1 SimGrid (3.4-svn) unstable; urgency=low
3 The "Easter in Cargese" release. Also known as (major changes):
5 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
6 ~> bindings were greatly improved
7 ~> new tracing infrastructure for better visualization introduced
9 * the "Welcome to configury modernity" release.
10 ~> we switched from autotools to cmake, and improved our cdash
13 A more detailled list of changes follow (full detail in svn log).
15 Java Bindings: Various Cleanups
16 * (install java-gcj-compat-dev on debian-like to use them)
17 * Remove put/get: no need to export deprecated interface in Java
18 Use send/receive instead.
19 * Cleanup the examples and add a README per directory
20 * Remove example autoDestination (that's the only way to go now)
21 * Remove example explicitDestination (was a plain copy of basic)
22 * Make JniException a runtime exception, so that there is no need to
23 declare the fact that you may encounter such a beast. I guess that
24 nobody will ever want to survive such error.
25 * Create specific errors for each MSG case of failure:
26 host failure, transfer failure, timeout, task cancelled
27 * Cleanup the exceptions that may get thrown by each function
28 * Other internal cleanups in Java bindings. Performance still bad :/
29 Ruby and Lua Bindings: create them
30 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
31 * That's new and great, you should try them out.
32 Same functionalities than Java bindings, only even less polished
34 * Kill the useless "rate" argument of SD_task_get_execution_time()
35 Everyone used to provide -1 as a value, it was not used, and the
36 semantic of a possible use wasn't even clear.
37 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
38 either as comm costs or compute costs to mean that there is no
39 such thing for that specific task.
40 * Fix SD_task_unschedule() on typed tasks
41 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
43 * Add an example masterslave_mailbox.c using send/receive and not
44 the deprecated put/get interface.
45 * Kill the MSG_paje_output() function. It's a noop since 2 years.
46 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
48 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
49 * Add a MSG_task_set_data() function
50 * About trace replay (see examples/msg/actions):
52 - Allow to work with splitted trace files for each process
53 Give the specific trace file as argument of each process,
54 and call MSG_action_trace_run(NULL)
55 You can still have one merged file for all processes.
56 - Fix implementation of collective operations
58 * This is the first release of SimGrid where SMPI is not considered
59 beta anymore (even if some corners should still be improved)
60 * Port over the new SIMIX_network submodule (internal refactoring)
61 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
62 * Implement more missing elements of the standard:
64 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
66 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
67 MPI_Scan() MPI_Get_processor_name()
68 - Added implementation of missing case for Alltoall (warning: it's
69 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
70 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
71 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
74 - MPI_Waitsome() was broken
75 - Allow relative includes in smpicc
76 - Command line cfg argument 'reference_speed' was ignored...
77 - Some functions did not properly lead to auto-benching of user code
78 - smpicc passes -O2 by default (just like openmpi one)
80 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
81 * Bug fixes about timeouts during communications
82 * add SIMIX_message_sizes_output() as a pimple to write to file the
83 amount of messages per size. Use gnuplot to get histogram.
84 Pimple because that's the only user-visible function of simix,
85 defined directly in xbt.h (irk, sorry)
87 - Add a SIMIX_sem_get_capacity() function
88 - Fix interactions with processe resume/suspende
89 - release_forever() was stupidly broken
90 - Fix SIMIX_display_process_status() for processes in a semaphore
91 - Make SIMIX_sem_block_onto() user-visible
92 * Refactoring context stuff:
93 - Use pseudo-OOP for better modularity
94 - reimplement SIMIX_process_kill() without process_schedule() so
95 that the latter can take as invariant that it is called from
97 - Merge context_start into context_new for sake of simplicity
99 * Rename configuration variables to start a hierarchy:
100 o cpu_model -> cpu/model
101 o network_model -> network/model
102 o workstation_model -> workstation/model
103 * New configuration variables:
104 o network/bandwidth_factor: correction to bandwith
105 o network/latency_factor: correction to latency
106 o netwotk/weight_S: correction to the weight of competing streams
107 * Add a long description to the models, that users can see with such
108 argument on the command line: --cfg=cpu/model:help
109 * --help-models display the long description of all known models
111 * config: add the ability to set a default value after registration
112 Does not override any previously set value (e.g. from cmd line)
113 * dict: allow to have integer key and data.
114 When so, you need to use the following functions
115 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
116 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
117 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
118 In contrary to regular dicts, the key is not malloced before copy.
119 Mixing scalar and regular elements in the same dict is not tested
121 * Allow to use xbt_dynar_shrink() to expend the dynar instead
122 Tracing for Visualization:
123 * SimGrid is now instrumented in order to generate a trace file for
124 visualization analysis: to use it, need to compile SimGrid with the
125 "tracing" option enabled, and instrument the program using SimGrid with
126 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
127 (among other functions).
128 * The instrumentation only traces the platform utilization for now
129 * Documentation to use the tracing functions and how to analyze the
130 traces with the Triva tool is written.
131 * More information about: SimGrid FAQ (in the section Tracing Simulations
134 * We moved to cmake as default build system. Autotools support will
135 be dropped soon. Check the FAQ for more info about how to use it.
136 * Greatly improved our cdash/ctest interactions
137 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
138 * Added memory checking tests with valgrind; lot of memleak fixing.
139 This may be the first release of simgrid with so few memory issues
140 * Added code coverage tests.
141 Our coverage is still improvable, but at least we see it on cdash.
143 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
146 SimGrid (3.3.4) stable; urgency=low
148 The "Desktop Grid needs love too" release (also called Xmas release).
151 * Major speedup in the maxmin system solving by using lazy evaluation
152 Instead of solving completely the maxmin system at each iteration,
153 only invalidate (and recompute) the modified parts.
154 This new feature is enabled in default models but you can try to
155 turn it on with "--cfg:maxmin-selective-update=1" for other models.
156 * Cas01 IMproved as default CPU model
157 This CPU model is the same Cas01 model, but it uses the
158 maxmin-selective-update flag and a heap structure to manage
159 actions on SURF kernel.
160 It reduces the complexity to find the next action to finish and,
161 consequently, it's faster than the old Cas01.
162 This is the new default CPU model (Cas01).
163 * Rename the old Cas01 model to Cas01_fullupdate
164 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
165 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
166 * CpuTI (CPU Trace Integration)
167 A new CPU model whose objective is simulate faster when using
168 availability trace files.
169 Instead of using a full featured, over engineered maxmin system for
170 CPU modeling, this model does the pre-integration of traces files
171 to calculate the amount of CPU power available, and so, executes
172 faster than the old CPU models.
173 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
174 * Use LV08 as default network model since it gives better accuracy
175 for small messages and shouldn't change things for big ones.
176 Use --cfg=network_model:CM02 to get the previous behavior.
179 ******************************************
180 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
181 ******************************************
182 * The new CPU model may changes simulations!
183 The point is that events occurring at the exact same timestamp
184 are not scheduled in the same order with the old and new
185 version. This may be enough to completely change the execution
186 of simulations in some cases.
187 * The new network model will change simulations!
188 This new model is more realistic than the previous one, so you
189 should consider redoing your old experiments with this model.
190 Sorry for the inconvenience.
193 * Introduce the supernovae compilation mode
194 When compiled that way, the whole SimGrid (or almost) is put in a
195 single compilation unit and compiled in one shoot.
196 This is to help gcc which has difficulties to inline stuff from one
198 The speedup seem to be above 15%, althrough more tests are needed on
199 amd64 to confirm that gain.
202 * Port of MSG's mailbox on top of SIMIX network
203 The put/get mechanism was greatly simplified on the way.
206 * New SIMIX network module. Provides:
207 - Mailbox: rendez-vous mecanism to find with who you want to speak
208 - Synchronous send/recv: easier and hopefully faster since the
209 logic is handled in the maestro process directly now
210 - Asynchronous send/recv: you dreamt of it? It's here now
211 Too bad that nobody cared enough to propagate the change to MSG.
212 * Add semaphores as SIMIX synchronization mechanism.
215 * new function SD_daxload(char*) to load a DAX file
216 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
217 * Introduce typed tasks. Specify its kind and cost at creation.
218 At scheduling, just give where it should be placed, and the cost
219 for each involved resource is automatically computed.
220 Existing constructors so far (more to come of course):
221 - SD_task_create_comm_e2e() for end-to-end communication
222 - SD_task_create_comp_seq() for sequential computation
223 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
224 * new function SD_task_dump() for debuging display
225 * new function SD_task_dotty(task,FILE*) writing to file the info
226 about the task in dotty format
227 * SD_task_dependency_exists() can now cope with having one of its
228 arguments NULL. If so, it tests whether the other argument has any
230 * Add getters on list of preceding/following tasks:
231 SD_task_get_parents(task) and SD_task_get_children(task)
232 * Add getters on amount of workstations and list:
233 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
234 * Add getter on task kind: SD_task_get_kind(task)
235 * Update the start_time and finish_time of tasks on completion/failure
236 * Bugfix: Remove task from state swags when destroyed
239 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
242 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
243 dynar with one separate thread per value of the dynar.
244 * Change the prototype of xbt_thread_create(), sorry.
245 Added a boolean parameter indicating whether we want to join this
246 thread (used in SG only for now)
247 * Implement xbt_thread_join and xbt_thread_yield in SG also.
250 * GTNetS wrappers should now be usable again (and betterly tested too)
251 * Fix a major regression from 3.2 where the timeout provided to
252 MSG_task_put_with_timeout() was used as absolute time before which
253 the comm should be done.
254 * Start to fix the <cluster> tag.
255 - Internal links should be good now (beside of the loopback, which
256 use the private link instead)
257 - paths to the external world is still rather broken
258 - the <route:multi> tag is just broken. Actually that's brain-dead.
259 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
260 to make it less stupid
261 ** Check your platform with teshsuite/simdag/platforms/flatifier **
262 * Fix a source-level compatibility glitch from 3.2: after defining
263 MSG_USE_DEPRECATED, you can use the old name
264 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
265 * Allow to compile from the SVN with automake 1.11
266 * Fix some problems when using the "start_time" tag in deployment XMLs.
267 * Fix #8569: XBT/synchro.h has redundant declarations
268 * Fix #8563: MSG return values and exceptions
269 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
270 * Integrate patch #8636: Obey DESTDIR when installing documentation.
271 Thanks to Robson Peixoto.
272 * Fix a vicious bug in dictionaries inducing that some elements were
273 not freed on xbt_dict_free()
275 Portability report of this version:
276 * Main portability targets:
277 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
279 - mac leopard on i386
280 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
281 but nothing critical.
282 * Other platforms: windows, AIX and others were not tested for this release
284 Timing report of this version:
285 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
286 scenario parameters). From 8h to a few seconds in desktop grid settings.
287 * Supernovae brings about 25% speedup on i386.
289 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
291 SimGrid (3.3.3) stable; urgency=low
293 The "Need for Speed" release.
295 The timings done to validate the 3.3.2 were faulty.
296 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
298 The problem was a conversion from a manually handled vector to
299 xbt_dynar_t on the critical path.
300 xbt_dynar_foreach calls functions, inducing stack management crap.
302 We inlined these functions and xbt_dynar_foreach is now breath taking.
303 We also inlined xbt_swag_belong on the way.
305 Here are some approximate speedup measurements (on master/slaves
306 simulations lasting between 10s and 20s each):
307 3.3.1 -> 3.3.2: about same performance
308 3.3.2 -> 3.3.3: 40% speedup
309 3.3.1 -> 3.3.3: 40% speedup
310 3.3.1 with inline patch -> 3.3.3: 30% speedup
312 Our reading is that the refactoring which occurred in 3.3.2 made us
313 suffer much more from the xbt_dynar_foreach low performance, but
314 once we solved this, this refactoring proved to be very performance
315 effective. From the 40% speedup, somehow, 10% are due to the
316 inlining and 30% to the refactoring.
318 That's a pitty that gcc cannot inline functions placed in other files
319 alone. We have to choose between:
320 - break the encapsulation (by putting private data structures and
321 accessors in headers files to help gcc)
322 - live with low performance
323 - switch to a decent compiler such as icc (not quite possible).
325 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
327 SimGrid (3.3.2) stable; urgency=low
329 The "Simplicity does not preceed complexity, but follows it" release.
331 The main contributors of this release were (lexical order):
332 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
335 * Extract the routing logic into its own object.
336 (was dupplicated in network.c and workstation_LV07.c;
337 Allows to implement other ways of storing that info)
338 => kill now useless network_card concept
339 - Use dynar to represent routes (instead of void** + int*)
340 - kill link_set (use surf_network_model->resource_set instead)
341 - Add a command-line option to choose the routing schema to use
342 - Add three new models:
343 * Floyd (shortest path computed at initialization)
344 * Dijikstra (shortest path recomputed all the time)
345 * Cached Dijikstra (shortest path computed on need)
346 All these models where contributed by Silas De Munck, and are
347 described in his ICCS09 paper.
349 * Simplify model declaration
350 (less redirections, less function to write when defining a model)
351 - Factorize stuff between models:
354 surf_model_resource_set(model)
355 surf_model_resource_by_name(model, name)
356 - Unify the types of models in s_surf_model_t (using an union)
357 - Embeed fields of common_public directly into s_surf_model_t
358 - Rename model methods:
359 action_free ~> action_unref
360 action_change_state ~> action_state_set
361 action_get_state ~> action_state_get
362 - Change model methods into functions :
363 (model)->common_public->action_use ~> surf_action_ref
365 * Implement a generic resource; use it as ancestor to specific ones
366 (allows to kill duplicated code in models)
367 Drawback: timer command don't need no name nor properties;
368 workstation_CLM03 don't need no properties
369 (but I guess we can live with those few bytes wasted)
371 * Improve the action object model
372 - implement a constructor avoiding dupplicated code about field
373 initialization in generic_action part.
375 * Kill the SDP model: it has an external dependency, is deprecated
376 in flavor of modern lmm models, and didn't compile since a while
379 * Relocation of the context module from XBT to SIMIX.
380 (the context were decoupled from the simix processes, duplicating a lot of code)
381 => a lot of code was factorized
382 - less overhead is introduced during scheduling
383 - simpler API for the context factory
384 - the logic for process creation,destruction and manipulation was simplified
385 * Simplification of the s_smx_process_t data structure.
386 => accesing the simix level data associated to a process is faster now,
387 and the code is a lot more readable.
390 * Implement some more MPI primitives:
391 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
392 -implementation: Bcast: flat or 2-ary tree (default),
395 Allreduce: Reduce then Bcast
396 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
397 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
398 Alltoallv: flat tree, like ompi
400 * Add support for optimized collectives (Bcast is now binomial by default)
401 * Port smpirun and smpicc to OS X
404 * Kill SD_link_get_properties: hard to maintain and makes very little sense
405 Shout out if you used it.
408 * Display the list of still queued messages in SG mode when existing
412 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
413 * Add xbt_graph_node_get_outedges() [Silas De Munck]
414 * Add xbt_str_from_file(FILE*)
415 * Add xbt_dict_get_key achieving a linear reverse search
416 * Remove the context module
418 Portability report of this version:
419 * Main portability targets:
420 - Linux(debian)/x86/context
421 - Linux(debian)/x86/pthreads
422 - Linux(debian)/amd64/context
423 - Linux(debian)/amd64/pthreads
424 On these, we still have the eratic breakages of gras/pmm and
425 amok/saturate_sg reported in previous version. We still think
426 that the tests are the cause of the fault, not the tested code.
428 - Mac OSX Leopard/x86/context
429 Still false negative in tesh autotesting.
430 Smpi still fails, but this time because readlink does not accept -f
431 Everything seems to work properly beside of that.
434 - AIX version 5.3 (only tested contexts this time)
435 Smpi still fails there because mktemp is not installed.
436 Everything seems to work properly beside of that.
438 I managed to compile it for the first time, but several breakages.
439 Won't delay the release for this exotic platform.
441 * Windows: it's still lagging behind. If you want to help, please
444 Timing report of this version:
445 This version seem to be more than 5% faster than 3.3.1 (on linux
446 64bits with contextes). The gain is less than expected, we are
447 investigating this for next release.
449 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
451 SimGrid (3.3.1) stable; urgency=low
454 * Implement a --cfg-help to show existing configuration variables
455 * Build chain do not require doxygen in maintainer mode
458 * fix a bug on struct sizeof computation, which prevented the
459 exchange of arrays of structs in some conditions
460 - added a regression test about this in datadesc_usage
461 * Allow the exchange of 0-long dynamic vectors.
462 - for that, use -1 as indicator of dynamic size instead of 0
463 - This implied to change any size from unsigned long to long,
464 reducing a bit communication abilities, but I guess that with
465 64bits being quite common, this is more than enough.
466 - This also induce a protocol change, thus bumping network protocol
467 version from 0 to 1 (if we have external users, we have to get
468 clean on that point too ;)
469 - added two regression tests about this in datadesc_usage
470 * Be more verbose when propagating local exceptions
471 This helps debugging.
472 * Display the status of simulated processes when receiving SIGINT in
476 * Allow to control the simulation from a trace file.
477 New functions MSG_action_register() and MSG_action_trace_run()
478 The first one allows to associate a function execution to each
479 kind of action while the second one parses a trace file and
480 triggers the corresponding actions within the system.
481 For now, only a toy example is provided in examples/msg/actions
482 * Add an exemple of process migration in examples/msg/migration
483 * Fix a bug in task exchange which broke MSG_task_get_sender()
484 Add a teshsuite regression test for that.
485 [Bug: if MSG_task_get_sender() is called after sender exit,
487 * Fix a bug which prevented suspend/resume to work properly
488 * Display the status of simulated processes when receiving SIGINT
489 This fixes a regression of v3.3. due to the introduction of SIMIX
490 * Bug fixing in failure management:
491 - trace could not start by a failure at time 0
492 - failure during communications were not working
495 * Add SIMIX_process_set_name() to change the name of the current
496 process in the log messages.
497 * Store smx_hosts in a dict since we only retrieve them by name
498 * Move the configuration infrastructure to surf
501 * Move the configuration infrastructure to surf
504 * Massive internal cleanups:
505 - Store internal structures on processes instead of hosts (allows
506 to have more than one process per host, in addition of being more
508 - Cleanup the initialization/finalization process
509 - Kill a whole bunch of unneeded synchronization:
510 processes run in exclusive manner within the simulator
511 - Move queues from global tables to process data fields
513 - now accept -platform and -hostfile arguments
514 - Pass the right rank value to processes according to the hostfile
515 * Compile the examples by default, and use them as regression tests
516 * Implement MPI_Wtime()
517 * Change the reference speed to a command line option
520 * TCP_gamma can now be specified as command line option using
521 --cfg=TCP_gamma:10000000.0
522 * Change the --surf-path cmd line option into --cfg=path:
525 * Also include strbuff from xbt.h public header
526 * xbt_ex_display(): do not free the exception after displaying
527 This allows to do more with the given exception afterward.
528 Users should call xbt_ex_free() themselves.
532 Portability report of this version:
533 * Main portability targets:
534 - Linux(debian)/x86/context
535 - Linux(debian)/x86/pthreads
536 - Linux(debian)/amd64/context
537 - Linux(debian)/amd64/pthreads
538 These targets fail about 1/10 of times on gras/pmm, but we believe
539 that this is because of the test, not because of simgrid.
540 amok/saturate_sg fails even more rarely, and the test may not be
543 - Mac OSX Leopard/x86/context
544 The test suite still spits tons of errors because some obscure
545 force prevents us from removing the temporary directories
546 arguing that they still contain some metadata I've never heard of.
547 Smpi fails because seq is not installed.
548 Everything seems to work properly beside of that.
551 - AIX version 5.3 (both contexts and pthread)
552 Smpi still fails there because mktemp is not installed.
553 XML inclusions seems rosty on AIX.
555 * Windows: it's still lagging behind. If you want to help, please
558 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
560 SimGrid (3.3) stable; urgency=high
564 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
567 * Introduce the SIMIX module: factorize code between MSG and GRAS.
570 Until now, GRAS were using MSG as an interface to SURF. It was
571 quite difficult because both interface have several differences
572 (MSG channels vs GRAS sockets were the most notable point).
574 This also opens the gate to SMPI (which should occur soon) and speed
575 up simulations by to 40% (even if it were not the main goal).
577 **************************************
578 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
579 **************************************
580 The point is that events occuring at the exact same timestamp are
581 not scheduled in the same order with the old and new version. This
582 may be enough to completely change the execution of simulations in
583 some cases. Sorry for the inconvenience.
585 * Cleanup and upgrade the XML format to push further scalability
586 issues (check http://hal.inria.fr/inria-00256883/ for more info)
588 * Improve the testing infrastructure with tesh. Now a very large part of
589 the code is tested not only by being run but also by checking that the
590 output match an expected output [Mt].
592 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
593 is really less memory-demanding, which should allow you to use
594 larger files in SimGrid [AL].
596 * Inform valgrind about our contextes, so that it becomes usable
597 with the default (and more effecient) version of SimGrid
598 [contributed by Sékou Diakite, many thanks]
601 * Introduce a listener thread in charge of receiving incomming
602 messages from the network. It allows to overlap communication and
603 computation but most notably, it removes some stupid deadlocks due
604 to the fact that so far, a process could not send and receive at
605 the same time. This made most non trivial communication schema
607 * Convert the PIDs from long int to int to match the MSG ones (and
609 * New function: gras_agent_spawn() to launch a new process on
610 current host. Only working in simulation for now. [Mt]
611 * New function: gras_os_hostport() returning a constant form (ie,
612 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
615 * Make the backtrace of exceptions more human readable [Mt]
616 * New module: xbt/str [Mt]
617 a ton of string utility functions (split, join, printf to a newly
618 allocated buffer, trim, etc)
619 * New module: xbt/hash [Mt]
620 SHA1 hashing algorithm (more to come if needed)
621 * New module: xbt/synchro [Mt]
622 synchronization tools (mutex and conditions) working the same way
623 in simulation and in real life (mainly useful for GRAS, but not
625 * New module: xbt/queue [Mt]
626 classical producer/consumer synchronization scheme
627 * xbt_dynar_new_sync() creates a synchronized dynar. All access
628 (using the classical functions will get serialized) [Mt]
629 * Make dictionary internal table dynamic. No need to specify its size
630 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
632 * Make sure the log channels are organized as a tree under windows
633 (because of ANSI C compatibility issue, any channel were child of
637 * Cleaned many thing in surf and fixed a few bugs [AL].
638 * Add a nice command line configuration mechanism to compose models [AL].
639 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
640 the previous one (KCCFLN05). It relies on something that looks like
641 a max-min sharing mechanism but cannot be written as such. A new solver
642 was thus designed [AL].
643 * Add a new solver to lmm. Based on Lagrange optimization and
644 gradient-based descent, it enables to efficiently maximise systems s.a
646 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
648 This solver enables to propose two new network models for TCP Reno and
649 TCP Vegas based on Low's work. These models still need to be fully
650 tested though [Pedro Velho].
653 * Bug fix in SD_simulate. Now the time bound given as argument is
655 * Use the new parallel task model (ptask_L07) as default.
656 * Use the SURF command line configuration mechanism.
657 * 0-size tasks (for synchronization) should now work.
659 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
661 SimGrid (3.2) stable; urgency=high
665 We still experience issues on this platform, but we believe that at
668 GRAS API BREAKAGE (for simplification purpose, sorry):
669 * the gras_msgtype_by_name is not used anymore. Instead of
670 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
671 you can write (and must)
672 gras_msg_send(toserver, "request", &request);
673 - If you still want to pass a gras_msgtype_t to the function (to cache
674 the type and avoid the lookup time), use the gras_msg_send_() variant.
675 - Impacted functions:
676 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
677 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
678 * The callbacks are now expected to return 0 when everything went well
679 (just like the main() function)
681 GRAS new features and improvements:
682 * New module mecanism where user code can use per process globals [Mt]
683 This is similar to gras_userdata_*() functions, but for libraries. It
684 factorize some code developped over and over in the examples and AMOK.
685 It has still to be documented and used (only amok/peermanagement is
687 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
688 loss when they were small enough to fit into the buffer and sent quickly
689 enough so that they can all get received in one shoot.
690 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
691 if not found. Use the *_or_null() variant for the old semantic.
692 * In gras_msg_handle, do not discard messages without callback.
693 They are probably messages to be explicitly awaited later (ie, proofs of
694 mis-synchronization in userland since they are sent before being awaited)
696 * gras_socket_meas_send/recv: semantic changed!
697 The numerical arguments used to be (1) the total amount of data to send
698 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
699 messages. This was need for the fool willing to send more than MAXINT
700 bytes on quite fat pipes.
703 * Do really rename the hostmanagement module to peermanagement. [Mt]
704 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
705 but this is rather new and this was documented in the module
706 documentation (poor excuses, I admit)
707 * Bandwidth measurement semantic changed! This follows the changes to
708 gras_socket_meas_send/recv explained above.
711 * A sequential mode has been added to the workstations. When a workstation
712 is in sequential mode, it can execute only one task, and the other tasks
713 are waiting in a FIFO. [Christophe Thiery]
716 * The KCCFLN05 workstation model now handles parallel tasks. It is the
717 model for SIMDAG. [Christophe Thiery]
718 * Bug fix in the maxmin solver: Some values were close to 0 instead of
719 equal to 0, which caused some bad behaviors in
720 saturated_constraint_set_update. I now use a threshold mechanism like in
724 * When running manually src/testall, you select specific units [Mt]
725 testall is the result of our cunit mecanism, and should replace all
726 the scripty thingy around since bash don't run easily on billware.
728 * A mallocator system has been added. [Christophe Thiery]
729 Mallocators allow you to recycle your unused objects instead of freeing them
730 and allocating new ones.
732 Documentation update:
733 * FAQ reworking + New FAQs:
734 - "Valgrind spits tons of errors!" [Mt]
735 - "How to repport bugs" [Mt]
736 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
737 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
738 - Communication time measurement within MSG [AL]
739 - I experience weird communication times when I change the latency [AL]
742 - an introduction to the framework and to the used communication model
743 - an initiatic tour introducing the most proheminent features:
745 . Lesson 0: Installing GRAS
746 . Lesson 1: Setting up your own project
747 o Part 2: Message passing
748 . Lesson 2: Exchanging simple messages
749 . Lesson 3: Passing arguments to the processes (in SG)
750 . Lesson 4: Attaching callbacks to messages
751 . Lesson 5: Using globals in processes
752 . Lesson 6: Logging informations properly
753 . Lesson 7: Using internal timers
754 . Lesson 8: Handling errors through exceptions
755 . Lesson 9: Exchanging simple data
756 . Lesson 10: Remote Procedure Calling (RPC)
757 . Lesson 11: Explicitely waiting for messages
758 . Recapping of message passing features in GRAS
759 - A HOWTO section containing:
760 o HOWTO design a GRAS application
761 More are due, of course. They will come latter. In the meanwhile, you can
762 check the examples which are still here.
764 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
766 SimGrid (3.1) stable; urgency=high
770 There was a stack corruption somewhere, visible only when optimizing
771 with these versions. [Vince]
774 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
775 module that enabled to play with Directed Acyclic Graphs. It is built
776 directly on top of SURF and provides an API rather close to the old
777 SG. Some old codes using SG are currently under rewrite to check that
778 all needful functions are provided. [Christophe Thiery]
781 * Complete rewrite of the KCCFLN05 workstation model. It is now an
782 extension of the classical CLM03 model that gracefully handles
783 failures. This is now the default model for MSG and GRAS. It doesn't
784 handle parallel tasks yet though. [AL]
785 * Bug fix: Weights were not correctly set in the network part.
786 WARNING: This may have resulted in incorrect results with simulations
787 where there are more than one flow on a given link. [AL]
790 * After a (long ?) discussion on simgrid-devel, we have decided that the
791 convention we had on units was stupid. That is why it has been decided
792 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
793 WARNING : This means that all previous platform files will not work as
794 such with this version! A warning is issued to ask users to update
796 A conversion script can be found in the contrib module of the CVS, under
797 the name contrib/platform_generation/surfxml_update.pl [MQ]
800 * Bug fix: Processes were started in reverse order, wrt deployment file.
801 WARNING: if your code relies on this bug, please fix it. [AL]
802 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
803 being executed on two different locations. [AL]
804 * Bug fix: Failures are now better supported thanks to Derrick's tests
805 (there was many failure situations I hadn't thought of and that weren't
806 correctly handled). [AL]
807 * New function: MSG_host_is_avail indicates you whether a given m_host_t
811 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
812 Exception occurring on server-side are propagated back to client (!).
814 API CHANGE: the callback changed their prototype. Change:
815 int my_handler(gras_socket_t expeditor, void *payload_data) {
817 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
818 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
820 * New! function: gras_msg_handleall to deal with all messages arriving
821 within a given period.
822 * New! function: gras_socket_server_range to get a server socket in a
823 range of port numbers (ease to avoid port number conflicts) [MQ]
824 * New! gras processes display their backtrace when they get a SIGUSR1
825 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
826 Sweet to debug RL processes [MQ]
830 - Do not force experiment sizes to be expressed in kb, or it becomes
831 impossible to measure the latency this way (needs one byte-long tests)
832 WARNING: this changes the amok_bw_* function semantic. [MQ]
833 - Implements the link saturation stuff. [MQ]
834 * Peer management module:
835 New! module factorizing code that we wrote over and over [MQ].
838 * New module: cunit (my jUnit implementation in ansi C) [MQ]
839 - Test units are placed directly into the library code, they get extracted
840 automatically and placed into the src/testall binary.
841 - Convert most of the XBT tests to this system.
842 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
843 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
844 live in the same C code. This required to change a little bit the API
845 of surfxml parsing but shouldn't be an issue for end-users. [AL]
846 * New module: sparse graph structure with basic algorithms (this is work
847 in progress and the API is not considered to be frozen yet). [AL]
848 * Display more information on backtraces: source line & function names are
849 now displayed just like valgrind does (rely on addr2line tool) [MQ]
850 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
851 * Reworked a little bit some #include statements to load only required
852 headers. Some user code that relied on SimGrid to include stdlib or
853 stdio may need to include it by themselves. [AL]
854 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
856 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
857 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
858 but I guess that almost nobody used those parts. [MQ]
860 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
862 SimGrid (3.0.1) stable; urgency=low
865 * Unfortunately, I had missed 5 misnamed functions:
866 xbt_fifo_item_t xbt_fifo_newitem(void);
867 void xbt_fifo_freeitem(xbt_fifo_item_t);
868 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
869 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
870 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
871 They're now deprecated. Please use their new versions:
872 xbt_fifo_item_t xbt_fifo_new_item(void);
873 void xbt_fifo_free_item(xbt_fifo_item_t);
874 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
875 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
876 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
878 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
879 * Documentation: xbt_log module unmercifully reworked [MQ]
880 * Bugfix: there was a problem with the ending of contexts with
881 the pthread backend. It caused some weird deadlock or behavior
882 depending on the pthread implementation. [AL]
883 * Bugfix: get the exceptions raised in the simulator repport where
884 and why they come from when they are not catched in time [AL, MQ]
887 * Bugfix: Do repport the error when two non-connected hosts try to
888 exchange data (Thanks to Flavien for stumbling into this one) [AL]
891 * Add additionnal checkings on communications. Assert that two
892 communicating hosts are connected by a set of links... [AL]
895 * Add additionnal checkings on channel values in communication [AL]
896 * New: MSG_task_get_source to see on which host a task was generated [HC]
897 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
898 the number of tasks waiting to be received on channel and sent
900 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
901 waits for the first task coming from a given host.. [AL]
903 GRAS new functionnalities: [MQ]
904 * Enhance the parsing macro to allow the size of multidimentional objects
905 to be given thru annotations.
906 * New example (and documentation): Matrix Multiplication a la RPC
907 (as when I was young!) and fix a bunch of bugs found on the way.
909 GRAS performance improvements: [MQ]
911 * Reduce the amount of cbps creation/destruction by making it static to
912 datadesc_send/recv() and using a (newly created) cbps_reset (based on
915 * Change libdata to a set so that we can search for stuff by ID (and thus
916 reduce the insane amount of dict lookups)
919 * Actually implement gras_datadesc_copy() so that we don't have to mimick
920 RL communication on top of SG since it's so uneffective.
921 It may also be used for inter-thread communication in RL, one day.
922 * Use gras_datadesc_copy() to exchange messages on top of SG
924 - improve message exchange performance on top of SG
925 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
926 * Don't exchange on the network the size of the used part of buffer,
927 instead, specify the possible buffer size to read().
929 - reduces the amount of read/write calls (one pair per exchange)
930 - reduces the amount of exchanged data (the size)
931 - allows to retrieve all arrived data on receiver side, if we don't need
932 it right now (subsequent read will peek the buffer)
933 - allows the receiver to proceed with the begining of the stream before
934 everything is arrived
935 - make it possible to build an iov transport (using readv/writev)
937 - take care of the data with non-stable storage (like stacked data),
939 * If possible, TCP send uses vector I/O (when writev() is here)
940 - Don't use it for receive since we send data sizes and data on the
941 same stream, so we wouldn't be able to chain large amount of chunks
942 before having to flush the stuff to read the size.
943 * Rework the transport plugin mecanism to simplify it and reduce the
944 amount of pointer dereferencement when searching for the right function
947 * I guess that now, we do almost as few system calls as possible while
948 doing as few data copy as possible.
950 To improve it further, we could try to send all the sizes first and then
951 all the data (to use iov on receiving size), but it's only a partial
952 solution: when you have 2 dimensional data, the sizes of the second
953 dimension is data of the first dimension, so you need 3 streams.
955 I'm not sure the potential performance gains justify the coding burden.
957 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
959 SimGrid (3.00) stable; urgency=high
962 * New! Give the possibility to hijack the surf parser and thus bypass
963 MSG_create_environment and MSG_launch_application. Have a look at
964 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
966 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
968 SimGrid (2.96) unstable; urgency=low
973 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
974 This deprecates the xbt_error_t mecanisms.
975 It modifies (simplifies) all XBT and GRAS API.
976 MSG API keeps unchanged (exceptions raised by XBT are catched from
977 within MSG and masked with existing error handling facilities)
980 * New! Add a FATPIPE model. [AL]
981 * New! Add a parallel task model. [AL]
982 * New! Add automatically a loopback interface (in the default
983 network model) if none was precised.
986 * Bugfix: MSG_process_resume now works with the current running process.
988 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
989 * Modification of MSG_task_get_compute_duration. Once a task has been
990 processed, the value returned by this function is now equal to 0. [AL]
991 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
992 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
993 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
994 of a m_task. For now, MSG_task_cancel only works with computation
996 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
997 of the processor (in Mflop/s) regardless of the current load on the
999 * API Change: use proper naming convention for MSG_getClock and
1000 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1002 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1003 This function changes the priority of a computation task. This priority
1004 doesn't affect the transfer rate. A priority of 2 will make a task
1005 receive two times more cpu power than the other ones. This function
1006 has been added to suit the needs of Nguyen The Loc and hasn't been that
1007 much tested yet. So if it fails, please report it and send me your code.
1009 * API Change: removed all functions and types that were marked "deprecated"
1010 since many months. Renamed MSG_global_init_args to MSG_global_init.
1012 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1014 SimGrid (2.95) unstable; urgency=low
1017 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1018 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1019 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1022 * Bugfix: Allow absolute paths for platform description files [MQ]
1023 * Bugfix: do free the variables after use. Leads to drastic performance
1025 * Implement max_duration (ie, timeouts) on resources [AL]
1028 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1030 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1031 the needed support (timeouts) [AL]
1034 * Implement measurement sockets. You can now get the bandwidth between two
1035 hosts thanks to AMOK (see below). [MQ]
1036 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1037 dynar over the network (yeah) [MQ]
1038 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1039 * simple/static token-ring example. [Alexandre Colucci and MQ]
1040 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1041 in place before [MQ]
1044 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1045 type parsing. "short" and "long" modifiers were also ignored; other
1046 modifier (such as reference level) are still ignored. [MQ]
1047 * Embeed the buffer size within the buffer itself on SG. [MQ]
1048 That way, send() are atomic and cannot get intermixed anymore (at least
1049 the ones which are less than 100k; bigger messages still have the issue)
1050 * Array size pushed by the field, not by the field type (or each
1051 and every long int will push stuff to the cbps) [MQ]
1052 * use select() to sleep since it allows to portably sleep less than one
1055 GRAS (minor cleanups)
1056 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1059 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1060 * type category 'ignored' killed as it was never used and were difficult
1062 * whether a type can cycle or not is now a flag, leaving room for more
1063 flags (as "ignored", if we feel the need one day ;)
1064 * Rename raw sockets to measurement sockets since "raw" has another
1065 meaning in networking community.
1068 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1069 and offers features not belonging to GRAS but that most applications
1070 need. One day, it may be a set of plugins loadable at runtime.
1071 * New module: bandwidth
1072 bandwidth measurement between arbitrary nodes running this module. [MQ]
1074 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1076 SimGrid (2.94) unstable; urgency=low
1078 The first beta release of SimGrid 3 !
1082 * Update the main page and the FAQ. Adding references to gforge.
1085 * Add a gras_os_getpid function.
1088 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1089 * Extend the logs so that they also print PID, hostname, date, ... if
1091 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1092 and kill the old version which were in testsuite/
1093 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1097 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1098 for users. Instead, each programming environment (GRAS, MSG,...) use it
1099 when needed to provide such a feature to users.
1100 Don't shortcut the mecanism or you will also shortcut the virtualization
1101 you need on the simulator.
1105 * Cleanups in configury with regard to compile optimization/warning flags.
1106 Also add -fno-loop-optimize to any powerpc since it's the optimization
1107 killing gcc (< 3.4.0).
1108 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1109 needed by MSG examples complications
1110 * Borrow configury beautifications from PHP
1113 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1114 warning (thanks loris for stumbling into this one).
1115 * Bugfix: stop loading private headers (gras_config.h) from the public
1119 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1120 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1121 RL would imply a DNS resolver, which is *hard* to do in a portable way
1122 (and therefore delayed).
1123 * Implement a real timer mecanism and use it in timing macros. This allows
1124 to avoid rounding errors and get a 0.000005 sec precision in timing
1125 macros. While I was at it, various cleanups:
1126 - allow to declare more than one timed section per file (fix a stupid bug)
1127 - move some private declaration to the right place
1128 - merge conditional execution and timing macros into emulation module
1129 - document the module
1130 - make sure the module cleanups its mess on gras_exit
1131 * Documentation improvements:
1132 - (new) how to compile applications using GRAS
1133 - (new) emulation support (timing macros)
1135 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1137 SimGrid (2.93) unstable; urgency=low
1139 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1142 - Use Paje properly where used. Still to be sanitized properly.
1143 - Portability fix: Add an implementation of the contexts using pthread
1147 - Add xbt_procname(): returns the name of the current process.
1148 Use it to show the current process's name in all logging.
1150 - fix detection of older flex version and the reaction, since we do
1151 depend on modern ones (we use lex_destroy)
1152 - Better separation of SG and RL in the libs: remove all simulation code
1153 from libgras. As a result, this lib is now only 200k when stripped.
1154 Some of the xbt modules may also be duplicated (two sets and such) and
1155 should be cleaned/killed before SG3.
1156 - Insist on using xlC on AIX because of weird problems involving gcc there.
1157 - Cleanup the make remote stuff. This is now done by scripts
1158 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1159 mainly for our private use, but we're working on changing them to user
1162 - Bugfix: flush the socket on close only if there is some *output*.
1163 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1165 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1168 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1170 SimGrid (2.92) unstable; urgency=low
1172 Alpha 3 on the path to SimGrid 3
1176 - New! First try of benchmarking macros.
1177 - New! First try so that gras_stub_generator generate deployment and
1178 remote compilation helpers.
1180 - Bugfix: Initialization fix in msg_test.
1184 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1186 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1188 - API Change: gras_os_sleep to take the amount of seconds as a double.
1189 Accepting an int was error prone since it was the only location where
1190 seconds were coded as such. It leaded to damn rounding errors.
1191 - Bugfix: Hard to belive that timers ever worked before this.
1193 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1195 SimGrid (2.91) unstable; urgency=low
1197 Alpha 2 on the path to SimGrid 3
1201 - Bug fix in the lmm_solver.
1203 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1204 through the function MSG_paje_output.
1205 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1206 - It is possible to bound the rate of a communication in MSG with
1207 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1208 to write it in the changelog).
1209 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1214 - Various cleanups to the autotools stuff
1215 - Begin to move Gras examples to examples/gras/
1216 - Let make distcheck work again (yeah!)
1218 - documentation overhauled using doxygen.
1219 gtk-doc-tools is dead in SimGrid now.
1220 - Automatically extract all existing logging categories, and add the list
1221 to the documentation (long standing one, to say the less)
1223 - Cleanup the known architecture table. Reorder the entries to group what
1224 should be, and use a more consistent naming scheme.
1225 (some of the test dataset are still to be regenerated)
1226 - New! Allow library to register globals on each process just as userdata
1228 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1229 lookup time (for now).
1230 Use it in msg and trp.
1231 This cleans a lot the internals and helps enforcing privacy of the
1232 headers between the gras components.
1233 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1234 - Bugfix: gras_os_time was delirious in RL.
1235 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1237 - Reenable GRAS now that it works.
1239 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1241 SimGrid (2.90) unstable; urgency=low
1243 Alpha 1 on the path to SimGrid 3
1245 * It is a long time since the last release of SimGrid. I'm sorry about
1246 that but as I had told you, I was rewriting a lot of things. I apologize
1247 to those who had been reporting bugs to me and that I had not answered.
1248 If your bug is still in the new version, please tell me. Here is a
1249 summary of the main changes.
1251 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1252 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1253 portability, speed, and a lot more but you'll figure it out later.
1254 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1255 I wanted to release the new MSG as soon as possible and I have
1256 broken GRAS, which is the reason why, for now, only MSG is fully
1257 functional. A laconic description of these projects is available
1258 in the documentation.
1260 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1261 kernel optimized for our needs (called SURF but only the developers
1262 should use it). Hence, MSG is now roughly 30 times faster and I think
1263 that by rewriting a little bit MSG, I could event speed it up a little
1264 bit more. Beside the gain in speed, it is also much easier to encode a
1265 new platform model with SURF than it was with SG. More to come...
1267 * REVOLUTION 3: I have tried to change a little as possible the API of
1268 MSG but a few things really had to disappear. The main differences
1269 with the previous version are :
1270 1) no more m_links_t and the corresponding functions. Platforms are
1271 directly read from a XML description and cannot be hard-coded
1272 anymore. The same format is used for application deployment
1273 description. The new format is described in the documentation.
1274 Have a look in tools/platform_generation. There is a tiny script
1275 that converts from the old platform format to the new one. Concerning
1276 the application deployment format, parsing the old one is tricky.
1277 I think most of you should however be able to convert your files. If
1278 it is really an issue, I can write a C code that does the conversion.
1280 2) the toolbox tbx does not exist anymore. We now have a library
1281 with much more data-structures but without the hash-tables (we have
1282 dictionaries that are much faster).
1284 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1286 *****************************************************************************
1287 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1288 * don't feel like converting the oldies. (Mt) *
1289 *****************************************************************************
1292 Version 2.90: "the long awaited one"
1293 - Finished rewriting and debugging MSG. Rewrote the documentation.
1294 - disable GRAS for now since it needs to be ported to the newest SG
1297 - Finish the port to windows (using mingw32 for cross-compile)
1300 - Main loop and datastructures of SURF. A cpu resource object is
1301 functional. Surf can thus be used to create cpu's with variable
1302 performance on which you can execute some actions.
1304 2004-11-15 Martin Quinson
1305 - Port to ARM. Simply added the alignment and size descriptions. Should
1306 work, but the ARM machines are so slow that I didn't had the opportunity
1307 to 'make check' over there yet.
1309 2004-11-15 Arnaud Legrand
1310 - Trace manager now written. It uses a heap structure and is therefore
1311 expected to be efficient. It may however be speeded up (particularly
1312 when many events occur at the same date) by using red and black
1313 trees. One day maybe...
1314 - Max-min linear system solver written. It uses a sparse matrix
1315 structure taking advantage of its expected use. Most operations are
1316 O(1) and free/calloc are called as few as possible. The computation of
1317 the minimum could however be improved by using a red and black tree
1320 2004-11-03 Arnaud Legrand
1321 - Rename every gras_* function that was in xbt/ to its xbt_
1323 - Add a heap and a doubly-linked list to xbt
1324 - Added a dichotomy to the dictionaries. make check works as well before
1325 so I assume that the patch is correct. I do not know however if things
1326 run effectively faster than before now. :)
1328 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1329 SimGrid, and the version number is bumped to 2.x
1331 2004-10-29 Martin Quinson
1332 - Introduction of the remote errors.
1333 They are the result of a RMI/RPC on the remote machine.
1334 ErrCodes being scalar values, you can't get the host on which those
1335 errors did happen. Extending the error mechanism as in Gnome is possible.
1336 No idea yet whether it is a good idea.
1338 2004-10-28 Martin Quinson
1339 - Interface revolution: the Starred Structure Eradication.
1340 I used to do typedef struct {} toto_t; and then handle *toto_t.
1341 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1342 - ???_t is a valid type (builded with typedef)
1343 - s_toto_t is a structure (access to fields with .)
1344 - s_toto is a structure needing 'struct' keyword to be used
1345 - e_toto_t is an enum
1346 - toto_t is an 'object' (struct*)
1348 typedef struct s_toto {} s_toto_t, *toto_t;
1349 typedef enum {} e_toto_t;
1350 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1351 s_toto_t) is private.
1353 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1354 so that it can set the variable to NULL. It was so for dicts and sets,
1355 it changed for dynars.
1357 - Fix a bunch of memleaks in dict_remove
1358 - Fix a bug in sg/server_socket opening: it failed all the time.
1360 2004-10-07 Martin Quinson
1361 - Speed up dynar lookup operation a bit.
1363 gras_dynar_get is dead.
1365 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1366 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1367 (which gives you the address of the stored data).
1369 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1370 copy of the data using an affectation to do the job and not a memcpy.
1372 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1373 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1375 It does not really speedup the dynar test because they are
1376 setting elements all the time (and look them seldom). But the dict does
1377 far more lookup than setting.
1379 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1381 2004-10-05 Martin Quinson
1382 - Allow to (en/dis)able the cycle detection at run time.
1384 Whether we should check for cycle or not is now a property of each
1385 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1386 datadesc_cycle_unset allow to remove this property when previously set.
1388 Note that the cycle detection is off by default since it impacts the
1389 performance. Watch the data you feed GRAS with ;)
1391 This property is hereditary. Any element embedded in a structure having it
1392 set have it set for the time of this data exchange.
1394 You should set it both on sender and receiver side. If you don't set it on
1395 sender side, it will enter an endless loop. If you forget on receiver
1396 side, the cycles won't be recreated after communication.
1398 - Header reorganization.
1399 Kill gras_private.h, each submodule must load the headers it needs.
1401 2004-10-04 Martin Quinson
1402 - Interface revolution: do not try to survive to malloc failure.
1404 Now, gras_malloc and friends call gras_abort() on failure.
1405 As a conclusion, malloc_error is not a valid error anymore, and all
1406 functions for which it was the only gras_error_t return value are
1407 changed. They now return void, or there result directly.
1408 This simplify the API a lot.
1410 2004-09-29 Martin Quinson
1411 - Re-enable raw sockets.
1412 Created by gras_socket_{client,server}_ext;
1413 Used with gras_raw_{send,recv}
1416 It should allow to kill the last bits of gras first version soon.
1418 This is not completely satisfactory yet (duplicate code with
1419 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1422 - Simplify transport plugin (internal) interface by not passing any
1423 argument to _server and _client, but embedding them in the socket
1426 2004-09-28 Martin Quinson
1427 - Finish the port to AIX.
1428 autoconf was my problem (segfault within the malloc replacement
1429 function. No idea why)
1431 2004-09-16 Martin Quinson
1432 - Fix some size_t madness on 64bit architectures.
1434 2004-09-08 Martin Quinson
1435 - Reduce the number of system headers loaded, overload some more system
1436 calls (such as malloc to cast the result of the system one, and work
1438 - Fix and reintroduce the config support
1440 2004-09-07 Martin Quinson
1441 - Source code reorganization to allow Arnaud to surf all over there.
1442 - Allow to document the logging categories.
1443 - Remove all uppercase from logging categories and useless cleanup in names.
1445 2004-08-18 Martin Quinson
1446 Version 0.6.2 (protocol not changed; API changed)
1447 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1448 gras_error_t), and NULL when not found. Functions expecting a msgtype
1449 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1450 hopefully usefull message.
1451 - Portability to prehistoric sparcs again
1453 2004-08-17 Martin Quinson
1454 Version 0.6.1 (protocol not changed; ABI not changed)
1455 - prealloc some buffers to speed things up
1457 2004-08-11 Martin Quinson
1458 Version 0.6 (protocol not changed; ABI expended)
1459 - The parsing macro can deal with the references, provided that you add
1460 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1462 2004-08-09 Martin Quinson
1463 Version 0.5 (protocol not changed; ABI changed)
1464 - Allow to off turn the cycle detection code in data exchange at
1465 compilation time. It should be at run time, but I'm short of time (and
1466 the config stuff is still broken). That way, we keep dict out of the
1467 critical path, which is good because the performance is poor:
1468 - search not dichotomial yet
1469 - dynar give no way to access their content and memcpy everytime
1470 - In composed data description (struct, ref and so on), stop foolness of
1471 keeping the subtype's ID, but store the type itself. This keeps sets out
1472 of the critical path, which is good since they rely on dynar and
1473 dictionnaries. The only loose of that is that we cannot detect the
1474 redeclaration of a structure/union with another content (but I'm not sure
1475 the code detected well this error before anyway). We still can detect
1476 the redefinition discrepancy for the other types.
1477 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1478 it breaks the code because of type-punning used all over the place).
1479 This breaks on all non-gcc architectures (for now).
1481 All those changes (plus the buffer of last time) allow me to gain 2 order
1482 of magnitude on cruel tests consisting of 800000 array of integers on two
1483 level of a hierarchical structure (200 secondes -> 4 secondes)
1486 - the selector of reference must now return the type it points to, not
1487 the ID of this type.
1489 2004-08-06 Martin Quinson
1490 Version 0.4 (protocol changed; ABI not changed)
1491 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1492 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1493 - Implement a buffer transport, and use it by default (it relies on tcp in
1494 real life and on sg in simulation).
1495 That's a bit hackish since I had a new field to the structure to store
1496 its data without interfering with the subtype ones. Inheritance
1497 is tricky in C. And that's a kind of reverse inheritance with one class
1498 derivating two classes. Or maybe a game with java interfaces. Anyway,
1499 that's damn hard in C (at least).
1500 Moreover, I got tired while trying to ensure plugin separation and
1501 genericity in SG mode. MSG wants me to do weird things, so let's go for
1502 cruel hacks (temporarily of course ;).
1503 See comment in transport_private.h:71
1504 - do not include all the _interface headers in private but in the files
1505 which really need them (to cut the compilation time when they are
1508 2004-07-26 Martin Quinson
1509 Version 0.3 (protocol not changed; ABI changed)
1510 - Major overhault of the datadesc interface to simplify it:
1511 - shorted the function names:
1512 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1513 - add a trivial way to push/pop integers into the cbps without malloc.
1514 This allows to make really generic sub_type description, which simply
1515 pop their size of the stack.
1516 - add a function gras_datadesc_ref_pop_arr() which does what users want
1517 most of the time: Declare a dynamic array (which pops its size of the
1518 stack) and declare a reference to it. Poor name, but anyway.
1519 - kill the post-send callback, add a post-receive one
1521 2004-07-23 Martin Quinson
1522 Version 0.2 (protocol changed; ABI changed)
1523 - add some testing for cpbs in the test cases, and fix some more bugs.
1524 This invalidate again the little64 data file, since I cannot regenerate
1526 - remove an awfull optimization in the logging stuff, allowing me to:
1527 - understand it again
1528 - learn gcc how to check that the argument match the provided format
1529 - fix all errors revealed by gcc after that
1530 - internal keys of dict are not \0 terminated. Deal with it properly in
1531 loggings instead of segfaulting when the user want to see the logs :-/
1533 2004-07-22 Martin Quinson
1534 - Fix some stupid bug preventing cbps (callback postit) from working
1536 2004-07-21 Martin Quinson
1537 - Some documentation cleanups
1538 - remove the useless last argument of msgtype_declare
1539 - rename the Virtu functions to fit into the 'os' namespace
1540 - move headers src/include -> src/include/gras/ and stop fooling with
1541 gras -> . symbolic link
1542 - make distcheck is now successful
1544 2004-07-19 Martin Quinson
1546 - Build shared library also
1547 - Install html doc to the right location
1548 - stop removing maintainer files in make clean
1549 - build tests only on make check
1551 2004-07-13 Martin Quinson
1553 - No major issue in previous version => change versionning schema
1554 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1555 corresponding dataset.
1557 2004-07-11 Martin Quinson
1559 - Get it working with any kind of structure (we can compute the padding
1560 bytes remotely for all the architectures I have access to)
1561 - Implement the structure parsing macro (still not quite robust/complete)
1562 - Improvement to the remote testing toysuite
1564 2004-07-10 Martin Quinson
1565 [autoconf mechanism]
1566 - get ride of a bunch of deprecated macros
1567 - actually run the test for two-compliment, not only compile it :-/
1568 - test whether the structures get packed (and bail out if yes. Damn.
1569 Alignment is a serious matter)
1570 - test whether the structures get compacted (but respecting the alignment
1571 constraints of each types)
1572 - test whether the array fields of structures can straddle alignment boundaries
1574 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1575 (took me 2 hours to find that bug, looking at the wrong place)
1576 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1577 before sending/receiving and are used to compute the offsets of fields
1578 - Given that padding size depend even on compiler options, keep track of
1579 alignment and aligned_size only for the current architecture. Not a big
1580 deal since we send structure fields one after the other (seems
1582 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1583 program, let it work on linux/gcc/little32. portability todo.
1585 2004-07-08 Martin Quinson
1586 - import and improve remote compilation support from FAST
1587 - make sure make check works on half a dozen of machines out there
1589 2004-07-07 Martin Quinson
1590 Let's say it's version 0.0.3 ;)
1591 - Implement conversions (yuhu!)
1592 - Let it work on solaris (beside conversion, of course)
1593 - Stupid me, using rand() to generate the conversion datatests in not wise.
1595 2004-07-06 Martin Quinson
1596 - Let make dist work, since I'm gonna need it to compile on remote hosts
1597 - Let Tests/datadesc_usage write the architecture on which the file was
1598 generated as first byte.
1599 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1600 architecture descriptions.
1601 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1602 on those architectures.
1603 - Optimization: send/recv array of scalar in one shoot
1605 2004-07-05 Martin Quinson
1606 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1608 - Plug a whole bunch of memleaks
1609 - each process now have to call gras_{init,exit}. One day, their log
1610 settings will be separated
1611 - Continue the code factorisation between SG, RL and common in Transport.
1613 2004-07-04 Martin Quinson
1615 - Redistribution between SG and RL.
1616 We wanna have to accept in SG, so move accepted related parts of RL in
1617 the common part. (more precisely, the dynar of all known sockets is no
1618 more a static in transport.c, but part of the process_data)
1620 [gras_stub_generator]
1621 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1623 2004-07-03 Martin Quinson
1624 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1625 them all up in one shot)
1627 - Fix the ugly case of reference to dynamic array.
1628 - New (semi-public) function gras_datadesc_size to allow the messaging
1629 layer to malloc the needed space for the buffer.
1631 - gras_socket_close now expect the socket to close (and not its address to
1632 put NULL in it after it). This is because the socket passed to handlers
1633 is one of their argument (=> not writable).
1635 - propagate the interface cleanup from last week in datadesc, ie remove a
1636 superfluous level of indirection. User pass adress of variable
1637 containing data (both when sending and receiving), and not of a variable
1638 being a pointer to the data. Let's say that I like it better ;)
1639 The price for that is constructs like "int msg=*(int*)payload" in
1640 handlers, but it's a fine price, IMHO.
1642 - Let it work in RL (yuhu)
1644 2004-06-21 Martin Quinson
1646 - porting SG plugin and SG select to new standards (works almost).
1647 - plug memleaks and fix bugs around.
1650 - cleanup the prototype of data recv and force users to specify when they
1651 want to handle references to objects. Test case working even for cycles.
1652 - plug memleaks. Valgrind is perfectly ok with this.
1654 2004-06-12 Martin Quinson
1656 - cleanup the separation between plugin and main code in plugin creation
1658 2004-06-11 Martin Quinson
1660 - Reput hook for raw sockets, needed for BW experiments
1661 - kill a few lines of dead code
1662 [Data description] Interface cleanup
1663 - gras_datadesc_by_name returns the searched type or NULL.
1664 That way, no variable is needed to use a type desc once, which makes
1666 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1667 two parameters were strings (field name, type name), leading to
1669 [Dicos] Interface cleanup
1670 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1671 This is consistant with the dynar API.
1673 2004-04-21 Martin Quinson
1675 - Porting to new standards.
1677 - interface cleanup.
1678 There is no bag anymore, no need to take extra provision to mask the
1679 pointers behind "ID".
1680 Better splitup of functions between files create/exchange/convert.
1681 This is still a bit artificial since convert and receive are so
1682 interleaved, but anyway.
1684 - add a queued message list to procdata (the ones not matching criteria
1686 - factorize some more code between SG and RL wrt procdata
1688 - use gras_exit in example to track memleaks
1689 - get rid of gs_example now that GS is properly integrated into gras
1690 - update run_test to integrate the lastest tests (datadesc)
1692 - rename WARNINGn macros to WARNn since it prooved error-prone
1694 2004-04-19 Martin Quinson
1696 - register init/exit functions within gras module mechanism
1697 - send/receive function.
1698 Convertion is not implemented, but short-cutted if not needed.
1699 struct/array elements are sent one by one (instead of block-wise), but
1700 nobody really cares (yet). Get a prototype before optimizing.
1701 - tests (using a file socket) for DD send/receive on:
1702 - base types: int, float
1703 - array: fixed size, string (ie ref to dynamic string)
1704 - structure: homogeneous, heterogeneous
1705 - chained list, graph with cycle
1706 Believe it or not, valgrind is not too unhappy with the results. The
1707 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1711 - Bugfix when using a filename explicitely (instead of '-')
1713 2004-04-09 Martin Quinson
1715 - factorize more code between RL and SG in socket creation
1716 - Complete the implementation and tests of:
1718 o file (only in RL, and mainly for debugging)
1720 I lost 3 days to design a portable address resolver, and then decided
1721 that the prototype mainly have to run on my box.
1722 Addressing portability too early may be like optimizing too early :-/
1724 - use gras_init in the Tests instead of the crappy parse_log_opt
1725 (the latter function is removed)
1726 [Conditional execution]
1727 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1728 [Code reorganisation]
1729 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1730 Build examples against the RL library, since there is no way to disable
1731 its creation for now.
1733 For information, the beginning of coding on GRAS was back in june
1734 2003. I guess that every line has been rewritten at least twice since