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 * Add a SD_task_set_name() function
41 * Fix SD_task_unschedule() on typed tasks
42 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
43 * In DAX loader, accept useless 'level' attributes to <job> since
44 LIGO DAGs have them (seem to be to ease graphical representation).
46 * Add an example masterslave_mailbox.c using send/receive and not
47 the deprecated put/get interface.
48 * Kill the MSG_paje_output() function. It's a noop since 2 years.
49 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
51 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
52 (declare MSG_USE_DEPRECATED to still have the old name)
53 * Add a MSG_task_set_data() function
54 * About trace replay (see examples/msg/actions):
56 - Allow to work with splitted trace files for each process
57 Give the specific trace file as argument of each process,
58 and call MSG_action_trace_run(NULL)
59 You can still have one merged file for all processes.
60 - Fix implementation of collective operations
62 * This is the first release of SimGrid where SMPI is not considered
63 beta anymore (even if some corners should still be improved)
64 * Port over the new SIMIX_network submodule (internal refactoring)
65 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
66 * Implement more missing elements of the standard:
68 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
70 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
71 MPI_Scan() MPI_Get_processor_name()
72 - Added implementation of missing case for Alltoall (warning: it's
73 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
74 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
75 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
78 - MPI_Waitsome() was broken
79 - Allow relative includes in smpicc
80 - Command line cfg argument 'reference_speed' was ignored...
81 - Some functions did not properly lead to auto-benching of user code
82 - smpicc passes -O2 by default (just like openmpi one)
84 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
85 * Bug fixes about timeouts during communications
86 * add SIMIX_message_sizes_output() as a pimple to write to file the
87 amount of messages per size. Use gnuplot to get histogram.
88 Pimple because that's the only user-visible function of simix,
89 defined directly in xbt.h (irk, sorry)
91 - Add a SIMIX_sem_get_capacity() function
92 - Fix interactions with processe resume/suspende
93 - release_forever() was stupidly broken
94 - Fix SIMIX_display_process_status() for processes in a semaphore
95 - Make SIMIX_sem_block_onto() user-visible
96 * Refactoring context stuff:
97 - Use pseudo-OOP for better modularity
98 - reimplement SIMIX_process_kill() without process_schedule() so
99 that the latter can take as invariant that it is called from
101 - Merge context_start into context_new for sake of simplicity
103 * Rename configuration variables to start a hierarchy:
104 o cpu_model -> cpu/model
105 o network_model -> network/model
106 o workstation_model -> workstation/model
107 * New configuration variables:
108 o network/bandwidth_factor: correction to bandwith
109 o network/latency_factor: correction to latency
110 o netwotk/weight_S: correction to the weight of competing streams
111 * Add a long description to the models, that users can see with such
112 argument on the command line: --cfg=cpu/model:help
113 * --help-models display the long description of all known models
115 * config: add the ability to set a default value after registration
116 Does not override any previously set value (e.g. from cmd line)
117 * dict: allow to have integer key and data.
118 When so, you need to use the following functions
119 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
120 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
121 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
122 In contrary to regular dicts, the key is not malloced before copy.
123 Mixing scalar and regular elements in the same dict is not tested
125 * Allow to use xbt_dynar_shrink() to expend the dynar instead
126 Tracing for Visualization:
127 * SimGrid is now instrumented in order to generate a trace file for
128 visualization analysis: to use it, need to compile SimGrid with the
129 "tracing" option enabled, and instrument the program using SimGrid with
130 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
131 (among other functions).
132 * The instrumentation only traces the platform utilization for now
133 * Documentation to use the tracing functions and how to analyze the
134 traces with the Triva tool is written.
135 * More information about: SimGrid FAQ (in the section Tracing Simulations
138 * We moved to cmake as default build system. Autotools support will
139 be dropped soon. Check the FAQ for more info about how to use it.
140 * Greatly improved our cdash/ctest interactions
141 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
142 * Added memory checking tests with valgrind; lot of memleak fixing.
143 This may be the first release of simgrid with so few memory issues
144 * Added code coverage tests.
145 Our coverage is still improvable, but at least we see it on cdash.
147 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
150 SimGrid (3.3.4) stable; urgency=low
152 The "Desktop Grid needs love too" release (also called Xmas release).
155 * Major speedup in the maxmin system solving by using lazy evaluation
156 Instead of solving completely the maxmin system at each iteration,
157 only invalidate (and recompute) the modified parts.
158 This new feature is enabled in default models but you can try to
159 turn it on with "--cfg:maxmin-selective-update=1" for other models.
160 * Cas01 IMproved as default CPU model
161 This CPU model is the same Cas01 model, but it uses the
162 maxmin-selective-update flag and a heap structure to manage
163 actions on SURF kernel.
164 It reduces the complexity to find the next action to finish and,
165 consequently, it's faster than the old Cas01.
166 This is the new default CPU model (Cas01).
167 * Rename the old Cas01 model to Cas01_fullupdate
168 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
169 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
170 * CpuTI (CPU Trace Integration)
171 A new CPU model whose objective is simulate faster when using
172 availability trace files.
173 Instead of using a full featured, over engineered maxmin system for
174 CPU modeling, this model does the pre-integration of traces files
175 to calculate the amount of CPU power available, and so, executes
176 faster than the old CPU models.
177 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
178 * Use LV08 as default network model since it gives better accuracy
179 for small messages and shouldn't change things for big ones.
180 Use --cfg=network_model:CM02 to get the previous behavior.
183 ******************************************
184 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
185 ******************************************
186 * The new CPU model may changes simulations!
187 The point is that events occurring at the exact same timestamp
188 are not scheduled in the same order with the old and new
189 version. This may be enough to completely change the execution
190 of simulations in some cases.
191 * The new network model will change simulations!
192 This new model is more realistic than the previous one, so you
193 should consider redoing your old experiments with this model.
194 Sorry for the inconvenience.
197 * Introduce the supernovae compilation mode
198 When compiled that way, the whole SimGrid (or almost) is put in a
199 single compilation unit and compiled in one shoot.
200 This is to help gcc which has difficulties to inline stuff from one
202 The speedup seem to be above 15%, althrough more tests are needed on
203 amd64 to confirm that gain.
206 * Port of MSG's mailbox on top of SIMIX network
207 The put/get mechanism was greatly simplified on the way.
210 * New SIMIX network module. Provides:
211 - Mailbox: rendez-vous mecanism to find with who you want to speak
212 - Synchronous send/recv: easier and hopefully faster since the
213 logic is handled in the maestro process directly now
214 - Asynchronous send/recv: you dreamt of it? It's here now
215 Too bad that nobody cared enough to propagate the change to MSG.
216 * Add semaphores as SIMIX synchronization mechanism.
219 * new function SD_daxload(char*) to load a DAX file
220 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
221 * Introduce typed tasks. Specify its kind and cost at creation.
222 At scheduling, just give where it should be placed, and the cost
223 for each involved resource is automatically computed.
224 Existing constructors so far (more to come of course):
225 - SD_task_create_comm_e2e() for end-to-end communication
226 - SD_task_create_comp_seq() for sequential computation
227 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
228 * new function SD_task_dump() for debuging display
229 * new function SD_task_dotty(task,FILE*) writing to file the info
230 about the task in dotty format
231 * SD_task_dependency_exists() can now cope with having one of its
232 arguments NULL. If so, it tests whether the other argument has any
234 * Add getters on list of preceding/following tasks:
235 SD_task_get_parents(task) and SD_task_get_children(task)
236 * Add getters on amount of workstations and list:
237 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
238 * Add getter on task kind: SD_task_get_kind(task)
239 * Update the start_time and finish_time of tasks on completion/failure
240 * Bugfix: Remove task from state swags when destroyed
243 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
246 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
247 dynar with one separate thread per value of the dynar.
248 * Change the prototype of xbt_thread_create(), sorry.
249 Added a boolean parameter indicating whether we want to join this
250 thread (used in SG only for now)
251 * Implement xbt_thread_join and xbt_thread_yield in SG also.
254 * GTNetS wrappers should now be usable again (and betterly tested too)
255 * Fix a major regression from 3.2 where the timeout provided to
256 MSG_task_put_with_timeout() was used as absolute time before which
257 the comm should be done.
258 * Start to fix the <cluster> tag.
259 - Internal links should be good now (beside of the loopback, which
260 use the private link instead)
261 - paths to the external world is still rather broken
262 - the <route:multi> tag is just broken. Actually that's brain-dead.
263 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
264 to make it less stupid
265 ** Check your platform with teshsuite/simdag/platforms/flatifier **
266 * Fix a source-level compatibility glitch from 3.2: after defining
267 MSG_USE_DEPRECATED, you can use the old name
268 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
269 * Allow to compile from the SVN with automake 1.11
270 * Fix some problems when using the "start_time" tag in deployment XMLs.
271 * Fix #8569: XBT/synchro.h has redundant declarations
272 * Fix #8563: MSG return values and exceptions
273 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
274 * Integrate patch #8636: Obey DESTDIR when installing documentation.
275 Thanks to Robson Peixoto.
276 * Fix a vicious bug in dictionaries inducing that some elements were
277 not freed on xbt_dict_free()
279 Portability report of this version:
280 * Main portability targets:
281 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
283 - mac leopard on i386
284 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
285 but nothing critical.
286 * Other platforms: windows, AIX and others were not tested for this release
288 Timing report of this version:
289 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
290 scenario parameters). From 8h to a few seconds in desktop grid settings.
291 * Supernovae brings about 25% speedup on i386.
293 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
295 SimGrid (3.3.3) stable; urgency=low
297 The "Need for Speed" release.
299 The timings done to validate the 3.3.2 were faulty.
300 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
302 The problem was a conversion from a manually handled vector to
303 xbt_dynar_t on the critical path.
304 xbt_dynar_foreach calls functions, inducing stack management crap.
306 We inlined these functions and xbt_dynar_foreach is now breath taking.
307 We also inlined xbt_swag_belong on the way.
309 Here are some approximate speedup measurements (on master/slaves
310 simulations lasting between 10s and 20s each):
311 3.3.1 -> 3.3.2: about same performance
312 3.3.2 -> 3.3.3: 40% speedup
313 3.3.1 -> 3.3.3: 40% speedup
314 3.3.1 with inline patch -> 3.3.3: 30% speedup
316 Our reading is that the refactoring which occurred in 3.3.2 made us
317 suffer much more from the xbt_dynar_foreach low performance, but
318 once we solved this, this refactoring proved to be very performance
319 effective. From the 40% speedup, somehow, 10% are due to the
320 inlining and 30% to the refactoring.
322 That's a pitty that gcc cannot inline functions placed in other files
323 alone. We have to choose between:
324 - break the encapsulation (by putting private data structures and
325 accessors in headers files to help gcc)
326 - live with low performance
327 - switch to a decent compiler such as icc (not quite possible).
329 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
331 SimGrid (3.3.2) stable; urgency=low
333 The "Simplicity does not preceed complexity, but follows it" release.
335 The main contributors of this release were (lexical order):
336 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
339 * Extract the routing logic into its own object.
340 (was dupplicated in network.c and workstation_LV07.c;
341 Allows to implement other ways of storing that info)
342 => kill now useless network_card concept
343 - Use dynar to represent routes (instead of void** + int*)
344 - kill link_set (use surf_network_model->resource_set instead)
345 - Add a command-line option to choose the routing schema to use
346 - Add three new models:
347 * Floyd (shortest path computed at initialization)
348 * Dijikstra (shortest path recomputed all the time)
349 * Cached Dijikstra (shortest path computed on need)
350 All these models where contributed by Silas De Munck, and are
351 described in his ICCS09 paper.
353 * Simplify model declaration
354 (less redirections, less function to write when defining a model)
355 - Factorize stuff between models:
358 surf_model_resource_set(model)
359 surf_model_resource_by_name(model, name)
360 - Unify the types of models in s_surf_model_t (using an union)
361 - Embeed fields of common_public directly into s_surf_model_t
362 - Rename model methods:
363 action_free ~> action_unref
364 action_change_state ~> action_state_set
365 action_get_state ~> action_state_get
366 - Change model methods into functions :
367 (model)->common_public->action_use ~> surf_action_ref
369 * Implement a generic resource; use it as ancestor to specific ones
370 (allows to kill duplicated code in models)
371 Drawback: timer command don't need no name nor properties;
372 workstation_CLM03 don't need no properties
373 (but I guess we can live with those few bytes wasted)
375 * Improve the action object model
376 - implement a constructor avoiding dupplicated code about field
377 initialization in generic_action part.
379 * Kill the SDP model: it has an external dependency, is deprecated
380 in flavor of modern lmm models, and didn't compile since a while
383 * Relocation of the context module from XBT to SIMIX.
384 (the context were decoupled from the simix processes, duplicating a lot of code)
385 => a lot of code was factorized
386 - less overhead is introduced during scheduling
387 - simpler API for the context factory
388 - the logic for process creation,destruction and manipulation was simplified
389 * Simplification of the s_smx_process_t data structure.
390 => accesing the simix level data associated to a process is faster now,
391 and the code is a lot more readable.
394 * Implement some more MPI primitives:
395 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
396 -implementation: Bcast: flat or 2-ary tree (default),
399 Allreduce: Reduce then Bcast
400 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
401 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
402 Alltoallv: flat tree, like ompi
404 * Add support for optimized collectives (Bcast is now binomial by default)
405 * Port smpirun and smpicc to OS X
408 * Kill SD_link_get_properties: hard to maintain and makes very little sense
409 Shout out if you used it.
412 * Display the list of still queued messages in SG mode when existing
416 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
417 * Add xbt_graph_node_get_outedges() [Silas De Munck]
418 * Add xbt_str_from_file(FILE*)
419 * Add xbt_dict_get_key achieving a linear reverse search
420 * Remove the context module
422 Portability report of this version:
423 * Main portability targets:
424 - Linux(debian)/x86/context
425 - Linux(debian)/x86/pthreads
426 - Linux(debian)/amd64/context
427 - Linux(debian)/amd64/pthreads
428 On these, we still have the eratic breakages of gras/pmm and
429 amok/saturate_sg reported in previous version. We still think
430 that the tests are the cause of the fault, not the tested code.
432 - Mac OSX Leopard/x86/context
433 Still false negative in tesh autotesting.
434 Smpi still fails, but this time because readlink does not accept -f
435 Everything seems to work properly beside of that.
438 - AIX version 5.3 (only tested contexts this time)
439 Smpi still fails there because mktemp is not installed.
440 Everything seems to work properly beside of that.
442 I managed to compile it for the first time, but several breakages.
443 Won't delay the release for this exotic platform.
445 * Windows: it's still lagging behind. If you want to help, please
448 Timing report of this version:
449 This version seem to be more than 5% faster than 3.3.1 (on linux
450 64bits with contextes). The gain is less than expected, we are
451 investigating this for next release.
453 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
455 SimGrid (3.3.1) stable; urgency=low
458 * Implement a --cfg-help to show existing configuration variables
459 * Build chain do not require doxygen in maintainer mode
462 * fix a bug on struct sizeof computation, which prevented the
463 exchange of arrays of structs in some conditions
464 - added a regression test about this in datadesc_usage
465 * Allow the exchange of 0-long dynamic vectors.
466 - for that, use -1 as indicator of dynamic size instead of 0
467 - This implied to change any size from unsigned long to long,
468 reducing a bit communication abilities, but I guess that with
469 64bits being quite common, this is more than enough.
470 - This also induce a protocol change, thus bumping network protocol
471 version from 0 to 1 (if we have external users, we have to get
472 clean on that point too ;)
473 - added two regression tests about this in datadesc_usage
474 * Be more verbose when propagating local exceptions
475 This helps debugging.
476 * Display the status of simulated processes when receiving SIGINT in
480 * Allow to control the simulation from a trace file.
481 New functions MSG_action_register() and MSG_action_trace_run()
482 The first one allows to associate a function execution to each
483 kind of action while the second one parses a trace file and
484 triggers the corresponding actions within the system.
485 For now, only a toy example is provided in examples/msg/actions
486 * Add an exemple of process migration in examples/msg/migration
487 * Fix a bug in task exchange which broke MSG_task_get_sender()
488 Add a teshsuite regression test for that.
489 [Bug: if MSG_task_get_sender() is called after sender exit,
491 * Fix a bug which prevented suspend/resume to work properly
492 * Display the status of simulated processes when receiving SIGINT
493 This fixes a regression of v3.3. due to the introduction of SIMIX
494 * Bug fixing in failure management:
495 - trace could not start by a failure at time 0
496 - failure during communications were not working
499 * Add SIMIX_process_set_name() to change the name of the current
500 process in the log messages.
501 * Store smx_hosts in a dict since we only retrieve them by name
502 * Move the configuration infrastructure to surf
505 * Move the configuration infrastructure to surf
508 * Massive internal cleanups:
509 - Store internal structures on processes instead of hosts (allows
510 to have more than one process per host, in addition of being more
512 - Cleanup the initialization/finalization process
513 - Kill a whole bunch of unneeded synchronization:
514 processes run in exclusive manner within the simulator
515 - Move queues from global tables to process data fields
517 - now accept -platform and -hostfile arguments
518 - Pass the right rank value to processes according to the hostfile
519 * Compile the examples by default, and use them as regression tests
520 * Implement MPI_Wtime()
521 * Change the reference speed to a command line option
524 * TCP_gamma can now be specified as command line option using
525 --cfg=TCP_gamma:10000000.0
526 * Change the --surf-path cmd line option into --cfg=path:
529 * Also include strbuff from xbt.h public header
530 * xbt_ex_display(): do not free the exception after displaying
531 This allows to do more with the given exception afterward.
532 Users should call xbt_ex_free() themselves.
536 Portability report of this version:
537 * Main portability targets:
538 - Linux(debian)/x86/context
539 - Linux(debian)/x86/pthreads
540 - Linux(debian)/amd64/context
541 - Linux(debian)/amd64/pthreads
542 These targets fail about 1/10 of times on gras/pmm, but we believe
543 that this is because of the test, not because of simgrid.
544 amok/saturate_sg fails even more rarely, and the test may not be
547 - Mac OSX Leopard/x86/context
548 The test suite still spits tons of errors because some obscure
549 force prevents us from removing the temporary directories
550 arguing that they still contain some metadata I've never heard of.
551 Smpi fails because seq is not installed.
552 Everything seems to work properly beside of that.
555 - AIX version 5.3 (both contexts and pthread)
556 Smpi still fails there because mktemp is not installed.
557 XML inclusions seems rosty on AIX.
559 * Windows: it's still lagging behind. If you want to help, please
562 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
564 SimGrid (3.3) stable; urgency=high
568 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
571 * Introduce the SIMIX module: factorize code between MSG and GRAS.
574 Until now, GRAS were using MSG as an interface to SURF. It was
575 quite difficult because both interface have several differences
576 (MSG channels vs GRAS sockets were the most notable point).
578 This also opens the gate to SMPI (which should occur soon) and speed
579 up simulations by to 40% (even if it were not the main goal).
581 **************************************
582 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
583 **************************************
584 The point is that events occuring at the exact same timestamp are
585 not scheduled in the same order with the old and new version. This
586 may be enough to completely change the execution of simulations in
587 some cases. Sorry for the inconvenience.
589 * Cleanup and upgrade the XML format to push further scalability
590 issues (check http://hal.inria.fr/inria-00256883/ for more info)
592 * Improve the testing infrastructure with tesh. Now a very large part of
593 the code is tested not only by being run but also by checking that the
594 output match an expected output [Mt].
596 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
597 is really less memory-demanding, which should allow you to use
598 larger files in SimGrid [AL].
600 * Inform valgrind about our contextes, so that it becomes usable
601 with the default (and more effecient) version of SimGrid
602 [contributed by Sékou Diakite, many thanks]
605 * Introduce a listener thread in charge of receiving incomming
606 messages from the network. It allows to overlap communication and
607 computation but most notably, it removes some stupid deadlocks due
608 to the fact that so far, a process could not send and receive at
609 the same time. This made most non trivial communication schema
611 * Convert the PIDs from long int to int to match the MSG ones (and
613 * New function: gras_agent_spawn() to launch a new process on
614 current host. Only working in simulation for now. [Mt]
615 * New function: gras_os_hostport() returning a constant form (ie,
616 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
619 * Make the backtrace of exceptions more human readable [Mt]
620 * New module: xbt/str [Mt]
621 a ton of string utility functions (split, join, printf to a newly
622 allocated buffer, trim, etc)
623 * New module: xbt/hash [Mt]
624 SHA1 hashing algorithm (more to come if needed)
625 * New module: xbt/synchro [Mt]
626 synchronization tools (mutex and conditions) working the same way
627 in simulation and in real life (mainly useful for GRAS, but not
629 * New module: xbt/queue [Mt]
630 classical producer/consumer synchronization scheme
631 * xbt_dynar_new_sync() creates a synchronized dynar. All access
632 (using the classical functions will get serialized) [Mt]
633 * Make dictionary internal table dynamic. No need to specify its size
634 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
636 * Make sure the log channels are organized as a tree under windows
637 (because of ANSI C compatibility issue, any channel were child of
641 * Cleaned many thing in surf and fixed a few bugs [AL].
642 * Add a nice command line configuration mechanism to compose models [AL].
643 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
644 the previous one (KCCFLN05). It relies on something that looks like
645 a max-min sharing mechanism but cannot be written as such. A new solver
646 was thus designed [AL].
647 * Add a new solver to lmm. Based on Lagrange optimization and
648 gradient-based descent, it enables to efficiently maximise systems s.a
650 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
652 This solver enables to propose two new network models for TCP Reno and
653 TCP Vegas based on Low's work. These models still need to be fully
654 tested though [Pedro Velho].
657 * Bug fix in SD_simulate. Now the time bound given as argument is
659 * Use the new parallel task model (ptask_L07) as default.
660 * Use the SURF command line configuration mechanism.
661 * 0-size tasks (for synchronization) should now work.
663 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
665 SimGrid (3.2) stable; urgency=high
669 We still experience issues on this platform, but we believe that at
672 GRAS API BREAKAGE (for simplification purpose, sorry):
673 * the gras_msgtype_by_name is not used anymore. Instead of
674 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
675 you can write (and must)
676 gras_msg_send(toserver, "request", &request);
677 - If you still want to pass a gras_msgtype_t to the function (to cache
678 the type and avoid the lookup time), use the gras_msg_send_() variant.
679 - Impacted functions:
680 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
681 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
682 * The callbacks are now expected to return 0 when everything went well
683 (just like the main() function)
685 GRAS new features and improvements:
686 * New module mecanism where user code can use per process globals [Mt]
687 This is similar to gras_userdata_*() functions, but for libraries. It
688 factorize some code developped over and over in the examples and AMOK.
689 It has still to be documented and used (only amok/peermanagement is
691 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
692 loss when they were small enough to fit into the buffer and sent quickly
693 enough so that they can all get received in one shoot.
694 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
695 if not found. Use the *_or_null() variant for the old semantic.
696 * In gras_msg_handle, do not discard messages without callback.
697 They are probably messages to be explicitly awaited later (ie, proofs of
698 mis-synchronization in userland since they are sent before being awaited)
700 * gras_socket_meas_send/recv: semantic changed!
701 The numerical arguments used to be (1) the total amount of data to send
702 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
703 messages. This was need for the fool willing to send more than MAXINT
704 bytes on quite fat pipes.
707 * Do really rename the hostmanagement module to peermanagement. [Mt]
708 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
709 but this is rather new and this was documented in the module
710 documentation (poor excuses, I admit)
711 * Bandwidth measurement semantic changed! This follows the changes to
712 gras_socket_meas_send/recv explained above.
715 * A sequential mode has been added to the workstations. When a workstation
716 is in sequential mode, it can execute only one task, and the other tasks
717 are waiting in a FIFO. [Christophe Thiery]
720 * The KCCFLN05 workstation model now handles parallel tasks. It is the
721 model for SIMDAG. [Christophe Thiery]
722 * Bug fix in the maxmin solver: Some values were close to 0 instead of
723 equal to 0, which caused some bad behaviors in
724 saturated_constraint_set_update. I now use a threshold mechanism like in
728 * When running manually src/testall, you select specific units [Mt]
729 testall is the result of our cunit mecanism, and should replace all
730 the scripty thingy around since bash don't run easily on billware.
732 * A mallocator system has been added. [Christophe Thiery]
733 Mallocators allow you to recycle your unused objects instead of freeing them
734 and allocating new ones.
736 Documentation update:
737 * FAQ reworking + New FAQs:
738 - "Valgrind spits tons of errors!" [Mt]
739 - "How to repport bugs" [Mt]
740 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
741 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
742 - Communication time measurement within MSG [AL]
743 - I experience weird communication times when I change the latency [AL]
746 - an introduction to the framework and to the used communication model
747 - an initiatic tour introducing the most proheminent features:
749 . Lesson 0: Installing GRAS
750 . Lesson 1: Setting up your own project
751 o Part 2: Message passing
752 . Lesson 2: Exchanging simple messages
753 . Lesson 3: Passing arguments to the processes (in SG)
754 . Lesson 4: Attaching callbacks to messages
755 . Lesson 5: Using globals in processes
756 . Lesson 6: Logging informations properly
757 . Lesson 7: Using internal timers
758 . Lesson 8: Handling errors through exceptions
759 . Lesson 9: Exchanging simple data
760 . Lesson 10: Remote Procedure Calling (RPC)
761 . Lesson 11: Explicitely waiting for messages
762 . Recapping of message passing features in GRAS
763 - A HOWTO section containing:
764 o HOWTO design a GRAS application
765 More are due, of course. They will come latter. In the meanwhile, you can
766 check the examples which are still here.
768 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
770 SimGrid (3.1) stable; urgency=high
774 There was a stack corruption somewhere, visible only when optimizing
775 with these versions. [Vince]
778 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
779 module that enabled to play with Directed Acyclic Graphs. It is built
780 directly on top of SURF and provides an API rather close to the old
781 SG. Some old codes using SG are currently under rewrite to check that
782 all needful functions are provided. [Christophe Thiery]
785 * Complete rewrite of the KCCFLN05 workstation model. It is now an
786 extension of the classical CLM03 model that gracefully handles
787 failures. This is now the default model for MSG and GRAS. It doesn't
788 handle parallel tasks yet though. [AL]
789 * Bug fix: Weights were not correctly set in the network part.
790 WARNING: This may have resulted in incorrect results with simulations
791 where there are more than one flow on a given link. [AL]
794 * After a (long ?) discussion on simgrid-devel, we have decided that the
795 convention we had on units was stupid. That is why it has been decided
796 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
797 WARNING : This means that all previous platform files will not work as
798 such with this version! A warning is issued to ask users to update
800 A conversion script can be found in the contrib module of the CVS, under
801 the name contrib/platform_generation/surfxml_update.pl [MQ]
804 * Bug fix: Processes were started in reverse order, wrt deployment file.
805 WARNING: if your code relies on this bug, please fix it. [AL]
806 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
807 being executed on two different locations. [AL]
808 * Bug fix: Failures are now better supported thanks to Derrick's tests
809 (there was many failure situations I hadn't thought of and that weren't
810 correctly handled). [AL]
811 * New function: MSG_host_is_avail indicates you whether a given m_host_t
815 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
816 Exception occurring on server-side are propagated back to client (!).
818 API CHANGE: the callback changed their prototype. Change:
819 int my_handler(gras_socket_t expeditor, void *payload_data) {
821 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
822 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
824 * New! function: gras_msg_handleall to deal with all messages arriving
825 within a given period.
826 * New! function: gras_socket_server_range to get a server socket in a
827 range of port numbers (ease to avoid port number conflicts) [MQ]
828 * New! gras processes display their backtrace when they get a SIGUSR1
829 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
830 Sweet to debug RL processes [MQ]
834 - Do not force experiment sizes to be expressed in kb, or it becomes
835 impossible to measure the latency this way (needs one byte-long tests)
836 WARNING: this changes the amok_bw_* function semantic. [MQ]
837 - Implements the link saturation stuff. [MQ]
838 * Peer management module:
839 New! module factorizing code that we wrote over and over [MQ].
842 * New module: cunit (my jUnit implementation in ansi C) [MQ]
843 - Test units are placed directly into the library code, they get extracted
844 automatically and placed into the src/testall binary.
845 - Convert most of the XBT tests to this system.
846 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
847 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
848 live in the same C code. This required to change a little bit the API
849 of surfxml parsing but shouldn't be an issue for end-users. [AL]
850 * New module: sparse graph structure with basic algorithms (this is work
851 in progress and the API is not considered to be frozen yet). [AL]
852 * Display more information on backtraces: source line & function names are
853 now displayed just like valgrind does (rely on addr2line tool) [MQ]
854 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
855 * Reworked a little bit some #include statements to load only required
856 headers. Some user code that relied on SimGrid to include stdlib or
857 stdio may need to include it by themselves. [AL]
858 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
860 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
861 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
862 but I guess that almost nobody used those parts. [MQ]
864 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
866 SimGrid (3.0.1) stable; urgency=low
869 * Unfortunately, I had missed 5 misnamed functions:
870 xbt_fifo_item_t xbt_fifo_newitem(void);
871 void xbt_fifo_freeitem(xbt_fifo_item_t);
872 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
873 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
874 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
875 They're now deprecated. Please use their new versions:
876 xbt_fifo_item_t xbt_fifo_new_item(void);
877 void xbt_fifo_free_item(xbt_fifo_item_t);
878 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
879 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
880 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
882 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
883 * Documentation: xbt_log module unmercifully reworked [MQ]
884 * Bugfix: there was a problem with the ending of contexts with
885 the pthread backend. It caused some weird deadlock or behavior
886 depending on the pthread implementation. [AL]
887 * Bugfix: get the exceptions raised in the simulator repport where
888 and why they come from when they are not catched in time [AL, MQ]
891 * Bugfix: Do repport the error when two non-connected hosts try to
892 exchange data (Thanks to Flavien for stumbling into this one) [AL]
895 * Add additionnal checkings on communications. Assert that two
896 communicating hosts are connected by a set of links... [AL]
899 * Add additionnal checkings on channel values in communication [AL]
900 * New: MSG_task_get_source to see on which host a task was generated [HC]
901 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
902 the number of tasks waiting to be received on channel and sent
904 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
905 waits for the first task coming from a given host.. [AL]
907 GRAS new functionnalities: [MQ]
908 * Enhance the parsing macro to allow the size of multidimentional objects
909 to be given thru annotations.
910 * New example (and documentation): Matrix Multiplication a la RPC
911 (as when I was young!) and fix a bunch of bugs found on the way.
913 GRAS performance improvements: [MQ]
915 * Reduce the amount of cbps creation/destruction by making it static to
916 datadesc_send/recv() and using a (newly created) cbps_reset (based on
919 * Change libdata to a set so that we can search for stuff by ID (and thus
920 reduce the insane amount of dict lookups)
923 * Actually implement gras_datadesc_copy() so that we don't have to mimick
924 RL communication on top of SG since it's so uneffective.
925 It may also be used for inter-thread communication in RL, one day.
926 * Use gras_datadesc_copy() to exchange messages on top of SG
928 - improve message exchange performance on top of SG
929 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
930 * Don't exchange on the network the size of the used part of buffer,
931 instead, specify the possible buffer size to read().
933 - reduces the amount of read/write calls (one pair per exchange)
934 - reduces the amount of exchanged data (the size)
935 - allows to retrieve all arrived data on receiver side, if we don't need
936 it right now (subsequent read will peek the buffer)
937 - allows the receiver to proceed with the begining of the stream before
938 everything is arrived
939 - make it possible to build an iov transport (using readv/writev)
941 - take care of the data with non-stable storage (like stacked data),
943 * If possible, TCP send uses vector I/O (when writev() is here)
944 - Don't use it for receive since we send data sizes and data on the
945 same stream, so we wouldn't be able to chain large amount of chunks
946 before having to flush the stuff to read the size.
947 * Rework the transport plugin mecanism to simplify it and reduce the
948 amount of pointer dereferencement when searching for the right function
951 * I guess that now, we do almost as few system calls as possible while
952 doing as few data copy as possible.
954 To improve it further, we could try to send all the sizes first and then
955 all the data (to use iov on receiving size), but it's only a partial
956 solution: when you have 2 dimensional data, the sizes of the second
957 dimension is data of the first dimension, so you need 3 streams.
959 I'm not sure the potential performance gains justify the coding burden.
961 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
963 SimGrid (3.00) stable; urgency=high
966 * New! Give the possibility to hijack the surf parser and thus bypass
967 MSG_create_environment and MSG_launch_application. Have a look at
968 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
970 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
972 SimGrid (2.96) unstable; urgency=low
977 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
978 This deprecates the xbt_error_t mecanisms.
979 It modifies (simplifies) all XBT and GRAS API.
980 MSG API keeps unchanged (exceptions raised by XBT are catched from
981 within MSG and masked with existing error handling facilities)
984 * New! Add a FATPIPE model. [AL]
985 * New! Add a parallel task model. [AL]
986 * New! Add automatically a loopback interface (in the default
987 network model) if none was precised.
990 * Bugfix: MSG_process_resume now works with the current running process.
992 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
993 * Modification of MSG_task_get_compute_duration. Once a task has been
994 processed, the value returned by this function is now equal to 0. [AL]
995 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
996 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
997 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
998 of a m_task. For now, MSG_task_cancel only works with computation
1000 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1001 of the processor (in Mflop/s) regardless of the current load on the
1003 * API Change: use proper naming convention for MSG_getClock and
1004 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1006 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1007 This function changes the priority of a computation task. This priority
1008 doesn't affect the transfer rate. A priority of 2 will make a task
1009 receive two times more cpu power than the other ones. This function
1010 has been added to suit the needs of Nguyen The Loc and hasn't been that
1011 much tested yet. So if it fails, please report it and send me your code.
1013 * API Change: removed all functions and types that were marked "deprecated"
1014 since many months. Renamed MSG_global_init_args to MSG_global_init.
1016 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1018 SimGrid (2.95) unstable; urgency=low
1021 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1022 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1023 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1026 * Bugfix: Allow absolute paths for platform description files [MQ]
1027 * Bugfix: do free the variables after use. Leads to drastic performance
1029 * Implement max_duration (ie, timeouts) on resources [AL]
1032 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1034 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1035 the needed support (timeouts) [AL]
1038 * Implement measurement sockets. You can now get the bandwidth between two
1039 hosts thanks to AMOK (see below). [MQ]
1040 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1041 dynar over the network (yeah) [MQ]
1042 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1043 * simple/static token-ring example. [Alexandre Colucci and MQ]
1044 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1045 in place before [MQ]
1048 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1049 type parsing. "short" and "long" modifiers were also ignored; other
1050 modifier (such as reference level) are still ignored. [MQ]
1051 * Embeed the buffer size within the buffer itself on SG. [MQ]
1052 That way, send() are atomic and cannot get intermixed anymore (at least
1053 the ones which are less than 100k; bigger messages still have the issue)
1054 * Array size pushed by the field, not by the field type (or each
1055 and every long int will push stuff to the cbps) [MQ]
1056 * use select() to sleep since it allows to portably sleep less than one
1059 GRAS (minor cleanups)
1060 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1063 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1064 * type category 'ignored' killed as it was never used and were difficult
1066 * whether a type can cycle or not is now a flag, leaving room for more
1067 flags (as "ignored", if we feel the need one day ;)
1068 * Rename raw sockets to measurement sockets since "raw" has another
1069 meaning in networking community.
1072 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1073 and offers features not belonging to GRAS but that most applications
1074 need. One day, it may be a set of plugins loadable at runtime.
1075 * New module: bandwidth
1076 bandwidth measurement between arbitrary nodes running this module. [MQ]
1078 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1080 SimGrid (2.94) unstable; urgency=low
1082 The first beta release of SimGrid 3 !
1086 * Update the main page and the FAQ. Adding references to gforge.
1089 * Add a gras_os_getpid function.
1092 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1093 * Extend the logs so that they also print PID, hostname, date, ... if
1095 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1096 and kill the old version which were in testsuite/
1097 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1101 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1102 for users. Instead, each programming environment (GRAS, MSG,...) use it
1103 when needed to provide such a feature to users.
1104 Don't shortcut the mecanism or you will also shortcut the virtualization
1105 you need on the simulator.
1109 * Cleanups in configury with regard to compile optimization/warning flags.
1110 Also add -fno-loop-optimize to any powerpc since it's the optimization
1111 killing gcc (< 3.4.0).
1112 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1113 needed by MSG examples complications
1114 * Borrow configury beautifications from PHP
1117 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1118 warning (thanks loris for stumbling into this one).
1119 * Bugfix: stop loading private headers (gras_config.h) from the public
1123 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1124 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1125 RL would imply a DNS resolver, which is *hard* to do in a portable way
1126 (and therefore delayed).
1127 * Implement a real timer mecanism and use it in timing macros. This allows
1128 to avoid rounding errors and get a 0.000005 sec precision in timing
1129 macros. While I was at it, various cleanups:
1130 - allow to declare more than one timed section per file (fix a stupid bug)
1131 - move some private declaration to the right place
1132 - merge conditional execution and timing macros into emulation module
1133 - document the module
1134 - make sure the module cleanups its mess on gras_exit
1135 * Documentation improvements:
1136 - (new) how to compile applications using GRAS
1137 - (new) emulation support (timing macros)
1139 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1141 SimGrid (2.93) unstable; urgency=low
1143 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1146 - Use Paje properly where used. Still to be sanitized properly.
1147 - Portability fix: Add an implementation of the contexts using pthread
1151 - Add xbt_procname(): returns the name of the current process.
1152 Use it to show the current process's name in all logging.
1154 - fix detection of older flex version and the reaction, since we do
1155 depend on modern ones (we use lex_destroy)
1156 - Better separation of SG and RL in the libs: remove all simulation code
1157 from libgras. As a result, this lib is now only 200k when stripped.
1158 Some of the xbt modules may also be duplicated (two sets and such) and
1159 should be cleaned/killed before SG3.
1160 - Insist on using xlC on AIX because of weird problems involving gcc there.
1161 - Cleanup the make remote stuff. This is now done by scripts
1162 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1163 mainly for our private use, but we're working on changing them to user
1166 - Bugfix: flush the socket on close only if there is some *output*.
1167 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1169 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1172 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1174 SimGrid (2.92) unstable; urgency=low
1176 Alpha 3 on the path to SimGrid 3
1180 - New! First try of benchmarking macros.
1181 - New! First try so that gras_stub_generator generate deployment and
1182 remote compilation helpers.
1184 - Bugfix: Initialization fix in msg_test.
1188 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1190 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1192 - API Change: gras_os_sleep to take the amount of seconds as a double.
1193 Accepting an int was error prone since it was the only location where
1194 seconds were coded as such. It leaded to damn rounding errors.
1195 - Bugfix: Hard to belive that timers ever worked before this.
1197 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1199 SimGrid (2.91) unstable; urgency=low
1201 Alpha 2 on the path to SimGrid 3
1205 - Bug fix in the lmm_solver.
1207 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1208 through the function MSG_paje_output.
1209 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1210 - It is possible to bound the rate of a communication in MSG with
1211 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1212 to write it in the changelog).
1213 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1218 - Various cleanups to the autotools stuff
1219 - Begin to move Gras examples to examples/gras/
1220 - Let make distcheck work again (yeah!)
1222 - documentation overhauled using doxygen.
1223 gtk-doc-tools is dead in SimGrid now.
1224 - Automatically extract all existing logging categories, and add the list
1225 to the documentation (long standing one, to say the less)
1227 - Cleanup the known architecture table. Reorder the entries to group what
1228 should be, and use a more consistent naming scheme.
1229 (some of the test dataset are still to be regenerated)
1230 - New! Allow library to register globals on each process just as userdata
1232 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1233 lookup time (for now).
1234 Use it in msg and trp.
1235 This cleans a lot the internals and helps enforcing privacy of the
1236 headers between the gras components.
1237 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1238 - Bugfix: gras_os_time was delirious in RL.
1239 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1241 - Reenable GRAS now that it works.
1243 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1245 SimGrid (2.90) unstable; urgency=low
1247 Alpha 1 on the path to SimGrid 3
1249 * It is a long time since the last release of SimGrid. I'm sorry about
1250 that but as I had told you, I was rewriting a lot of things. I apologize
1251 to those who had been reporting bugs to me and that I had not answered.
1252 If your bug is still in the new version, please tell me. Here is a
1253 summary of the main changes.
1255 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1256 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1257 portability, speed, and a lot more but you'll figure it out later.
1258 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1259 I wanted to release the new MSG as soon as possible and I have
1260 broken GRAS, which is the reason why, for now, only MSG is fully
1261 functional. A laconic description of these projects is available
1262 in the documentation.
1264 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1265 kernel optimized for our needs (called SURF but only the developers
1266 should use it). Hence, MSG is now roughly 30 times faster and I think
1267 that by rewriting a little bit MSG, I could event speed it up a little
1268 bit more. Beside the gain in speed, it is also much easier to encode a
1269 new platform model with SURF than it was with SG. More to come...
1271 * REVOLUTION 3: I have tried to change a little as possible the API of
1272 MSG but a few things really had to disappear. The main differences
1273 with the previous version are :
1274 1) no more m_links_t and the corresponding functions. Platforms are
1275 directly read from a XML description and cannot be hard-coded
1276 anymore. The same format is used for application deployment
1277 description. The new format is described in the documentation.
1278 Have a look in tools/platform_generation. There is a tiny script
1279 that converts from the old platform format to the new one. Concerning
1280 the application deployment format, parsing the old one is tricky.
1281 I think most of you should however be able to convert your files. If
1282 it is really an issue, I can write a C code that does the conversion.
1284 2) the toolbox tbx does not exist anymore. We now have a library
1285 with much more data-structures but without the hash-tables (we have
1286 dictionaries that are much faster).
1288 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1290 *****************************************************************************
1291 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1292 * don't feel like converting the oldies. (Mt) *
1293 *****************************************************************************
1296 Version 2.90: "the long awaited one"
1297 - Finished rewriting and debugging MSG. Rewrote the documentation.
1298 - disable GRAS for now since it needs to be ported to the newest SG
1301 - Finish the port to windows (using mingw32 for cross-compile)
1304 - Main loop and datastructures of SURF. A cpu resource object is
1305 functional. Surf can thus be used to create cpu's with variable
1306 performance on which you can execute some actions.
1308 2004-11-15 Martin Quinson
1309 - Port to ARM. Simply added the alignment and size descriptions. Should
1310 work, but the ARM machines are so slow that I didn't had the opportunity
1311 to 'make check' over there yet.
1313 2004-11-15 Arnaud Legrand
1314 - Trace manager now written. It uses a heap structure and is therefore
1315 expected to be efficient. It may however be speeded up (particularly
1316 when many events occur at the same date) by using red and black
1317 trees. One day maybe...
1318 - Max-min linear system solver written. It uses a sparse matrix
1319 structure taking advantage of its expected use. Most operations are
1320 O(1) and free/calloc are called as few as possible. The computation of
1321 the minimum could however be improved by using a red and black tree
1324 2004-11-03 Arnaud Legrand
1325 - Rename every gras_* function that was in xbt/ to its xbt_
1327 - Add a heap and a doubly-linked list to xbt
1328 - Added a dichotomy to the dictionaries. make check works as well before
1329 so I assume that the patch is correct. I do not know however if things
1330 run effectively faster than before now. :)
1332 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1333 SimGrid, and the version number is bumped to 2.x
1335 2004-10-29 Martin Quinson
1336 - Introduction of the remote errors.
1337 They are the result of a RMI/RPC on the remote machine.
1338 ErrCodes being scalar values, you can't get the host on which those
1339 errors did happen. Extending the error mechanism as in Gnome is possible.
1340 No idea yet whether it is a good idea.
1342 2004-10-28 Martin Quinson
1343 - Interface revolution: the Starred Structure Eradication.
1344 I used to do typedef struct {} toto_t; and then handle *toto_t.
1345 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1346 - ???_t is a valid type (builded with typedef)
1347 - s_toto_t is a structure (access to fields with .)
1348 - s_toto is a structure needing 'struct' keyword to be used
1349 - e_toto_t is an enum
1350 - toto_t is an 'object' (struct*)
1352 typedef struct s_toto {} s_toto_t, *toto_t;
1353 typedef enum {} e_toto_t;
1354 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1355 s_toto_t) is private.
1357 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1358 so that it can set the variable to NULL. It was so for dicts and sets,
1359 it changed for dynars.
1361 - Fix a bunch of memleaks in dict_remove
1362 - Fix a bug in sg/server_socket opening: it failed all the time.
1364 2004-10-07 Martin Quinson
1365 - Speed up dynar lookup operation a bit.
1367 gras_dynar_get is dead.
1369 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1370 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1371 (which gives you the address of the stored data).
1373 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1374 copy of the data using an affectation to do the job and not a memcpy.
1376 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1377 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1379 It does not really speedup the dynar test because they are
1380 setting elements all the time (and look them seldom). But the dict does
1381 far more lookup than setting.
1383 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1385 2004-10-05 Martin Quinson
1386 - Allow to (en/dis)able the cycle detection at run time.
1388 Whether we should check for cycle or not is now a property of each
1389 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1390 datadesc_cycle_unset allow to remove this property when previously set.
1392 Note that the cycle detection is off by default since it impacts the
1393 performance. Watch the data you feed GRAS with ;)
1395 This property is hereditary. Any element embedded in a structure having it
1396 set have it set for the time of this data exchange.
1398 You should set it both on sender and receiver side. If you don't set it on
1399 sender side, it will enter an endless loop. If you forget on receiver
1400 side, the cycles won't be recreated after communication.
1402 - Header reorganization.
1403 Kill gras_private.h, each submodule must load the headers it needs.
1405 2004-10-04 Martin Quinson
1406 - Interface revolution: do not try to survive to malloc failure.
1408 Now, gras_malloc and friends call gras_abort() on failure.
1409 As a conclusion, malloc_error is not a valid error anymore, and all
1410 functions for which it was the only gras_error_t return value are
1411 changed. They now return void, or there result directly.
1412 This simplify the API a lot.
1414 2004-09-29 Martin Quinson
1415 - Re-enable raw sockets.
1416 Created by gras_socket_{client,server}_ext;
1417 Used with gras_raw_{send,recv}
1420 It should allow to kill the last bits of gras first version soon.
1422 This is not completely satisfactory yet (duplicate code with
1423 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1426 - Simplify transport plugin (internal) interface by not passing any
1427 argument to _server and _client, but embedding them in the socket
1430 2004-09-28 Martin Quinson
1431 - Finish the port to AIX.
1432 autoconf was my problem (segfault within the malloc replacement
1433 function. No idea why)
1435 2004-09-16 Martin Quinson
1436 - Fix some size_t madness on 64bit architectures.
1438 2004-09-08 Martin Quinson
1439 - Reduce the number of system headers loaded, overload some more system
1440 calls (such as malloc to cast the result of the system one, and work
1442 - Fix and reintroduce the config support
1444 2004-09-07 Martin Quinson
1445 - Source code reorganization to allow Arnaud to surf all over there.
1446 - Allow to document the logging categories.
1447 - Remove all uppercase from logging categories and useless cleanup in names.
1449 2004-08-18 Martin Quinson
1450 Version 0.6.2 (protocol not changed; API changed)
1451 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1452 gras_error_t), and NULL when not found. Functions expecting a msgtype
1453 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1454 hopefully usefull message.
1455 - Portability to prehistoric sparcs again
1457 2004-08-17 Martin Quinson
1458 Version 0.6.1 (protocol not changed; ABI not changed)
1459 - prealloc some buffers to speed things up
1461 2004-08-11 Martin Quinson
1462 Version 0.6 (protocol not changed; ABI expended)
1463 - The parsing macro can deal with the references, provided that you add
1464 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1466 2004-08-09 Martin Quinson
1467 Version 0.5 (protocol not changed; ABI changed)
1468 - Allow to off turn the cycle detection code in data exchange at
1469 compilation time. It should be at run time, but I'm short of time (and
1470 the config stuff is still broken). That way, we keep dict out of the
1471 critical path, which is good because the performance is poor:
1472 - search not dichotomial yet
1473 - dynar give no way to access their content and memcpy everytime
1474 - In composed data description (struct, ref and so on), stop foolness of
1475 keeping the subtype's ID, but store the type itself. This keeps sets out
1476 of the critical path, which is good since they rely on dynar and
1477 dictionnaries. The only loose of that is that we cannot detect the
1478 redeclaration of a structure/union with another content (but I'm not sure
1479 the code detected well this error before anyway). We still can detect
1480 the redefinition discrepancy for the other types.
1481 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1482 it breaks the code because of type-punning used all over the place).
1483 This breaks on all non-gcc architectures (for now).
1485 All those changes (plus the buffer of last time) allow me to gain 2 order
1486 of magnitude on cruel tests consisting of 800000 array of integers on two
1487 level of a hierarchical structure (200 secondes -> 4 secondes)
1490 - the selector of reference must now return the type it points to, not
1491 the ID of this type.
1493 2004-08-06 Martin Quinson
1494 Version 0.4 (protocol changed; ABI not changed)
1495 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1496 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1497 - Implement a buffer transport, and use it by default (it relies on tcp in
1498 real life and on sg in simulation).
1499 That's a bit hackish since I had a new field to the structure to store
1500 its data without interfering with the subtype ones. Inheritance
1501 is tricky in C. And that's a kind of reverse inheritance with one class
1502 derivating two classes. Or maybe a game with java interfaces. Anyway,
1503 that's damn hard in C (at least).
1504 Moreover, I got tired while trying to ensure plugin separation and
1505 genericity in SG mode. MSG wants me to do weird things, so let's go for
1506 cruel hacks (temporarily of course ;).
1507 See comment in transport_private.h:71
1508 - do not include all the _interface headers in private but in the files
1509 which really need them (to cut the compilation time when they are
1512 2004-07-26 Martin Quinson
1513 Version 0.3 (protocol not changed; ABI changed)
1514 - Major overhault of the datadesc interface to simplify it:
1515 - shorted the function names:
1516 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1517 - add a trivial way to push/pop integers into the cbps without malloc.
1518 This allows to make really generic sub_type description, which simply
1519 pop their size of the stack.
1520 - add a function gras_datadesc_ref_pop_arr() which does what users want
1521 most of the time: Declare a dynamic array (which pops its size of the
1522 stack) and declare a reference to it. Poor name, but anyway.
1523 - kill the post-send callback, add a post-receive one
1525 2004-07-23 Martin Quinson
1526 Version 0.2 (protocol changed; ABI changed)
1527 - add some testing for cpbs in the test cases, and fix some more bugs.
1528 This invalidate again the little64 data file, since I cannot regenerate
1530 - remove an awfull optimization in the logging stuff, allowing me to:
1531 - understand it again
1532 - learn gcc how to check that the argument match the provided format
1533 - fix all errors revealed by gcc after that
1534 - internal keys of dict are not \0 terminated. Deal with it properly in
1535 loggings instead of segfaulting when the user want to see the logs :-/
1537 2004-07-22 Martin Quinson
1538 - Fix some stupid bug preventing cbps (callback postit) from working
1540 2004-07-21 Martin Quinson
1541 - Some documentation cleanups
1542 - remove the useless last argument of msgtype_declare
1543 - rename the Virtu functions to fit into the 'os' namespace
1544 - move headers src/include -> src/include/gras/ and stop fooling with
1545 gras -> . symbolic link
1546 - make distcheck is now successful
1548 2004-07-19 Martin Quinson
1550 - Build shared library also
1551 - Install html doc to the right location
1552 - stop removing maintainer files in make clean
1553 - build tests only on make check
1555 2004-07-13 Martin Quinson
1557 - No major issue in previous version => change versionning schema
1558 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1559 corresponding dataset.
1561 2004-07-11 Martin Quinson
1563 - Get it working with any kind of structure (we can compute the padding
1564 bytes remotely for all the architectures I have access to)
1565 - Implement the structure parsing macro (still not quite robust/complete)
1566 - Improvement to the remote testing toysuite
1568 2004-07-10 Martin Quinson
1569 [autoconf mechanism]
1570 - get ride of a bunch of deprecated macros
1571 - actually run the test for two-compliment, not only compile it :-/
1572 - test whether the structures get packed (and bail out if yes. Damn.
1573 Alignment is a serious matter)
1574 - test whether the structures get compacted (but respecting the alignment
1575 constraints of each types)
1576 - test whether the array fields of structures can straddle alignment boundaries
1578 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1579 (took me 2 hours to find that bug, looking at the wrong place)
1580 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1581 before sending/receiving and are used to compute the offsets of fields
1582 - Given that padding size depend even on compiler options, keep track of
1583 alignment and aligned_size only for the current architecture. Not a big
1584 deal since we send structure fields one after the other (seems
1586 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1587 program, let it work on linux/gcc/little32. portability todo.
1589 2004-07-08 Martin Quinson
1590 - import and improve remote compilation support from FAST
1591 - make sure make check works on half a dozen of machines out there
1593 2004-07-07 Martin Quinson
1594 Let's say it's version 0.0.3 ;)
1595 - Implement conversions (yuhu!)
1596 - Let it work on solaris (beside conversion, of course)
1597 - Stupid me, using rand() to generate the conversion datatests in not wise.
1599 2004-07-06 Martin Quinson
1600 - Let make dist work, since I'm gonna need it to compile on remote hosts
1601 - Let Tests/datadesc_usage write the architecture on which the file was
1602 generated as first byte.
1603 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1604 architecture descriptions.
1605 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1606 on those architectures.
1607 - Optimization: send/recv array of scalar in one shoot
1609 2004-07-05 Martin Quinson
1610 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1612 - Plug a whole bunch of memleaks
1613 - each process now have to call gras_{init,exit}. One day, their log
1614 settings will be separated
1615 - Continue the code factorisation between SG, RL and common in Transport.
1617 2004-07-04 Martin Quinson
1619 - Redistribution between SG and RL.
1620 We wanna have to accept in SG, so move accepted related parts of RL in
1621 the common part. (more precisely, the dynar of all known sockets is no
1622 more a static in transport.c, but part of the process_data)
1624 [gras_stub_generator]
1625 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1627 2004-07-03 Martin Quinson
1628 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1629 them all up in one shot)
1631 - Fix the ugly case of reference to dynamic array.
1632 - New (semi-public) function gras_datadesc_size to allow the messaging
1633 layer to malloc the needed space for the buffer.
1635 - gras_socket_close now expect the socket to close (and not its address to
1636 put NULL in it after it). This is because the socket passed to handlers
1637 is one of their argument (=> not writable).
1639 - propagate the interface cleanup from last week in datadesc, ie remove a
1640 superfluous level of indirection. User pass adress of variable
1641 containing data (both when sending and receiving), and not of a variable
1642 being a pointer to the data. Let's say that I like it better ;)
1643 The price for that is constructs like "int msg=*(int*)payload" in
1644 handlers, but it's a fine price, IMHO.
1646 - Let it work in RL (yuhu)
1648 2004-06-21 Martin Quinson
1650 - porting SG plugin and SG select to new standards (works almost).
1651 - plug memleaks and fix bugs around.
1654 - cleanup the prototype of data recv and force users to specify when they
1655 want to handle references to objects. Test case working even for cycles.
1656 - plug memleaks. Valgrind is perfectly ok with this.
1658 2004-06-12 Martin Quinson
1660 - cleanup the separation between plugin and main code in plugin creation
1662 2004-06-11 Martin Quinson
1664 - Reput hook for raw sockets, needed for BW experiments
1665 - kill a few lines of dead code
1666 [Data description] Interface cleanup
1667 - gras_datadesc_by_name returns the searched type or NULL.
1668 That way, no variable is needed to use a type desc once, which makes
1670 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1671 two parameters were strings (field name, type name), leading to
1673 [Dicos] Interface cleanup
1674 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1675 This is consistant with the dynar API.
1677 2004-04-21 Martin Quinson
1679 - Porting to new standards.
1681 - interface cleanup.
1682 There is no bag anymore, no need to take extra provision to mask the
1683 pointers behind "ID".
1684 Better splitup of functions between files create/exchange/convert.
1685 This is still a bit artificial since convert and receive are so
1686 interleaved, but anyway.
1688 - add a queued message list to procdata (the ones not matching criteria
1690 - factorize some more code between SG and RL wrt procdata
1692 - use gras_exit in example to track memleaks
1693 - get rid of gs_example now that GS is properly integrated into gras
1694 - update run_test to integrate the lastest tests (datadesc)
1696 - rename WARNINGn macros to WARNn since it prooved error-prone
1698 2004-04-19 Martin Quinson
1700 - register init/exit functions within gras module mechanism
1701 - send/receive function.
1702 Convertion is not implemented, but short-cutted if not needed.
1703 struct/array elements are sent one by one (instead of block-wise), but
1704 nobody really cares (yet). Get a prototype before optimizing.
1705 - tests (using a file socket) for DD send/receive on:
1706 - base types: int, float
1707 - array: fixed size, string (ie ref to dynamic string)
1708 - structure: homogeneous, heterogeneous
1709 - chained list, graph with cycle
1710 Believe it or not, valgrind is not too unhappy with the results. The
1711 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1715 - Bugfix when using a filename explicitely (instead of '-')
1717 2004-04-09 Martin Quinson
1719 - factorize more code between RL and SG in socket creation
1720 - Complete the implementation and tests of:
1722 o file (only in RL, and mainly for debugging)
1724 I lost 3 days to design a portable address resolver, and then decided
1725 that the prototype mainly have to run on my box.
1726 Addressing portability too early may be like optimizing too early :-/
1728 - use gras_init in the Tests instead of the crappy parse_log_opt
1729 (the latter function is removed)
1730 [Conditional execution]
1731 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1732 [Code reorganisation]
1733 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1734 Build examples against the RL library, since there is no way to disable
1735 its creation for now.
1737 For information, the beginning of coding on GRAS was back in june
1738 2003. I guess that every line has been rewritten at least twice since