multi

1 ############################################################################### 2 # # 3 # Copyright (C) 2007 Gary S Thompson # 4 # Copyright (C) 2008,2012 Edward d'Auvergne # 5 # # 6 # This file is part of the program relax (http://www.nmr-relax.com). # 7 # # 8 # This program is free software: you can redistribute it and/or modify # 9 # it under the terms of the GNU General Public License as published by # 10 # the Free Software Foundation, either version 3 of the License, or # 11 # (at your option) any later version. # 12 # # 13 # This program is distributed in the hope that it will be useful, # 14 # but WITHOUT ANY WARRANTY; without even the implied warranty of # 15 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # 16 # GNU General Public License for more details. # 17 # # 18 # You should have received a copy of the GNU General Public License # 19 # along with this program. If not, see <http://www.gnu.org/licenses/>. # 20 # # 21 ############################################################################### 22 23 # Package docstring. 24 """The multi-processor package. 25 26 1 Introduction 27 ============== 28 29 This package is an abstraction of specific multi-processor implementations or fabrics such as MPI via mpi4py. It is designed to be extended for use on other fabrics such as grid computing via SSH tunnelling, threading, etc. It also has a uni-processor mode as the default fabric. 30 31 32 2 API 33 ===== 34 35 The public API is available via the __init__ module. It consists of a number of functions and classes. Using this basic interface, code can be parallelised and executed via an MPI implementation, or default back to a single CPU when needed. The choice of processor fabric is up to the calling program (via multi.load_multiprocessor). 36 37 38 2.1 Program initialisation 39 -------------------------- 40 41 The function multi.load_multiprocessor() is the interface for how a program can load and set up a specific processor fabric. This function returns the set up processor, which itself provides a run() method which is used to execute your application. 42 43 44 2.2 Access to the processor instance 45 ------------------------------------ 46 47 The multi.Processor_box class is a special singleton object which provides access to the processor object. This is required for a number of actions: 48 49 - Queuing of slave commands and memos via Processor_box().processor.add_to_queue(). 50 - Returning results (as a Results_command) from the slave processor to the master via Processor_box().processor.return_object(). 51 - Determining the number of processes via Processor_box().processor.processor_size(). 52 - Waiting for completion of the queued slave processors via Processor_box().processor.run_queue(). 53 54 55 2.3 Slaves 56 ---------- 57 58 Slave processors are created via the multi.Slave_command class. This is special base class which must be subclassed. The run() function should be overridden, this provides the code to execute on the slave processors. 59 60 61 2.4 Results handling 62 -------------------- 63 64 The multi.Result_command class is a special base class which must be subclassed. The run() function should be overridden, this provides the code for the master to process the results from the slaves. 65 66 In addition, the multi.Memo should also be used. This is a special base class which must be subclassed. This is a data store used by the Results_command to help process the results from the slave on the master processor. 67 68 69 70 3 Parallelisation 71 ================= 72 73 The following are the steps required to parallelise a calculation via the multi-processor package API. It is assumed that the multi.load_multiprocessor() function has been set up at the highest level so that the entire program will be executed by the returned processor's run() method. 74 75 76 3.1 Subclassing command and memo objects 77 ---------------------------------------- 78 79 The first step is that the Slave_command, Result_command, and Memo classes need to be subclassed. The Slave_command.run() method must be provided and is used for running the calculations on the slave processors. The Result_command is used to unpack the results from the slave. It is initialised by the Slave_command itself with the results from the calculation as arguments of __init__(). Its run() method processes the results on the master processor. The Memo object holds data other than the calculation results required by the Result_command.run() method to process the results. 80 81 82 3.2 Initialisation and queuing 83 ------------------------------ 84 85 The second step is to initialise the Slave_command and Memo and add these to the processor queue. But first access to the processor is required. The singleton multi.Processor_box should be imported, and the processor accessed with code such as:: 86 87 # Initialise the Processor box singleton. 88 processor_box = Processor_box() 89 90 The slave command is then initialised and all required data by the slave for the calculation (via its run() method) is stored within the class instance. The memo is also initialised with its data required for the result command for processing on the master of the results from the slave. These are then queued on the processor:: 91 92 # Queue the slave command and memo. 93 processor_box.processor.add_to_queue(command, memo) 94 95 96 3.3 Calculation 97 --------------- 98 99 To execute the calculations, the final part of the calculation code on the master must feature a call to:: 100 101 processor_box.processor.run_queue(). 102 103 104 4 Example 105 ========= 106 107 See the script 'test_implementation.py' for a basic example of a reference, and full, implementation of the multi-processor package. 108 109 110 5 Issues 111 ======== 112 113 For multi-core systems and Linux 2.6, the following might be required to prevent the master processor from taking 100% of one CPU core while waiting for the slaves: 114 115 # echo "1" > /proc/sys/kernel/sched_compat_yield 116 117 This appears to be an OpenMPI problem with late 2.6 Linux kernels. 118 """ 119 120 121 __all__ = ['memo', 122 'misc', 123 'mpi4py_processor', 124 'multi_processor_base', 125 'processor', 126 'processor_io', 127 'result_commands', 128 'result_queue', 129 'slave_commands', 130 'uni_processor'] 131 132 # Python module imports. 133 import sys as _sys 134 import traceback as _traceback 135 136 # Multi-processor module imports. 137 from multi.memo import Memo 138 from multi.misc import import_module as _import_module 139 from multi.misc import Verbosity as _Verbosity; _verbosity = _Verbosity() 140 from multi.result_commands import Result_command 141 from multi.slave_commands import Slave_command 142 143 144 #FIXME error checking for if module required not found. 145 #FIXME module loading code needs to be in a util module. 146 #FIXME: remove parameters that are not required to load the module (processor_size).

147 -def load_multiprocessor(processor_name, callback, processor_size, verbosity=1):

148 """Load a multi processor given its name. 149 150 Dynamically load a multi processor, the current algorithm is to search in module multi for a 151 module called <processor_name>.<Processor_name> (note capitalisation). 152 153 154 @todo: This algorithm needs to be improved to allow users to load processors without altering the relax source code. 155 156 @todo: Remove non-essential parameters. 157 158 @param processor_name: Name of the processor module/class to load. 159 @type processor_name: str 160 @keyword verbosity: The verbosity level at initialisation. This can be changed during program execution. A value of 0 suppresses all output. A value of 1 causes the basic multi-processor information to be printed. A value of 2 will switch on a number of debugging printouts. Values greater than 2 currently do nothing, though this might change in the future. 161 @type verbosity: int 162 @return: A loaded processor object or None to indicate failure. 163 @rtype: multi.processor.Processor instance 164 """ 165 166 # Check that the processor type is supported. 167 if processor_name not in ['uni', 'mpi4py']: 168 _sys.stderr.write("The processor type '%s' is not supported.\n" % processor_name) 169 _sys.exit() 170 171 # Store the verbosity level. 172 _verbosity.set(verbosity) 173 174 # The Processor details. 175 processor_name = processor_name + '_processor' 176 class_name = processor_name[0].upper() + processor_name[1:] 177 module_path = '.'.join(('multi', processor_name)) 178 179 # Load the module containing the specific processor. 180 modules = _import_module(module_path) 181 182 # Access the class from within the module. 183 if hasattr(modules[-1], class_name): 184 clazz = getattr(modules[-1], class_name) 185 else: 186 raise Exception("can't load class %s from module %s" % (class_name, module_path)) 187 188 # Instantiate the Processor. 189 object = clazz(callback=callback, processor_size=processor_size) 190 191 # Load the Processor_box container and store the details and Processor instance. 192 processor_box = Processor_box() 193 processor_box.processor = object 194 processor_box.processor_name = processor_name 195 processor_box.class_name = class_name 196 197 # Return the Processor instance. 198 return object

199 200

201 -def fetch_data(name=None):

202 """API function for obtaining data from the Processor instance's data store. 203 204 This is for fetching data from the data store of the Processor instance. If run on the master, then the master's data store will be accessed. If run on the slave, then the slave's data store will be accessed. 205 206 207 @attention: No inter-processor communications are performed. 208 209 @keyword name: The name of the data structure to fetch. 210 @type name: str 211 @return: The value of the associated data structure. 212 @rtype: anything 213 """ 214 215 # Load the Processor_box. 216 processor_box = Processor_box() 217 218 # Forward the call to the processor instance. 219 return processor_box.processor.fetch_data(name=name)

220 221

222 -def fetch_data_store():

223 """API function for obtaining the data store object from the Processor instance. 224 225 If run on the master, then the master's data store will be returned. If run on the slave, then the slave's data store will be returned. 226 227 228 @attention: No inter-processor communications are performed. 229 230 @return: The data store of the processor (of the same rank as the calling code). 231 @rtype: class instance 232 """ 233 234 # Load the Processor_box. 235 processor_box = Processor_box() 236 237 # Return the data store. 238 return processor_box.processor.data_store

239 240

241 -def send_data_to_slaves(name=None, value=None):

242 """API function for sending data from the master to all slaves processors. 243 244 @attention: Inter-processor communications are performed. 245 246 @keyword name: The name of the data structure to store. 247 @type name: str 248 @keyword value: The data structure. 249 @type value: anything 250 """ 251 252 # Load the Processor_box. 253 processor_box = Processor_box() 254 255 # Forward the call to the processor instance. 256 processor_box.processor.send_data_to_slaves(name=name, value=value)

257 258 259

260 -class Application_callback(object):

261 """Call backs provided to the host application by the multi processor framework. 262 263 This class allows for independence from the host class/application. 264 265 @note: B{The logic behind the design} the callbacks are defined as two attributes 266 self.init_master and self.handle_exception as handle_exception can be null (which is 267 used to request the use of the processors default error handling code). Note, however, 268 that a class with the equivalent methods would also works as python effectively handles 269 methods as attributes of a class. The signatures for the callback methods are documented 270 by the default methods default_init_master & default_handle_exception. 271 """ 272

273 - def __init__(self, master):

274 """Initialise the callback interface. 275 276 @param master: The data for the host application. In the default implementation this is an 277 object we call methods on but it could be anything... 278 @type master: object 279 """ 280 281 self.master = master 282 """The host application.""" 283 284 self.init_master = self.default_init_master 285 self.handle_exception = self.default_handle_exception

286 287

288 - def default_handle_exception(self, processor, exception):

289 """Handle an exception raised in the processor framework. 290 291 The function is responsible for aborting the processor by calling processor.abort() as its 292 final act. 293 294 @param processor: The processor instance. 295 @type processor: multi.processor.Processor instance 296 @param exception: The exception raised by the processor or slave processor. In the case of 297 a slave processor exception this may well be a wrapped exception of type 298 multi.processor.Capturing_exception which was raised at the point the 299 exception was received on the master processor but contains an enclosed 300 exception from a slave. 301 @type exception: Exception instance 302 """ 303 304 # Print the traceback. 305 _traceback.print_exc(file=_sys.stderr) 306 307 # Stop the processor. 308 processor.abort()

309 310

311 - def default_init_master(self, processor):

312 """Start the main loop of the host application. 313 314 @param processor: The processor instance. 315 @type processor: multi.processor.Processor instance 316 """ 317 318 self.master.run()

319 320 321

322 -class Processor_box(object):

323 """A storage class for the Processor instance and its attributes. 324 325 This singleton contains Processor instances and information about these Processors. Importantly 326 this container gives the calling code access to the Processor. 327 """ 328 329 # Class variable for storing the class instance. 330 instance = None 331

332 - def __new__(self, *args, **kargs):

333 """Replacement function for implementing the singleton design pattern.""" 334 335 # First initialisation. 336 if self.instance is None: 337 self.instance = object.__new__(self, *args, **kargs) 338 339 # Already initialised, so return the instance. 340 return self.instance

341

Source Code for Package multi