created cluster profile by editing |
|
246 | Next, you will need to configure the newly created cluster profile by editing | |
219 | the following configuration files in the cluster directory: |
|
247 | the following configuration files in the cluster directory: | |
220 |
|
248 | |||
221 | * :file:`ipcluster_config.py` |
|
249 | * :file:`ipcluster_config.py` | |
222 | * :file:`ipcontroller_config.py` |
|
250 | * :file:`ipcontroller_config.py` | |
223 | * :file:`ipengine_config.py` |
|
251 | * :file:`ipengine_config.py` | |
224 |
|
252 | |||
225 | When :command:`ipcluster` is run, these configuration files are used to |
|
253 | When :command:`ipcluster` is run, these configuration files are used to | |
226 | determine how the engines and controller will be started. In most cases, |
|
254 | determine how the engines and controller will be started. In most cases, | |
227 | you will only have to set a few of the attributes in these files. |
|
255 | you will only have to set a few of the attributes in these files. | |
228 |
|
256 | |||
229 | To configure :command:`ipcluster` to use the Windows HPC job scheduler, you |
|
257 | To configure :command:`ipcluster` to use the Windows HPC job scheduler, you | |
230 | will need to edit the following attributes in the file |
|
258 | will need to edit the following attributes in the file | |
231 | :file:`ipcluster_config.py`:: |
|
259 | :file:`ipcluster_config.py`:: | |
232 |
|
260 | |||
233 | # Set these at the top of the file to tell ipcluster to use the |
|
261 | # Set these at the top of the file to tell ipcluster to use the | |
234 | # Windows HPC job scheduler. |
|
262 | # Windows HPC job scheduler. | |
235 | c.IPClusterStart.controller_launcher_class = 'WindowsHPCControllerLauncher' |
|
263 | c.IPClusterStart.controller_launcher_class = 'WindowsHPCControllerLauncher' | |
236 | c.IPClusterEngines.engine_launcher_class = 'WindowsHPCEngineSetLauncher' |
|
264 | c.IPClusterEngines.engine_launcher_class = 'WindowsHPCEngineSetLauncher' | |
237 |
|
265 | |||
238 | # Set these to the host name of the scheduler (head node) of your cluster. |
|
266 | # Set these to the host name of the scheduler (head node) of your cluster. | |
239 | c.WindowsHPCControllerLauncher.scheduler = 'HEADNODE' |
|
267 | c.WindowsHPCControllerLauncher.scheduler = 'HEADNODE' | |
240 | c.WindowsHPCEngineSetLauncher.scheduler = 'HEADNODE' |
|
268 | c.WindowsHPCEngineSetLauncher.scheduler = 'HEADNODE' | |
241 |
|
269 | |||
242 | There are a number of other configuration attributes that can be set, but |
|
270 | There are a number of other configuration attributes that can be set, but | |
243 | in most cases these will be sufficient to get you started. |
|
271 | in most cases these will be sufficient to get you started. | |
244 |
|
272 | |||
245 | .. warning:: |
|
273 | .. warning:: | |
246 | If any of your configuration attributes involve specifying the location |
|
274 | If any of your configuration attributes involve specifying the location | |
247 | of shared directories or files, you must make sure that you use UNC paths |
|
275 | of shared directories or files, you must make sure that you use UNC paths | |
248 |
like :file:`\\\\host\\share`. It is |
|
276 | like :file:`\\\\host\\share`. It is helpful to specify | |
249 | these paths using raw Python strings: ``r'\\host\share'`` to make sure |
|
277 | these paths using raw Python strings: ``r'\\host\share'`` to make sure | |
250 | that the backslashes are properly escaped. |
|
278 | that the backslashes are properly escaped. | |
251 |
|
279 | |||
252 | Starting the cluster profile |
|
280 | Starting the cluster profile | |
253 | ---------------------------- |
|
281 | ---------------------------- | |
254 |
|
282 | |||
255 | Once a cluster profile has been configured, starting an IPython cluster using |
|
283 | Once a cluster profile has been configured, starting an IPython cluster using | |
256 | the profile is simple:: |
|
284 | the profile is simple:: | |
257 |
|
285 | |||
258 | ipcluster start --profile=mycluster -n 32 |
|
286 | ipcluster start --profile=mycluster -n 32 | |
259 |
|
287 | |||
260 | The ``-n`` option tells :command:`ipcluster` how many engines to start (in |
|
288 | The ``-n`` option tells :command:`ipcluster` how many engines to start (in | |
261 | this case 32). Stopping the cluster is as simple as typing Control-C. |
|
289 | this case 32). Stopping the cluster is as simple as typing Control-C. | |
262 |
|
290 | |||
263 | Using the HPC Job Manager |
|
291 | Using the HPC Job Manager | |
264 | ------------------------- |
|
292 | ------------------------- | |
265 |
|
293 | føø | ||
266 | When ``ipcluster start`` is run the first time, :command:`ipcluster` creates |
|
294 | When ``ipcluster start`` is run the first time, :command:`ipcluster` creates | |
267 | two XML job description files in the cluster directory: |
|
295 | two XML job description files in the cluster directory: | |
268 |
|
296 | |||
269 | * :file:`ipcontroller_job.xml` |
|
297 | * :file:`ipcontroller_job.xml` | |
270 | * :file:`ipengineset_job.xml` |
|
298 | * :file:`ipengineset_job.xml` | |
271 |
|
299 | |||
272 | Once these files have been created, they can be imported into the HPC Job |
|
300 | Once these files have been created, they can be imported into the HPC Job | |
273 | Manager application. Then, the controller and engines for that profile can be |
|
301 | Manager application. Then, the controller and engines for that profile can be | |
274 | started using the HPC Job Manager directly, without using :command:`ipcluster`. |
|
302 | started using the HPC Job Manager directly, without using :command:`ipcluster`. | |
275 | However, anytime the cluster profile is re-configured, ``ipcluster start`` |
|
303 | However, anytime the cluster profile is re-configured, ``ipcluster start`` | |
276 | must be run again to regenerate the XML job description files. The |
|
304 | must be run again to regenerate the XML job description files. The | |
277 | following screenshot shows what the HPC Job Manager interface looks like |
|
305 | following screenshot shows what the HPC Job Manager interface looks like | |
278 | with a running IPython cluster. |
|
306 | with a running IPython cluster. | |
279 |
|
307 | |||
280 | .. image:: figs/hpc_job_manager.* |
|
308 | .. image:: figs/hpc_job_manager.* | |
281 |
|
309 | |||
282 | Performing a simple interactive parallel computation |
|
310 | Performing a simple interactive parallel computation | |
283 | ==================================================== |
|
311 | ==================================================== | |
284 |
|
312 | |||
285 | Once you have started your IPython cluster, you can start to use it. To do |
|
313 | Once you have started your IPython cluster, you can start to use it. To do | |
286 | this, open up a new Windows Command Prompt and start up IPython's interactive |
|
314 | this, open up a new Windows Command Prompt and start up IPython's interactive | |
287 | shell by typing:: |
|
315 | shell by typing:: | |
288 |
|
316 | |||
289 | ipython |
|
317 | ipython | |
290 |
|
318 | |||
291 |
Then you can create a :class:` |
|
319 | Then you can create a :class:`DirectView` instance for your profile and | |
292 | use the resulting instance to do a simple interactive parallel computation. In |
|
320 | use the resulting instance to do a simple interactive parallel computation. In | |
293 | the code and screenshot that follows, we take a simple Python function and |
|
321 | the code and screenshot that follows, we take a simple Python function and | |
294 | apply it to each element of an array of integers in parallel using the |
|
322 | apply it to each element of an array of integers in parallel using the | |
295 |
:meth:` |
|
323 | :meth:`DirectView.map` method: | |
296 |
|
324 | |||
297 | .. sourcecode:: ipython |
|
325 | .. sourcecode:: ipython | |
298 |
|
326 | |||
299 | In [1]: from IPython.parallel import * |
|
327 | In [1]: from IPython.parallel import * | |
300 |
|
328 | |||
301 |
In [2]: c = |
|
329 | In [2]: c = Client(profile='mycluster') | |
|
330 | ||||
|
331 | In [3]: view = c[:] | |||
302 |
|
332 | |||
303 |
In [ |
|
333 | In [4]: c.ids | |
304 |
Out[ |
|
334 | Out[4]: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14] | |
305 |
|
335 | |||
306 |
In [ |
|
336 | In [5]: def f(x): | |
307 | ...: return x**10 |
|
337 | ...: return x**10 | |
308 |
|
338 | |||
309 |
In [ |
|
339 | In [6]: view.map(f, range(15)) # f is applied in parallel | |
310 |
Out[ |
|
340 | Out[6]: | |
311 | [0, |
|
341 | [0, | |
312 | 1, |
|
342 | 1, | |
313 | 1024, |
|
343 | 1024, | |
314 | 59049, |
|
344 | 59049, | |
315 | 1048576, |
|
345 | 1048576, | |
316 | 9765625, |
|
346 | 9765625, | |
317 | 60466176, |
|
347 | 60466176, | |
318 | 282475249, |
|
348 | 282475249, | |
319 | 1073741824, |
|
349 | 1073741824, | |
320 | 3486784401L, |
|
350 | 3486784401L, | |
321 | 10000000000L, |
|
351 | 10000000000L, | |
322 | 25937424601L, |
|
352 | 25937424601L, | |
323 | 61917364224L, |
|
353 | 61917364224L, | |
324 | 137858491849L, |
|
354 | 137858491849L, | |
325 | 289254654976L] |
|
355 | 289254654976L] | |
326 |
|
356 | |||
327 | The :meth:`map` method has the same signature as Python's builtin :func:`map` |
|
357 | The :meth:`map` method has the same signature as Python's builtin :func:`map` | |
328 | function, but runs the calculation in parallel. More involved examples of using |
|
358 | function, but runs the calculation in parallel. More involved examples of using | |
329 |
:class:` |
|
359 | :class:`DirectView` are provided in the examples that follow. | |
330 |
|
||||
331 | .. image:: figs/mec_simple.* |
|
|||
332 |
|
360 |
1 | NO CONTENT: file was removed, binary diff hidden |
|
NO CONTENT: file was removed, binary diff hidden |
1 | NO CONTENT: file was removed, binary diff hidden |
|
NO CONTENT: file was removed, binary diff hidden |
1 | NO CONTENT: file was removed, binary diff hidden |
|
NO CONTENT: file was removed, binary diff hidden |
1 | NO CONTENT: file was removed, binary diff hidden |
|
NO CONTENT: file was removed, binary diff hidden |
1 | NO CONTENT: file was removed, binary diff hidden |
|
NO CONTENT: file was removed, binary diff hidden |
1 | NO CONTENT: file was removed, binary diff hidden |
|
NO CONTENT: file was removed, binary diff hidden |
General Comments 0
You need to be logged in to leave comments.
Login now