1 SimGrid (3.4.1-svn) unstable; urgency=low
4 * Fix a bug preventing the tasks from begin garbage collected.
7 * Fix a bug occuring when a host involved in a communication fails.
8 This was not detected properly by the other peer involved in the
9 communication. Now, it's reported as a network error.
12 * Warn the user about loop dependencies in data flow of DAX files
13 * Obey the control-flow dependencies of DAX files
16 * Add option "enable_smpi" for fix mac problem with smpi
17 * Add verification of lua5.1 and ruby1.8
19 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
21 SimGrid (3.4) stable; urgency=low
23 The "Easter in Cargese" release. Also known as (major changes):
25 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
26 ~> bindings were greatly improved
27 ~> new tracing infrastructure for better visualization introduced
29 * the "Welcome to configury modernity" release.
30 ~> we switched from autotools to cmake, and improved our cdash
33 A more detailled list of changes follow (full detail in svn log).
35 Java Bindings: Various Cleanups
36 * (install java-gcj-compat-dev on debian-like to use them)
37 * Remove put/get: no need to export deprecated interface in Java
38 Use send/receive instead.
39 * Cleanup the examples and add a README per directory
40 * Remove example autoDestination (that's the only way to go now)
41 * Remove example explicitDestination (was a plain copy of basic)
42 * Make JniException a runtime exception, so that there is no need to
43 declare the fact that you may encounter such a beast. I guess that
44 nobody will ever want to survive such error.
45 * Create specific errors for each MSG case of failure:
46 host failure, transfer failure, timeout, task cancelled
47 * Cleanup the exceptions that may get thrown by each function
48 * Other internal cleanups in Java bindings. Performance still bad :/
49 Ruby and Lua Bindings: create them
50 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
51 * That's new and great, you should try them out.
52 Same functionalities than Java bindings, only even less polished
54 * Kill the useless "rate" argument of SD_task_get_execution_time()
55 Everyone used to provide -1 as a value, it was not used, and the
56 semantic of a possible use wasn't even clear.
57 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
58 either as comm costs or compute costs to mean that there is no
59 such thing for that specific task.
60 * Add a SD_task_set_name() function
61 * Fix SD_task_unschedule() on typed tasks
62 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
63 * In DAX loader, accept useless 'level' attributes to <job> since
64 LIGO DAGs have them (seem to be to ease graphical representation).
66 * Add an example masterslave_mailbox.c using send/receive and not
67 the deprecated put/get interface.
68 * Kill the MSG_paje_output() function. It's a noop since 2 years.
69 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
71 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
72 (declare MSG_USE_DEPRECATED to still have the old name)
73 * Add a MSG_task_set_data() function
74 * About trace replay (see examples/msg/actions):
76 - Allow to work with splitted trace files for each process
77 Give the specific trace file as argument of each process,
78 and call MSG_action_trace_run(NULL)
79 You can still have one merged file for all processes.
80 - Fix implementation of collective operations
81 * Allow task_execute() on 0-sized tasks (closes #10063)
83 * This is the first release of SimGrid where SMPI is not considered
84 beta anymore (even if some corners should still be improved)
85 * Port over the new SIMIX_network submodule (internal refactoring)
86 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
87 * Implement more missing elements of the standard:
89 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
91 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
92 MPI_Scan() MPI_Get_processor_name()
93 - Added implementation of missing case for Alltoall (warning: it's
94 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
95 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
96 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
99 - MPI_Waitsome() was broken
100 - Allow relative includes in smpicc
101 - Command line cfg argument 'reference_speed' was ignored...
102 - Some functions did not properly lead to auto-benching of user code
103 - smpicc passes -O2 by default (just like openmpi one)
105 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
106 * Bug fixes about timeouts during communications
107 * add SIMIX_message_sizes_output() as a pimple to write to file the
108 amount of messages per size. Use gnuplot to get histogram.
109 Pimple because that's the only user-visible function of simix,
110 defined directly in xbt.h (irk, sorry)
112 - Add a SIMIX_sem_get_capacity() function
113 - Fix interactions with processe resume/suspende
114 - release_forever() was stupidly broken
115 - Fix SIMIX_display_process_status() for processes in a semaphore
116 - Make SIMIX_sem_block_onto() user-visible
117 * Refactoring context stuff:
118 - Use pseudo-OOP for better modularity
119 - reimplement SIMIX_process_kill() without process_schedule() so
120 that the latter can take as invariant that it is called from
122 - Merge context_start into context_new for sake of simplicity
124 * Add a Vivaldi network model, coded live during SUD'10 ;)
125 * Rename configuration variables to start a hierarchy:
126 o cpu_model -> cpu/model
127 o network_model -> network/model
128 o workstation_model -> workstation/model
129 * New configuration variables:
130 o network/bandwidth_factor: correction to bandwith
131 o network/latency_factor: correction to latency
132 o netwotk/weight_S: correction to the weight of competing streams
133 * Add a long description to the models, that users can see with such
134 argument on the command line: --cfg=cpu/model:help
135 * --help-models display the long description of all known models
137 * config: add the ability to set a default value after registration
138 Does not override any previously set value (e.g. from cmd line)
139 * dict: allow to have integer key and data.
140 When so, you need to use the following functions
141 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
142 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
143 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
144 In contrary to regular dicts, the key is not malloced before copy.
145 Mixing scalar and regular elements in the same dict is not tested
147 * Allow to use xbt_dynar_shrink() to expend the dynar instead
148 Tracing for Visualization:
149 * SimGrid is now instrumented in order to generate a trace file for
150 visualization analysis: to use it, need to compile SimGrid with the
151 "tracing" option enabled, and instrument the program using SimGrid with
152 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
153 (among other functions).
154 * The instrumentation only traces the platform utilization for now
155 * Documentation to use the tracing functions and how to analyze the
156 traces with the Triva tool is written.
157 * More information about: SimGrid FAQ (in the section Tracing Simulations
160 * We moved to cmake as default build system. Autotools support will
161 be dropped soon. Check the FAQ for more info about how to use it.
162 * Greatly improved our cdash/ctest interactions
163 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
164 * Added memory checking tests with valgrind; lot of memleak fixing.
165 This may be the first release of simgrid with so few memory issues
166 * Added code coverage tests.
167 Our coverage is still improvable, but at least we see it on cdash.
169 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
171 SimGrid (3.3.4) stable; urgency=low
173 The "Desktop Grid needs love too" release (also called Xmas release).
176 * Major speedup in the maxmin system solving by using lazy evaluation
177 Instead of solving completely the maxmin system at each iteration,
178 only invalidate (and recompute) the modified parts.
179 This new feature is enabled in default models but you can try to
180 turn it on with "--cfg:maxmin-selective-update=1" for other models.
181 * Cas01 IMproved as default CPU model
182 This CPU model is the same Cas01 model, but it uses the
183 maxmin-selective-update flag and a heap structure to manage
184 actions on SURF kernel.
185 It reduces the complexity to find the next action to finish and,
186 consequently, it's faster than the old Cas01.
187 This is the new default CPU model (Cas01).
188 * Rename the old Cas01 model to Cas01_fullupdate
189 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
190 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
191 * CpuTI (CPU Trace Integration)
192 A new CPU model whose objective is simulate faster when using
193 availability trace files.
194 Instead of using a full featured, over engineered maxmin system for
195 CPU modeling, this model does the pre-integration of traces files
196 to calculate the amount of CPU power available, and so, executes
197 faster than the old CPU models.
198 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
199 * Use LV08 as default network model since it gives better accuracy
200 for small messages and shouldn't change things for big ones.
201 Use --cfg=network_model:CM02 to get the previous behavior.
204 ******************************************
205 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
206 ******************************************
207 * The new CPU model may changes simulations!
208 The point is that events occurring at the exact same timestamp
209 are not scheduled in the same order with the old and new
210 version. This may be enough to completely change the execution
211 of simulations in some cases.
212 * The new network model will change simulations!
213 This new model is more realistic than the previous one, so you
214 should consider redoing your old experiments with this model.
215 Sorry for the inconvenience.
218 * Introduce the supernovae compilation mode
219 When compiled that way, the whole SimGrid (or almost) is put in a
220 single compilation unit and compiled in one shoot.
221 This is to help gcc which has difficulties to inline stuff from one
223 The speedup seem to be above 15%, althrough more tests are needed on
224 amd64 to confirm that gain.
227 * Port of MSG's mailbox on top of SIMIX network
228 The put/get mechanism was greatly simplified on the way.
231 * New SIMIX network module. Provides:
232 - Mailbox: rendez-vous mecanism to find with who you want to speak
233 - Synchronous send/recv: easier and hopefully faster since the
234 logic is handled in the maestro process directly now
235 - Asynchronous send/recv: you dreamt of it? It's here now
236 Too bad that nobody cared enough to propagate the change to MSG.
237 * Add semaphores as SIMIX synchronization mechanism.
240 * new function SD_daxload(char*) to load a DAX file
241 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
242 * Introduce typed tasks. Specify its kind and cost at creation.
243 At scheduling, just give where it should be placed, and the cost
244 for each involved resource is automatically computed.
245 Existing constructors so far (more to come of course):
246 - SD_task_create_comm_e2e() for end-to-end communication
247 - SD_task_create_comp_seq() for sequential computation
248 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
249 * new function SD_task_dump() for debuging display
250 * new function SD_task_dotty(task,FILE*) writing to file the info
251 about the task in dotty format
252 * SD_task_dependency_exists() can now cope with having one of its
253 arguments NULL. If so, it tests whether the other argument has any
255 * Add getters on list of preceding/following tasks:
256 SD_task_get_parents(task) and SD_task_get_children(task)
257 * Add getters on amount of workstations and list:
258 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
259 * Add getter on task kind: SD_task_get_kind(task)
260 * Update the start_time and finish_time of tasks on completion/failure
261 * Bugfix: Remove task from state swags when destroyed
264 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
267 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
268 dynar with one separate thread per value of the dynar.
269 * Change the prototype of xbt_thread_create(), sorry.
270 Added a boolean parameter indicating whether we want to join this
271 thread (used in SG only for now)
272 * Implement xbt_thread_join and xbt_thread_yield in SG also.
275 * GTNetS wrappers should now be usable again (and betterly tested too)
276 * Fix a major regression from 3.2 where the timeout provided to
277 MSG_task_put_with_timeout() was used as absolute time before which
278 the comm should be done.
279 * Start to fix the <cluster> tag.
280 - Internal links should be good now (beside of the loopback, which
281 use the private link instead)
282 - paths to the external world is still rather broken
283 - the <route:multi> tag is just broken. Actually that's brain-dead.
284 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
285 to make it less stupid
286 ** Check your platform with teshsuite/simdag/platforms/flatifier **
287 * Fix a source-level compatibility glitch from 3.2: after defining
288 MSG_USE_DEPRECATED, you can use the old name
289 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
290 * Allow to compile from the SVN with automake 1.11
291 * Fix some problems when using the "start_time" tag in deployment XMLs.
292 * Fix #8569: XBT/synchro.h has redundant declarations
293 * Fix #8563: MSG return values and exceptions
294 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
295 * Integrate patch #8636: Obey DESTDIR when installing documentation.
296 Thanks to Robson Peixoto.
297 * Fix a vicious bug in dictionaries inducing that some elements were
298 not freed on xbt_dict_free()
300 Portability report of this version:
301 * Main portability targets:
302 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
304 - mac leopard on i386
305 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
306 but nothing critical.
307 * Other platforms: windows, AIX and others were not tested for this release
309 Timing report of this version:
310 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
311 scenario parameters). From 8h to a few seconds in desktop grid settings.
312 * Supernovae brings about 25% speedup on i386.
314 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
316 SimGrid (3.3.3) stable; urgency=low
318 The "Need for Speed" release.
320 The timings done to validate the 3.3.2 were faulty.
321 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
323 The problem was a conversion from a manually handled vector to
324 xbt_dynar_t on the critical path.
325 xbt_dynar_foreach calls functions, inducing stack management crap.
327 We inlined these functions and xbt_dynar_foreach is now breath taking.
328 We also inlined xbt_swag_belong on the way.
330 Here are some approximate speedup measurements (on master/slaves
331 simulations lasting between 10s and 20s each):
332 3.3.1 -> 3.3.2: about same performance
333 3.3.2 -> 3.3.3: 40% speedup
334 3.3.1 -> 3.3.3: 40% speedup
335 3.3.1 with inline patch -> 3.3.3: 30% speedup
337 Our reading is that the refactoring which occurred in 3.3.2 made us
338 suffer much more from the xbt_dynar_foreach low performance, but
339 once we solved this, this refactoring proved to be very performance
340 effective. From the 40% speedup, somehow, 10% are due to the
341 inlining and 30% to the refactoring.
343 That's a pitty that gcc cannot inline functions placed in other files
344 alone. We have to choose between:
345 - break the encapsulation (by putting private data structures and
346 accessors in headers files to help gcc)
347 - live with low performance
348 - switch to a decent compiler such as icc (not quite possible).
350 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
352 SimGrid (3.3.2) stable; urgency=low
354 The "Simplicity does not preceed complexity, but follows it" release.
356 The main contributors of this release were (lexical order):
357 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
360 * Extract the routing logic into its own object.
361 (was dupplicated in network.c and workstation_LV07.c;
362 Allows to implement other ways of storing that info)
363 => kill now useless network_card concept
364 - Use dynar to represent routes (instead of void** + int*)
365 - kill link_set (use surf_network_model->resource_set instead)
366 - Add a command-line option to choose the routing schema to use
367 - Add three new models:
368 * Floyd (shortest path computed at initialization)
369 * Dijikstra (shortest path recomputed all the time)
370 * Cached Dijikstra (shortest path computed on need)
371 All these models where contributed by Silas De Munck, and are
372 described in his ICCS09 paper.
374 * Simplify model declaration
375 (less redirections, less function to write when defining a model)
376 - Factorize stuff between models:
379 surf_model_resource_set(model)
380 surf_model_resource_by_name(model, name)
381 - Unify the types of models in s_surf_model_t (using an union)
382 - Embeed fields of common_public directly into s_surf_model_t
383 - Rename model methods:
384 action_free ~> action_unref
385 action_change_state ~> action_state_set
386 action_get_state ~> action_state_get
387 - Change model methods into functions :
388 (model)->common_public->action_use ~> surf_action_ref
390 * Implement a generic resource; use it as ancestor to specific ones
391 (allows to kill duplicated code in models)
392 Drawback: timer command don't need no name nor properties;
393 workstation_CLM03 don't need no properties
394 (but I guess we can live with those few bytes wasted)
396 * Improve the action object model
397 - implement a constructor avoiding dupplicated code about field
398 initialization in generic_action part.
400 * Kill the SDP model: it has an external dependency, is deprecated
401 in flavor of modern lmm models, and didn't compile since a while
404 * Relocation of the context module from XBT to SIMIX.
405 (the context were decoupled from the simix processes, duplicating a lot of code)
406 => a lot of code was factorized
407 - less overhead is introduced during scheduling
408 - simpler API for the context factory
409 - the logic for process creation,destruction and manipulation was simplified
410 * Simplification of the s_smx_process_t data structure.
411 => accesing the simix level data associated to a process is faster now,
412 and the code is a lot more readable.
415 * Implement some more MPI primitives:
416 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
417 -implementation: Bcast: flat or 2-ary tree (default),
420 Allreduce: Reduce then Bcast
421 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
422 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
423 Alltoallv: flat tree, like ompi
425 * Add support for optimized collectives (Bcast is now binomial by default)
426 * Port smpirun and smpicc to OS X
429 * Kill SD_link_get_properties: hard to maintain and makes very little sense
430 Shout out if you used it.
433 * Display the list of still queued messages in SG mode when existing
437 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
438 * Add xbt_graph_node_get_outedges() [Silas De Munck]
439 * Add xbt_str_from_file(FILE*)
440 * Add xbt_dict_get_key achieving a linear reverse search
441 * Remove the context module
443 Portability report of this version:
444 * Main portability targets:
445 - Linux(debian)/x86/context
446 - Linux(debian)/x86/pthreads
447 - Linux(debian)/amd64/context
448 - Linux(debian)/amd64/pthreads
449 On these, we still have the eratic breakages of gras/pmm and
450 amok/saturate_sg reported in previous version. We still think
451 that the tests are the cause of the fault, not the tested code.
453 - Mac OSX Leopard/x86/context
454 Still false negative in tesh autotesting.
455 Smpi still fails, but this time because readlink does not accept -f
456 Everything seems to work properly beside of that.
459 - AIX version 5.3 (only tested contexts this time)
460 Smpi still fails there because mktemp is not installed.
461 Everything seems to work properly beside of that.
463 I managed to compile it for the first time, but several breakages.
464 Won't delay the release for this exotic platform.
466 * Windows: it's still lagging behind. If you want to help, please
469 Timing report of this version:
470 This version seem to be more than 5% faster than 3.3.1 (on linux
471 64bits with contextes). The gain is less than expected, we are
472 investigating this for next release.
474 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
476 SimGrid (3.3.1) stable; urgency=low
479 * Implement a --cfg-help to show existing configuration variables
480 * Build chain do not require doxygen in maintainer mode
483 * fix a bug on struct sizeof computation, which prevented the
484 exchange of arrays of structs in some conditions
485 - added a regression test about this in datadesc_usage
486 * Allow the exchange of 0-long dynamic vectors.
487 - for that, use -1 as indicator of dynamic size instead of 0
488 - This implied to change any size from unsigned long to long,
489 reducing a bit communication abilities, but I guess that with
490 64bits being quite common, this is more than enough.
491 - This also induce a protocol change, thus bumping network protocol
492 version from 0 to 1 (if we have external users, we have to get
493 clean on that point too ;)
494 - added two regression tests about this in datadesc_usage
495 * Be more verbose when propagating local exceptions
496 This helps debugging.
497 * Display the status of simulated processes when receiving SIGINT in
501 * Allow to control the simulation from a trace file.
502 New functions MSG_action_register() and MSG_action_trace_run()
503 The first one allows to associate a function execution to each
504 kind of action while the second one parses a trace file and
505 triggers the corresponding actions within the system.
506 For now, only a toy example is provided in examples/msg/actions
507 * Add an exemple of process migration in examples/msg/migration
508 * Fix a bug in task exchange which broke MSG_task_get_sender()
509 Add a teshsuite regression test for that.
510 [Bug: if MSG_task_get_sender() is called after sender exit,
512 * Fix a bug which prevented suspend/resume to work properly
513 * Display the status of simulated processes when receiving SIGINT
514 This fixes a regression of v3.3. due to the introduction of SIMIX
515 * Bug fixing in failure management:
516 - trace could not start by a failure at time 0
517 - failure during communications were not working
520 * Add SIMIX_process_set_name() to change the name of the current
521 process in the log messages.
522 * Store smx_hosts in a dict since we only retrieve them by name
523 * Move the configuration infrastructure to surf
526 * Move the configuration infrastructure to surf
529 * Massive internal cleanups:
530 - Store internal structures on processes instead of hosts (allows
531 to have more than one process per host, in addition of being more
533 - Cleanup the initialization/finalization process
534 - Kill a whole bunch of unneeded synchronization:
535 processes run in exclusive manner within the simulator
536 - Move queues from global tables to process data fields
538 - now accept -platform and -hostfile arguments
539 - Pass the right rank value to processes according to the hostfile
540 * Compile the examples by default, and use them as regression tests
541 * Implement MPI_Wtime()
542 * Change the reference speed to a command line option
545 * TCP_gamma can now be specified as command line option using
546 --cfg=TCP_gamma:10000000.0
547 * Change the --surf-path cmd line option into --cfg=path:
550 * Also include strbuff from xbt.h public header
551 * xbt_ex_display(): do not free the exception after displaying
552 This allows to do more with the given exception afterward.
553 Users should call xbt_ex_free() themselves.
557 Portability report of this version:
558 * Main portability targets:
559 - Linux(debian)/x86/context
560 - Linux(debian)/x86/pthreads
561 - Linux(debian)/amd64/context
562 - Linux(debian)/amd64/pthreads
563 These targets fail about 1/10 of times on gras/pmm, but we believe
564 that this is because of the test, not because of simgrid.
565 amok/saturate_sg fails even more rarely, and the test may not be
568 - Mac OSX Leopard/x86/context
569 The test suite still spits tons of errors because some obscure
570 force prevents us from removing the temporary directories
571 arguing that they still contain some metadata I've never heard of.
572 Smpi fails because seq is not installed.
573 Everything seems to work properly beside of that.
576 - AIX version 5.3 (both contexts and pthread)
577 Smpi still fails there because mktemp is not installed.
578 XML inclusions seems rosty on AIX.
580 * Windows: it's still lagging behind. If you want to help, please
583 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
585 SimGrid (3.3) stable; urgency=high
589 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
592 * Introduce the SIMIX module: factorize code between MSG and GRAS.
595 Until now, GRAS were using MSG as an interface to SURF. It was
596 quite difficult because both interface have several differences
597 (MSG channels vs GRAS sockets were the most notable point).
599 This also opens the gate to SMPI (which should occur soon) and speed
600 up simulations by to 40% (even if it were not the main goal).
602 **************************************
603 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
604 **************************************
605 The point is that events occuring at the exact same timestamp are
606 not scheduled in the same order with the old and new version. This
607 may be enough to completely change the execution of simulations in
608 some cases. Sorry for the inconvenience.
610 * Cleanup and upgrade the XML format to push further scalability
611 issues (check http://hal.inria.fr/inria-00256883/ for more info)
613 * Improve the testing infrastructure with tesh. Now a very large part of
614 the code is tested not only by being run but also by checking that the
615 output match an expected output [Mt].
617 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
618 is really less memory-demanding, which should allow you to use
619 larger files in SimGrid [AL].
621 * Inform valgrind about our contextes, so that it becomes usable
622 with the default (and more effecient) version of SimGrid
623 [contributed by Sékou Diakite, many thanks]
626 * Introduce a listener thread in charge of receiving incomming
627 messages from the network. It allows to overlap communication and
628 computation but most notably, it removes some stupid deadlocks due
629 to the fact that so far, a process could not send and receive at
630 the same time. This made most non trivial communication schema
632 * Convert the PIDs from long int to int to match the MSG ones (and
634 * New function: gras_agent_spawn() to launch a new process on
635 current host. Only working in simulation for now. [Mt]
636 * New function: gras_os_hostport() returning a constant form (ie,
637 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
640 * Make the backtrace of exceptions more human readable [Mt]
641 * New module: xbt/str [Mt]
642 a ton of string utility functions (split, join, printf to a newly
643 allocated buffer, trim, etc)
644 * New module: xbt/hash [Mt]
645 SHA1 hashing algorithm (more to come if needed)
646 * New module: xbt/synchro [Mt]
647 synchronization tools (mutex and conditions) working the same way
648 in simulation and in real life (mainly useful for GRAS, but not
650 * New module: xbt/queue [Mt]
651 classical producer/consumer synchronization scheme
652 * xbt_dynar_new_sync() creates a synchronized dynar. All access
653 (using the classical functions will get serialized) [Mt]
654 * Make dictionary internal table dynamic. No need to specify its size
655 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
657 * Make sure the log channels are organized as a tree under windows
658 (because of ANSI C compatibility issue, any channel were child of
662 * Cleaned many thing in surf and fixed a few bugs [AL].
663 * Add a nice command line configuration mechanism to compose models [AL].
664 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
665 the previous one (KCCFLN05). It relies on something that looks like
666 a max-min sharing mechanism but cannot be written as such. A new solver
667 was thus designed [AL].
668 * Add a new solver to lmm. Based on Lagrange optimization and
669 gradient-based descent, it enables to efficiently maximise systems s.a
671 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
673 This solver enables to propose two new network models for TCP Reno and
674 TCP Vegas based on Low's work. These models still need to be fully
675 tested though [Pedro Velho].
678 * Bug fix in SD_simulate. Now the time bound given as argument is
680 * Use the new parallel task model (ptask_L07) as default.
681 * Use the SURF command line configuration mechanism.
682 * 0-size tasks (for synchronization) should now work.
684 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
686 SimGrid (3.2) stable; urgency=high
690 We still experience issues on this platform, but we believe that at
693 GRAS API BREAKAGE (for simplification purpose, sorry):
694 * the gras_msgtype_by_name is not used anymore. Instead of
695 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
696 you can write (and must)
697 gras_msg_send(toserver, "request", &request);
698 - If you still want to pass a gras_msgtype_t to the function (to cache
699 the type and avoid the lookup time), use the gras_msg_send_() variant.
700 - Impacted functions:
701 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
702 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
703 * The callbacks are now expected to return 0 when everything went well
704 (just like the main() function)
706 GRAS new features and improvements:
707 * New module mecanism where user code can use per process globals [Mt]
708 This is similar to gras_userdata_*() functions, but for libraries. It
709 factorize some code developped over and over in the examples and AMOK.
710 It has still to be documented and used (only amok/peermanagement is
712 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
713 loss when they were small enough to fit into the buffer and sent quickly
714 enough so that they can all get received in one shoot.
715 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
716 if not found. Use the *_or_null() variant for the old semantic.
717 * In gras_msg_handle, do not discard messages without callback.
718 They are probably messages to be explicitly awaited later (ie, proofs of
719 mis-synchronization in userland since they are sent before being awaited)
721 * gras_socket_meas_send/recv: semantic changed!
722 The numerical arguments used to be (1) the total amount of data to send
723 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
724 messages. This was need for the fool willing to send more than MAXINT
725 bytes on quite fat pipes.
728 * Do really rename the hostmanagement module to peermanagement. [Mt]
729 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
730 but this is rather new and this was documented in the module
731 documentation (poor excuses, I admit)
732 * Bandwidth measurement semantic changed! This follows the changes to
733 gras_socket_meas_send/recv explained above.
736 * A sequential mode has been added to the workstations. When a workstation
737 is in sequential mode, it can execute only one task, and the other tasks
738 are waiting in a FIFO. [Christophe Thiery]
741 * The KCCFLN05 workstation model now handles parallel tasks. It is the
742 model for SIMDAG. [Christophe Thiery]
743 * Bug fix in the maxmin solver: Some values were close to 0 instead of
744 equal to 0, which caused some bad behaviors in
745 saturated_constraint_set_update. I now use a threshold mechanism like in
749 * When running manually src/testall, you select specific units [Mt]
750 testall is the result of our cunit mecanism, and should replace all
751 the scripty thingy around since bash don't run easily on billware.
753 * A mallocator system has been added. [Christophe Thiery]
754 Mallocators allow you to recycle your unused objects instead of freeing them
755 and allocating new ones.
757 Documentation update:
758 * FAQ reworking + New FAQs:
759 - "Valgrind spits tons of errors!" [Mt]
760 - "How to repport bugs" [Mt]
761 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
762 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
763 - Communication time measurement within MSG [AL]
764 - I experience weird communication times when I change the latency [AL]
767 - an introduction to the framework and to the used communication model
768 - an initiatic tour introducing the most proheminent features:
770 . Lesson 0: Installing GRAS
771 . Lesson 1: Setting up your own project
772 o Part 2: Message passing
773 . Lesson 2: Exchanging simple messages
774 . Lesson 3: Passing arguments to the processes (in SG)
775 . Lesson 4: Attaching callbacks to messages
776 . Lesson 5: Using globals in processes
777 . Lesson 6: Logging informations properly
778 . Lesson 7: Using internal timers
779 . Lesson 8: Handling errors through exceptions
780 . Lesson 9: Exchanging simple data
781 . Lesson 10: Remote Procedure Calling (RPC)
782 . Lesson 11: Explicitely waiting for messages
783 . Recapping of message passing features in GRAS
784 - A HOWTO section containing:
785 o HOWTO design a GRAS application
786 More are due, of course. They will come latter. In the meanwhile, you can
787 check the examples which are still here.
789 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
791 SimGrid (3.1) stable; urgency=high
795 There was a stack corruption somewhere, visible only when optimizing
796 with these versions. [Vince]
799 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
800 module that enabled to play with Directed Acyclic Graphs. It is built
801 directly on top of SURF and provides an API rather close to the old
802 SG. Some old codes using SG are currently under rewrite to check that
803 all needful functions are provided. [Christophe Thiery]
806 * Complete rewrite of the KCCFLN05 workstation model. It is now an
807 extension of the classical CLM03 model that gracefully handles
808 failures. This is now the default model for MSG and GRAS. It doesn't
809 handle parallel tasks yet though. [AL]
810 * Bug fix: Weights were not correctly set in the network part.
811 WARNING: This may have resulted in incorrect results with simulations
812 where there are more than one flow on a given link. [AL]
815 * After a (long ?) discussion on simgrid-devel, we have decided that the
816 convention we had on units was stupid. That is why it has been decided
817 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
818 WARNING : This means that all previous platform files will not work as
819 such with this version! A warning is issued to ask users to update
821 A conversion script can be found in the contrib module of the CVS, under
822 the name contrib/platform_generation/surfxml_update.pl [MQ]
825 * Bug fix: Processes were started in reverse order, wrt deployment file.
826 WARNING: if your code relies on this bug, please fix it. [AL]
827 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
828 being executed on two different locations. [AL]
829 * Bug fix: Failures are now better supported thanks to Derrick's tests
830 (there was many failure situations I hadn't thought of and that weren't
831 correctly handled). [AL]
832 * New function: MSG_host_is_avail indicates you whether a given m_host_t
836 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
837 Exception occurring on server-side are propagated back to client (!).
839 API CHANGE: the callback changed their prototype. Change:
840 int my_handler(gras_socket_t expeditor, void *payload_data) {
842 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
843 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
845 * New! function: gras_msg_handleall to deal with all messages arriving
846 within a given period.
847 * New! function: gras_socket_server_range to get a server socket in a
848 range of port numbers (ease to avoid port number conflicts) [MQ]
849 * New! gras processes display their backtrace when they get a SIGUSR1
850 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
851 Sweet to debug RL processes [MQ]
855 - Do not force experiment sizes to be expressed in kb, or it becomes
856 impossible to measure the latency this way (needs one byte-long tests)
857 WARNING: this changes the amok_bw_* function semantic. [MQ]
858 - Implements the link saturation stuff. [MQ]
859 * Peer management module:
860 New! module factorizing code that we wrote over and over [MQ].
863 * New module: cunit (my jUnit implementation in ansi C) [MQ]
864 - Test units are placed directly into the library code, they get extracted
865 automatically and placed into the src/testall binary.
866 - Convert most of the XBT tests to this system.
867 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
868 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
869 live in the same C code. This required to change a little bit the API
870 of surfxml parsing but shouldn't be an issue for end-users. [AL]
871 * New module: sparse graph structure with basic algorithms (this is work
872 in progress and the API is not considered to be frozen yet). [AL]
873 * Display more information on backtraces: source line & function names are
874 now displayed just like valgrind does (rely on addr2line tool) [MQ]
875 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
876 * Reworked a little bit some #include statements to load only required
877 headers. Some user code that relied on SimGrid to include stdlib or
878 stdio may need to include it by themselves. [AL]
879 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
881 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
882 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
883 but I guess that almost nobody used those parts. [MQ]
885 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
887 SimGrid (3.0.1) stable; urgency=low
890 * Unfortunately, I had missed 5 misnamed functions:
891 xbt_fifo_item_t xbt_fifo_newitem(void);
892 void xbt_fifo_freeitem(xbt_fifo_item_t);
893 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
894 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
895 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
896 They're now deprecated. Please use their new versions:
897 xbt_fifo_item_t xbt_fifo_new_item(void);
898 void xbt_fifo_free_item(xbt_fifo_item_t);
899 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
900 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
901 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
903 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
904 * Documentation: xbt_log module unmercifully reworked [MQ]
905 * Bugfix: there was a problem with the ending of contexts with
906 the pthread backend. It caused some weird deadlock or behavior
907 depending on the pthread implementation. [AL]
908 * Bugfix: get the exceptions raised in the simulator repport where
909 and why they come from when they are not catched in time [AL, MQ]
912 * Bugfix: Do repport the error when two non-connected hosts try to
913 exchange data (Thanks to Flavien for stumbling into this one) [AL]
916 * Add additionnal checkings on communications. Assert that two
917 communicating hosts are connected by a set of links... [AL]
920 * Add additionnal checkings on channel values in communication [AL]
921 * New: MSG_task_get_source to see on which host a task was generated [HC]
922 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
923 the number of tasks waiting to be received on channel and sent
925 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
926 waits for the first task coming from a given host.. [AL]
928 GRAS new functionnalities: [MQ]
929 * Enhance the parsing macro to allow the size of multidimentional objects
930 to be given thru annotations.
931 * New example (and documentation): Matrix Multiplication a la RPC
932 (as when I was young!) and fix a bunch of bugs found on the way.
934 GRAS performance improvements: [MQ]
936 * Reduce the amount of cbps creation/destruction by making it static to
937 datadesc_send/recv() and using a (newly created) cbps_reset (based on
940 * Change libdata to a set so that we can search for stuff by ID (and thus
941 reduce the insane amount of dict lookups)
944 * Actually implement gras_datadesc_copy() so that we don't have to mimick
945 RL communication on top of SG since it's so uneffective.
946 It may also be used for inter-thread communication in RL, one day.
947 * Use gras_datadesc_copy() to exchange messages on top of SG
949 - improve message exchange performance on top of SG
950 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
951 * Don't exchange on the network the size of the used part of buffer,
952 instead, specify the possible buffer size to read().
954 - reduces the amount of read/write calls (one pair per exchange)
955 - reduces the amount of exchanged data (the size)
956 - allows to retrieve all arrived data on receiver side, if we don't need
957 it right now (subsequent read will peek the buffer)
958 - allows the receiver to proceed with the begining of the stream before
959 everything is arrived
960 - make it possible to build an iov transport (using readv/writev)
962 - take care of the data with non-stable storage (like stacked data),
964 * If possible, TCP send uses vector I/O (when writev() is here)
965 - Don't use it for receive since we send data sizes and data on the
966 same stream, so we wouldn't be able to chain large amount of chunks
967 before having to flush the stuff to read the size.
968 * Rework the transport plugin mecanism to simplify it and reduce the
969 amount of pointer dereferencement when searching for the right function
972 * I guess that now, we do almost as few system calls as possible while
973 doing as few data copy as possible.
975 To improve it further, we could try to send all the sizes first and then
976 all the data (to use iov on receiving size), but it's only a partial
977 solution: when you have 2 dimensional data, the sizes of the second
978 dimension is data of the first dimension, so you need 3 streams.
980 I'm not sure the potential performance gains justify the coding burden.
982 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
984 SimGrid (3.00) stable; urgency=high
987 * New! Give the possibility to hijack the surf parser and thus bypass
988 MSG_create_environment and MSG_launch_application. Have a look at
989 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
991 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
993 SimGrid (2.96) unstable; urgency=low
998 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
999 This deprecates the xbt_error_t mecanisms.
1000 It modifies (simplifies) all XBT and GRAS API.
1001 MSG API keeps unchanged (exceptions raised by XBT are catched from
1002 within MSG and masked with existing error handling facilities)
1005 * New! Add a FATPIPE model. [AL]
1006 * New! Add a parallel task model. [AL]
1007 * New! Add automatically a loopback interface (in the default
1008 network model) if none was precised.
1011 * Bugfix: MSG_process_resume now works with the current running process.
1013 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1014 * Modification of MSG_task_get_compute_duration. Once a task has been
1015 processed, the value returned by this function is now equal to 0. [AL]
1016 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1017 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1018 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1019 of a m_task. For now, MSG_task_cancel only works with computation
1021 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1022 of the processor (in Mflop/s) regardless of the current load on the
1024 * API Change: use proper naming convention for MSG_getClock and
1025 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1027 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1028 This function changes the priority of a computation task. This priority
1029 doesn't affect the transfer rate. A priority of 2 will make a task
1030 receive two times more cpu power than the other ones. This function
1031 has been added to suit the needs of Nguyen The Loc and hasn't been that
1032 much tested yet. So if it fails, please report it and send me your code.
1034 * API Change: removed all functions and types that were marked "deprecated"
1035 since many months. Renamed MSG_global_init_args to MSG_global_init.
1037 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1039 SimGrid (2.95) unstable; urgency=low
1042 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1043 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1044 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1047 * Bugfix: Allow absolute paths for platform description files [MQ]
1048 * Bugfix: do free the variables after use. Leads to drastic performance
1050 * Implement max_duration (ie, timeouts) on resources [AL]
1053 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1055 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1056 the needed support (timeouts) [AL]
1059 * Implement measurement sockets. You can now get the bandwidth between two
1060 hosts thanks to AMOK (see below). [MQ]
1061 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1062 dynar over the network (yeah) [MQ]
1063 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1064 * simple/static token-ring example. [Alexandre Colucci and MQ]
1065 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1066 in place before [MQ]
1069 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1070 type parsing. "short" and "long" modifiers were also ignored; other
1071 modifier (such as reference level) are still ignored. [MQ]
1072 * Embeed the buffer size within the buffer itself on SG. [MQ]
1073 That way, send() are atomic and cannot get intermixed anymore (at least
1074 the ones which are less than 100k; bigger messages still have the issue)
1075 * Array size pushed by the field, not by the field type (or each
1076 and every long int will push stuff to the cbps) [MQ]
1077 * use select() to sleep since it allows to portably sleep less than one
1080 GRAS (minor cleanups)
1081 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1084 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1085 * type category 'ignored' killed as it was never used and were difficult
1087 * whether a type can cycle or not is now a flag, leaving room for more
1088 flags (as "ignored", if we feel the need one day ;)
1089 * Rename raw sockets to measurement sockets since "raw" has another
1090 meaning in networking community.
1093 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1094 and offers features not belonging to GRAS but that most applications
1095 need. One day, it may be a set of plugins loadable at runtime.
1096 * New module: bandwidth
1097 bandwidth measurement between arbitrary nodes running this module. [MQ]
1099 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1101 SimGrid (2.94) unstable; urgency=low
1103 The first beta release of SimGrid 3 !
1107 * Update the main page and the FAQ. Adding references to gforge.
1110 * Add a gras_os_getpid function.
1113 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1114 * Extend the logs so that they also print PID, hostname, date, ... if
1116 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1117 and kill the old version which were in testsuite/
1118 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1122 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1123 for users. Instead, each programming environment (GRAS, MSG,...) use it
1124 when needed to provide such a feature to users.
1125 Don't shortcut the mecanism or you will also shortcut the virtualization
1126 you need on the simulator.
1130 * Cleanups in configury with regard to compile optimization/warning flags.
1131 Also add -fno-loop-optimize to any powerpc since it's the optimization
1132 killing gcc (< 3.4.0).
1133 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1134 needed by MSG examples complications
1135 * Borrow configury beautifications from PHP
1138 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1139 warning (thanks loris for stumbling into this one).
1140 * Bugfix: stop loading private headers (gras_config.h) from the public
1144 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1145 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1146 RL would imply a DNS resolver, which is *hard* to do in a portable way
1147 (and therefore delayed).
1148 * Implement a real timer mecanism and use it in timing macros. This allows
1149 to avoid rounding errors and get a 0.000005 sec precision in timing
1150 macros. While I was at it, various cleanups:
1151 - allow to declare more than one timed section per file (fix a stupid bug)
1152 - move some private declaration to the right place
1153 - merge conditional execution and timing macros into emulation module
1154 - document the module
1155 - make sure the module cleanups its mess on gras_exit
1156 * Documentation improvements:
1157 - (new) how to compile applications using GRAS
1158 - (new) emulation support (timing macros)
1160 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1162 SimGrid (2.93) unstable; urgency=low
1164 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1167 - Use Paje properly where used. Still to be sanitized properly.
1168 - Portability fix: Add an implementation of the contexts using pthread
1172 - Add xbt_procname(): returns the name of the current process.
1173 Use it to show the current process's name in all logging.
1175 - fix detection of older flex version and the reaction, since we do
1176 depend on modern ones (we use lex_destroy)
1177 - Better separation of SG and RL in the libs: remove all simulation code
1178 from libgras. As a result, this lib is now only 200k when stripped.
1179 Some of the xbt modules may also be duplicated (two sets and such) and
1180 should be cleaned/killed before SG3.
1181 - Insist on using xlC on AIX because of weird problems involving gcc there.
1182 - Cleanup the make remote stuff. This is now done by scripts
1183 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1184 mainly for our private use, but we're working on changing them to user
1187 - Bugfix: flush the socket on close only if there is some *output*.
1188 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1190 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1193 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1195 SimGrid (2.92) unstable; urgency=low
1197 Alpha 3 on the path to SimGrid 3
1201 - New! First try of benchmarking macros.
1202 - New! First try so that gras_stub_generator generate deployment and
1203 remote compilation helpers.
1205 - Bugfix: Initialization fix in msg_test.
1209 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1211 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1213 - API Change: gras_os_sleep to take the amount of seconds as a double.
1214 Accepting an int was error prone since it was the only location where
1215 seconds were coded as such. It leaded to damn rounding errors.
1216 - Bugfix: Hard to belive that timers ever worked before this.
1218 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1220 SimGrid (2.91) unstable; urgency=low
1222 Alpha 2 on the path to SimGrid 3
1226 - Bug fix in the lmm_solver.
1228 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1229 through the function MSG_paje_output.
1230 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1231 - It is possible to bound the rate of a communication in MSG with
1232 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1233 to write it in the changelog).
1234 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1239 - Various cleanups to the autotools stuff
1240 - Begin to move Gras examples to examples/gras/
1241 - Let make distcheck work again (yeah!)
1243 - documentation overhauled using doxygen.
1244 gtk-doc-tools is dead in SimGrid now.
1245 - Automatically extract all existing logging categories, and add the list
1246 to the documentation (long standing one, to say the less)
1248 - Cleanup the known architecture table. Reorder the entries to group what
1249 should be, and use a more consistent naming scheme.
1250 (some of the test dataset are still to be regenerated)
1251 - New! Allow library to register globals on each process just as userdata
1253 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1254 lookup time (for now).
1255 Use it in msg and trp.
1256 This cleans a lot the internals and helps enforcing privacy of the
1257 headers between the gras components.
1258 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1259 - Bugfix: gras_os_time was delirious in RL.
1260 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1262 - Reenable GRAS now that it works.
1264 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1266 SimGrid (2.90) unstable; urgency=low
1268 Alpha 1 on the path to SimGrid 3
1270 * It is a long time since the last release of SimGrid. I'm sorry about
1271 that but as I had told you, I was rewriting a lot of things. I apologize
1272 to those who had been reporting bugs to me and that I had not answered.
1273 If your bug is still in the new version, please tell me. Here is a
1274 summary of the main changes.
1276 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1277 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1278 portability, speed, and a lot more but you'll figure it out later.
1279 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1280 I wanted to release the new MSG as soon as possible and I have
1281 broken GRAS, which is the reason why, for now, only MSG is fully
1282 functional. A laconic description of these projects is available
1283 in the documentation.
1285 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1286 kernel optimized for our needs (called SURF but only the developers
1287 should use it). Hence, MSG is now roughly 30 times faster and I think
1288 that by rewriting a little bit MSG, I could event speed it up a little
1289 bit more. Beside the gain in speed, it is also much easier to encode a
1290 new platform model with SURF than it was with SG. More to come...
1292 * REVOLUTION 3: I have tried to change a little as possible the API of
1293 MSG but a few things really had to disappear. The main differences
1294 with the previous version are :
1295 1) no more m_links_t and the corresponding functions. Platforms are
1296 directly read from a XML description and cannot be hard-coded
1297 anymore. The same format is used for application deployment
1298 description. The new format is described in the documentation.
1299 Have a look in tools/platform_generation. There is a tiny script
1300 that converts from the old platform format to the new one. Concerning
1301 the application deployment format, parsing the old one is tricky.
1302 I think most of you should however be able to convert your files. If
1303 it is really an issue, I can write a C code that does the conversion.
1305 2) the toolbox tbx does not exist anymore. We now have a library
1306 with much more data-structures but without the hash-tables (we have
1307 dictionaries that are much faster).
1309 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1311 *****************************************************************************
1312 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1313 * don't feel like converting the oldies. (Mt) *
1314 *****************************************************************************
1317 Version 2.90: "the long awaited one"
1318 - Finished rewriting and debugging MSG. Rewrote the documentation.
1319 - disable GRAS for now since it needs to be ported to the newest SG
1322 - Finish the port to windows (using mingw32 for cross-compile)
1325 - Main loop and datastructures of SURF. A cpu resource object is
1326 functional. Surf can thus be used to create cpu's with variable
1327 performance on which you can execute some actions.
1329 2004-11-15 Martin Quinson
1330 - Port to ARM. Simply added the alignment and size descriptions. Should
1331 work, but the ARM machines are so slow that I didn't had the opportunity
1332 to 'make check' over there yet.
1334 2004-11-15 Arnaud Legrand
1335 - Trace manager now written. It uses a heap structure and is therefore
1336 expected to be efficient. It may however be speeded up (particularly
1337 when many events occur at the same date) by using red and black
1338 trees. One day maybe...
1339 - Max-min linear system solver written. It uses a sparse matrix
1340 structure taking advantage of its expected use. Most operations are
1341 O(1) and free/calloc are called as few as possible. The computation of
1342 the minimum could however be improved by using a red and black tree
1345 2004-11-03 Arnaud Legrand
1346 - Rename every gras_* function that was in xbt/ to its xbt_
1348 - Add a heap and a doubly-linked list to xbt
1349 - Added a dichotomy to the dictionaries. make check works as well before
1350 so I assume that the patch is correct. I do not know however if things
1351 run effectively faster than before now. :)
1353 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1354 SimGrid, and the version number is bumped to 2.x
1356 2004-10-29 Martin Quinson
1357 - Introduction of the remote errors.
1358 They are the result of a RMI/RPC on the remote machine.
1359 ErrCodes being scalar values, you can't get the host on which those
1360 errors did happen. Extending the error mechanism as in Gnome is possible.
1361 No idea yet whether it is a good idea.
1363 2004-10-28 Martin Quinson
1364 - Interface revolution: the Starred Structure Eradication.
1365 I used to do typedef struct {} toto_t; and then handle *toto_t.
1366 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1367 - ???_t is a valid type (builded with typedef)
1368 - s_toto_t is a structure (access to fields with .)
1369 - s_toto is a structure needing 'struct' keyword to be used
1370 - e_toto_t is an enum
1371 - toto_t is an 'object' (struct*)
1373 typedef struct s_toto {} s_toto_t, *toto_t;
1374 typedef enum {} e_toto_t;
1375 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1376 s_toto_t) is private.
1378 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1379 so that it can set the variable to NULL. It was so for dicts and sets,
1380 it changed for dynars.
1382 - Fix a bunch of memleaks in dict_remove
1383 - Fix a bug in sg/server_socket opening: it failed all the time.
1385 2004-10-07 Martin Quinson
1386 - Speed up dynar lookup operation a bit.
1388 gras_dynar_get is dead.
1390 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1391 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1392 (which gives you the address of the stored data).
1394 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1395 copy of the data using an affectation to do the job and not a memcpy.
1397 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1398 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1400 It does not really speedup the dynar test because they are
1401 setting elements all the time (and look them seldom). But the dict does
1402 far more lookup than setting.
1404 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1406 2004-10-05 Martin Quinson
1407 - Allow to (en/dis)able the cycle detection at run time.
1409 Whether we should check for cycle or not is now a property of each
1410 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1411 datadesc_cycle_unset allow to remove this property when previously set.
1413 Note that the cycle detection is off by default since it impacts the
1414 performance. Watch the data you feed GRAS with ;)
1416 This property is hereditary. Any element embedded in a structure having it
1417 set have it set for the time of this data exchange.
1419 You should set it both on sender and receiver side. If you don't set it on
1420 sender side, it will enter an endless loop. If you forget on receiver
1421 side, the cycles won't be recreated after communication.
1423 - Header reorganization.
1424 Kill gras_private.h, each submodule must load the headers it needs.
1426 2004-10-04 Martin Quinson
1427 - Interface revolution: do not try to survive to malloc failure.
1429 Now, gras_malloc and friends call gras_abort() on failure.
1430 As a conclusion, malloc_error is not a valid error anymore, and all
1431 functions for which it was the only gras_error_t return value are
1432 changed. They now return void, or there result directly.
1433 This simplify the API a lot.
1435 2004-09-29 Martin Quinson
1436 - Re-enable raw sockets.
1437 Created by gras_socket_{client,server}_ext;
1438 Used with gras_raw_{send,recv}
1441 It should allow to kill the last bits of gras first version soon.
1443 This is not completely satisfactory yet (duplicate code with
1444 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1447 - Simplify transport plugin (internal) interface by not passing any
1448 argument to _server and _client, but embedding them in the socket
1451 2004-09-28 Martin Quinson
1452 - Finish the port to AIX.
1453 autoconf was my problem (segfault within the malloc replacement
1454 function. No idea why)
1456 2004-09-16 Martin Quinson
1457 - Fix some size_t madness on 64bit architectures.
1459 2004-09-08 Martin Quinson
1460 - Reduce the number of system headers loaded, overload some more system
1461 calls (such as malloc to cast the result of the system one, and work
1463 - Fix and reintroduce the config support
1465 2004-09-07 Martin Quinson
1466 - Source code reorganization to allow Arnaud to surf all over there.
1467 - Allow to document the logging categories.
1468 - Remove all uppercase from logging categories and useless cleanup in names.
1470 2004-08-18 Martin Quinson
1471 Version 0.6.2 (protocol not changed; API changed)
1472 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1473 gras_error_t), and NULL when not found. Functions expecting a msgtype
1474 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1475 hopefully usefull message.
1476 - Portability to prehistoric sparcs again
1478 2004-08-17 Martin Quinson
1479 Version 0.6.1 (protocol not changed; ABI not changed)
1480 - prealloc some buffers to speed things up
1482 2004-08-11 Martin Quinson
1483 Version 0.6 (protocol not changed; ABI expended)
1484 - The parsing macro can deal with the references, provided that you add
1485 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1487 2004-08-09 Martin Quinson
1488 Version 0.5 (protocol not changed; ABI changed)
1489 - Allow to off turn the cycle detection code in data exchange at
1490 compilation time. It should be at run time, but I'm short of time (and
1491 the config stuff is still broken). That way, we keep dict out of the
1492 critical path, which is good because the performance is poor:
1493 - search not dichotomial yet
1494 - dynar give no way to access their content and memcpy everytime
1495 - In composed data description (struct, ref and so on), stop foolness of
1496 keeping the subtype's ID, but store the type itself. This keeps sets out
1497 of the critical path, which is good since they rely on dynar and
1498 dictionnaries. The only loose of that is that we cannot detect the
1499 redeclaration of a structure/union with another content (but I'm not sure
1500 the code detected well this error before anyway). We still can detect
1501 the redefinition discrepancy for the other types.
1502 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1503 it breaks the code because of type-punning used all over the place).
1504 This breaks on all non-gcc architectures (for now).
1506 All those changes (plus the buffer of last time) allow me to gain 2 order
1507 of magnitude on cruel tests consisting of 800000 array of integers on two
1508 level of a hierarchical structure (200 secondes -> 4 secondes)
1511 - the selector of reference must now return the type it points to, not
1512 the ID of this type.
1514 2004-08-06 Martin Quinson
1515 Version 0.4 (protocol changed; ABI not changed)
1516 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1517 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1518 - Implement a buffer transport, and use it by default (it relies on tcp in
1519 real life and on sg in simulation).
1520 That's a bit hackish since I had a new field to the structure to store
1521 its data without interfering with the subtype ones. Inheritance
1522 is tricky in C. And that's a kind of reverse inheritance with one class
1523 derivating two classes. Or maybe a game with java interfaces. Anyway,
1524 that's damn hard in C (at least).
1525 Moreover, I got tired while trying to ensure plugin separation and
1526 genericity in SG mode. MSG wants me to do weird things, so let's go for
1527 cruel hacks (temporarily of course ;).
1528 See comment in transport_private.h:71
1529 - do not include all the _interface headers in private but in the files
1530 which really need them (to cut the compilation time when they are
1533 2004-07-26 Martin Quinson
1534 Version 0.3 (protocol not changed; ABI changed)
1535 - Major overhault of the datadesc interface to simplify it:
1536 - shorted the function names:
1537 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1538 - add a trivial way to push/pop integers into the cbps without malloc.
1539 This allows to make really generic sub_type description, which simply
1540 pop their size of the stack.
1541 - add a function gras_datadesc_ref_pop_arr() which does what users want
1542 most of the time: Declare a dynamic array (which pops its size of the
1543 stack) and declare a reference to it. Poor name, but anyway.
1544 - kill the post-send callback, add a post-receive one
1546 2004-07-23 Martin Quinson
1547 Version 0.2 (protocol changed; ABI changed)
1548 - add some testing for cpbs in the test cases, and fix some more bugs.
1549 This invalidate again the little64 data file, since I cannot regenerate
1551 - remove an awfull optimization in the logging stuff, allowing me to:
1552 - understand it again
1553 - learn gcc how to check that the argument match the provided format
1554 - fix all errors revealed by gcc after that
1555 - internal keys of dict are not \0 terminated. Deal with it properly in
1556 loggings instead of segfaulting when the user want to see the logs :-/
1558 2004-07-22 Martin Quinson
1559 - Fix some stupid bug preventing cbps (callback postit) from working
1561 2004-07-21 Martin Quinson
1562 - Some documentation cleanups
1563 - remove the useless last argument of msgtype_declare
1564 - rename the Virtu functions to fit into the 'os' namespace
1565 - move headers src/include -> src/include/gras/ and stop fooling with
1566 gras -> . symbolic link
1567 - make distcheck is now successful
1569 2004-07-19 Martin Quinson
1571 - Build shared library also
1572 - Install html doc to the right location
1573 - stop removing maintainer files in make clean
1574 - build tests only on make check
1576 2004-07-13 Martin Quinson
1578 - No major issue in previous version => change versionning schema
1579 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1580 corresponding dataset.
1582 2004-07-11 Martin Quinson
1584 - Get it working with any kind of structure (we can compute the padding
1585 bytes remotely for all the architectures I have access to)
1586 - Implement the structure parsing macro (still not quite robust/complete)
1587 - Improvement to the remote testing toysuite
1589 2004-07-10 Martin Quinson
1590 [autoconf mechanism]
1591 - get ride of a bunch of deprecated macros
1592 - actually run the test for two-compliment, not only compile it :-/
1593 - test whether the structures get packed (and bail out if yes. Damn.
1594 Alignment is a serious matter)
1595 - test whether the structures get compacted (but respecting the alignment
1596 constraints of each types)
1597 - test whether the array fields of structures can straddle alignment boundaries
1599 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1600 (took me 2 hours to find that bug, looking at the wrong place)
1601 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1602 before sending/receiving and are used to compute the offsets of fields
1603 - Given that padding size depend even on compiler options, keep track of
1604 alignment and aligned_size only for the current architecture. Not a big
1605 deal since we send structure fields one after the other (seems
1607 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1608 program, let it work on linux/gcc/little32. portability todo.
1610 2004-07-08 Martin Quinson
1611 - import and improve remote compilation support from FAST
1612 - make sure make check works on half a dozen of machines out there
1614 2004-07-07 Martin Quinson
1615 Let's say it's version 0.0.3 ;)
1616 - Implement conversions (yuhu!)
1617 - Let it work on solaris (beside conversion, of course)
1618 - Stupid me, using rand() to generate the conversion datatests in not wise.
1620 2004-07-06 Martin Quinson
1621 - Let make dist work, since I'm gonna need it to compile on remote hosts
1622 - Let Tests/datadesc_usage write the architecture on which the file was
1623 generated as first byte.
1624 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1625 architecture descriptions.
1626 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1627 on those architectures.
1628 - Optimization: send/recv array of scalar in one shoot
1630 2004-07-05 Martin Quinson
1631 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1633 - Plug a whole bunch of memleaks
1634 - each process now have to call gras_{init,exit}. One day, their log
1635 settings will be separated
1636 - Continue the code factorisation between SG, RL and common in Transport.
1638 2004-07-04 Martin Quinson
1640 - Redistribution between SG and RL.
1641 We wanna have to accept in SG, so move accepted related parts of RL in
1642 the common part. (more precisely, the dynar of all known sockets is no
1643 more a static in transport.c, but part of the process_data)
1645 [gras_stub_generator]
1646 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1648 2004-07-03 Martin Quinson
1649 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1650 them all up in one shot)
1652 - Fix the ugly case of reference to dynamic array.
1653 - New (semi-public) function gras_datadesc_size to allow the messaging
1654 layer to malloc the needed space for the buffer.
1656 - gras_socket_close now expect the socket to close (and not its address to
1657 put NULL in it after it). This is because the socket passed to handlers
1658 is one of their argument (=> not writable).
1660 - propagate the interface cleanup from last week in datadesc, ie remove a
1661 superfluous level of indirection. User pass adress of variable
1662 containing data (both when sending and receiving), and not of a variable
1663 being a pointer to the data. Let's say that I like it better ;)
1664 The price for that is constructs like "int msg=*(int*)payload" in
1665 handlers, but it's a fine price, IMHO.
1667 - Let it work in RL (yuhu)
1669 2004-06-21 Martin Quinson
1671 - porting SG plugin and SG select to new standards (works almost).
1672 - plug memleaks and fix bugs around.
1675 - cleanup the prototype of data recv and force users to specify when they
1676 want to handle references to objects. Test case working even for cycles.
1677 - plug memleaks. Valgrind is perfectly ok with this.
1679 2004-06-12 Martin Quinson
1681 - cleanup the separation between plugin and main code in plugin creation
1683 2004-06-11 Martin Quinson
1685 - Reput hook for raw sockets, needed for BW experiments
1686 - kill a few lines of dead code
1687 [Data description] Interface cleanup
1688 - gras_datadesc_by_name returns the searched type or NULL.
1689 That way, no variable is needed to use a type desc once, which makes
1691 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1692 two parameters were strings (field name, type name), leading to
1694 [Dicos] Interface cleanup
1695 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1696 This is consistant with the dynar API.
1698 2004-04-21 Martin Quinson
1700 - Porting to new standards.
1702 - interface cleanup.
1703 There is no bag anymore, no need to take extra provision to mask the
1704 pointers behind "ID".
1705 Better splitup of functions between files create/exchange/convert.
1706 This is still a bit artificial since convert and receive are so
1707 interleaved, but anyway.
1709 - add a queued message list to procdata (the ones not matching criteria
1711 - factorize some more code between SG and RL wrt procdata
1713 - use gras_exit in example to track memleaks
1714 - get rid of gs_example now that GS is properly integrated into gras
1715 - update run_test to integrate the lastest tests (datadesc)
1717 - rename WARNINGn macros to WARNn since it prooved error-prone
1719 2004-04-19 Martin Quinson
1721 - register init/exit functions within gras module mechanism
1722 - send/receive function.
1723 Convertion is not implemented, but short-cutted if not needed.
1724 struct/array elements are sent one by one (instead of block-wise), but
1725 nobody really cares (yet). Get a prototype before optimizing.
1726 - tests (using a file socket) for DD send/receive on:
1727 - base types: int, float
1728 - array: fixed size, string (ie ref to dynamic string)
1729 - structure: homogeneous, heterogeneous
1730 - chained list, graph with cycle
1731 Believe it or not, valgrind is not too unhappy with the results. The
1732 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1736 - Bugfix when using a filename explicitely (instead of '-')
1738 2004-04-09 Martin Quinson
1740 - factorize more code between RL and SG in socket creation
1741 - Complete the implementation and tests of:
1743 o file (only in RL, and mainly for debugging)
1745 I lost 3 days to design a portable address resolver, and then decided
1746 that the prototype mainly have to run on my box.
1747 Addressing portability too early may be like optimizing too early :-/
1749 - use gras_init in the Tests instead of the crappy parse_log_opt
1750 (the latter function is removed)
1751 [Conditional execution]
1752 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1753 [Code reorganisation]
1754 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1755 Build examples against the RL library, since there is no way to disable
1756 its creation for now.
1758 For information, the beginning of coding on GRAS was back in june
1759 2003. I guess that every line has been rewritten at least twice since