1 SimGrid (3.3.5-svn) unstable; urgency=low
3 The "C spoken, se habla Java, Ruby 話せます, fala-se Lua" release.
5 Java Bindings: Various Cleanups
6 * Remove put/get: no need to export deprecated interface in Java
7 Use send/receive instead.
8 * Cleanup the examples and add a README per directory
9 * Remove example autoDestination (that's the only way to go now)
10 * Remove example explicitDestination (was a plain copy of basic)
11 * Make JniException a runtime exception, so that there is no need to
12 declare the fact that you may encounter such a beast. I guess that
13 nobody will ever want to survive such error.
14 * Create specific errors for each MSG case of failure:
15 host failure, transfer failure, timeout, task cancelled
16 Ruby and Lua Bindings: create them
17 * That's new and great, you should try them out.
18 Same functionalities than Java bindings, only even less polished
20 * Kill the useless "rate" argument of SD_task_get_execution_time()
21 Everyone used to provide -1 as a value, it was not used, and the
22 semantic of a possible use wasn't even clear.
23 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
24 either as comm costs or compute costs to mean that there is no
25 such thing for that specific task.
27 * In trace replay, allow to have one trace file per process.
28 Give the specific trace file as argument of each process,
29 and call MSG_action_trace_run(NULL)
30 You can still have one merged file for each processes.
31 * Kill the MSG_paje_output() function. It's a noop since 2 years.
32 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
34 * Add a MSG_task_set_data() function
36 * add a SIMIX_sem_get_capacity() function
37 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
38 * Bugfix: Handle correctly the resume/suspend of processes when they
39 are blocked on semaphores.
41 * Rename configuration variables to start a hierarchy:
42 o cpu_model -> cpu/model
43 o network_model -> network/model
44 o workstation_model -> workstation/model
45 * New configuration variables:
46 o network/bandwidth_factor: correction to bandwith
47 o network/latency_factor: correction to latency
48 o netwotk/weight_S: correction to the weight of competing streams
49 * Add a long description to the models, that users can see with such
50 argument on the command line: --cfg=cpu/model:help
51 * --help-models display the long description of all known models
53 * Implement MPI_Get_count, MPI_MAXLOC, MPI_MINLOC
55 * config: add the ability to set a default value after registration
56 Does not override any previously set value (e.g. from cmd line)
57 * dict: allow to have integer key and data.
58 When so, you need to use the following functions
59 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
60 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
61 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
62 In contrary to regular dicts, the key is not malloced before copy.
63 Mixing scalar and regular elements in the same dict is not tested
65 Tracing for Visualization:
66 * SimGrid is now instrumented in order to generate a trace file for
67 visualization analysis: to use it, need to compile SimGrid with the
68 "tracing" option enabled, and instrument the program using SimGrid with
69 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
70 (among other functions).
71 * The instrumentation only traces the platform utilization for now
72 * Documentation to use the tracing functions and how to analyze the
73 traces with the Triva tool is written.
74 * More information about: SimGrid FAQ (in the section Tracing Simulations
77 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
80 SimGrid (3.3.4) stable; urgency=low
82 The "Desktop Grid needs love too" release (also called Xmas release).
85 * Major speedup in the maxmin system solving by using lazy evaluation
86 Instead of solving completely the maxmin system at each iteration,
87 only invalidate (and recompute) the modified parts.
88 This new feature is enabled in default models but you can try to
89 turn it on with "--cfg:maxmin-selective-update=1" for other models.
90 * Cas01 IMproved as default CPU model
91 This CPU model is the same Cas01 model, but it uses the
92 maxmin-selective-update flag and a heap structure to manage
93 actions on SURF kernel.
94 It reduces the complexity to find the next action to finish and,
95 consequently, it's faster than the old Cas01.
96 This is the new default CPU model (Cas01).
97 * Rename the old Cas01 model to Cas01_fullupdate
98 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
99 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
100 * CpuTI (CPU Trace Integration)
101 A new CPU model whose objective is simulate faster when using
102 availability trace files.
103 Instead of using a full featured, over engineered maxmin system for
104 CPU modeling, this model does the pre-integration of traces files
105 to calculate the amount of CPU power available, and so, executes
106 faster than the old CPU models.
107 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
108 * Use LV08 as default network model since it gives better accuracy
109 for small messages and shouldn't change things for big ones.
110 Use --cfg=network_model:CM02 to get the previous behavior.
113 ******************************************
114 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
115 ******************************************
116 * The new CPU model may changes simulations!
117 The point is that events occurring at the exact same timestamp
118 are not scheduled in the same order with the old and new
119 version. This may be enough to completely change the execution
120 of simulations in some cases.
121 * The new network model will change simulations!
122 This new model is more realistic than the previous one, so you
123 should consider redoing your old experiments with this model.
124 Sorry for the inconvenience.
127 * Introduce the supernovae compilation mode
128 When compiled that way, the whole SimGrid (or almost) is put in a
129 single compilation unit and compiled in one shoot.
130 This is to help gcc which has difficulties to inline stuff from one
132 The speedup seem to be above 15%, althrough more tests are needed on
133 amd64 to confirm that gain.
136 * Port of MSG's mailbox on top of SIMIX network
137 The put/get mechanism was greatly simplified on the way.
140 * New SIMIX network module. Provides:
141 - Mailbox: rendez-vous mecanism to find with who you want to speak
142 - Synchronous send/recv: easier and hopefully faster since the
143 logic is handled in the maestro process directly now
144 - Asynchronous send/recv: you dreamt of it? It's here now
145 Too bad that nobody cared enough to propagate the change to MSG.
146 * Add semaphores as SIMIX synchronization mechanism.
149 * new function SD_daxload(char*) to load a DAX file
150 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
151 * Introduce typed tasks. Specify its kind and cost at creation.
152 At scheduling, just give where it should be placed, and the cost
153 for each involved resource is automatically computed.
154 Existing constructors so far (more to come of course):
155 - SD_task_create_comm_e2e() for end-to-end communication
156 - SD_task_create_comp_seq() for sequential computation
157 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
158 * new function SD_task_dump() for debuging display
159 * new function SD_task_dotty(task,FILE*) writing to file the info
160 about the task in dotty format
161 * SD_task_dependency_exists() can now cope with having one of its
162 arguments NULL. If so, it tests whether the other argument has any
164 * Add getters on list of preceding/following tasks:
165 SD_task_get_parents(task) and SD_task_get_children(task)
166 * Add getters on amount of workstations and list:
167 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
168 * Add getter on task kind: SD_task_get_kind(task)
169 * Update the start_time and finish_time of tasks on completion/failure
170 * Bugfix: Remove task from state swags when destroyed
173 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
176 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
177 dynar with one separate thread per value of the dynar.
178 * Change the prototype of xbt_thread_create(), sorry.
179 Added a boolean parameter indicating whether we want to join this
180 thread (used in SG only for now)
181 * Implement xbt_thread_join and xbt_thread_yield in SG also.
184 * GTNetS wrappers should now be usable again (and betterly tested too)
185 * Fix a major regression from 3.2 where the timeout provided to
186 MSG_task_put_with_timeout() was used as absolute time before which
187 the comm should be done.
188 * Start to fix the <cluster> tag.
189 - Internal links should be good now (beside of the loopback, which
190 use the private link instead)
191 - paths to the external world is still rather broken
192 - the <route:multi> tag is just broken. Actually that's brain-dead.
193 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
194 to make it less stupid
195 ** Check your platform with teshsuite/simdag/platforms/flatifier **
196 * Fix a source-level compatibility glitch from 3.2: after defining
197 MSG_USE_DEPRECATED, you can use the old name
198 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
199 * Allow to compile from the SVN with automake 1.11
200 * Fix some problems when using the "start_time" tag in deployment XMLs.
201 * Fix #8569: XBT/synchro.h has redundant declarations
202 * Fix #8563: MSG return values and exceptions
203 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
204 * Integrate patch #8636: Obey DESTDIR when installing documentation.
205 Thanks to Robson Peixoto.
206 * Fix a vicious bug in dictionaries inducing that some elements were
207 not freed on xbt_dict_free()
209 Portability report of this version:
210 * Main portability targets:
211 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
213 - mac leopard on i386
214 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
215 but nothing critical.
216 * Other platforms: windows, AIX and others were not tested for this release
218 Timing report of this version:
219 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
220 scenario parameters). From 8h to a few seconds in desktop grid settings.
221 * Supernovae brings about 25% speedup on i386.
223 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
225 SimGrid (3.3.3) stable; urgency=low
227 The "Need for Speed" release.
229 The timings done to validate the 3.3.2 were faulty.
230 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
232 The problem was a conversion from a manually handled vector to
233 xbt_dynar_t on the critical path.
234 xbt_dynar_foreach calls functions, inducing stack management crap.
236 We inlined these functions and xbt_dynar_foreach is now breath taking.
237 We also inlined xbt_swag_belong on the way.
239 Here are some approximate speedup measurements (on master/slaves
240 simulations lasting between 10s and 20s each):
241 3.3.1 -> 3.3.2: about same performance
242 3.3.2 -> 3.3.3: 40% speedup
243 3.3.1 -> 3.3.3: 40% speedup
244 3.3.1 with inline patch -> 3.3.3: 30% speedup
246 Our reading is that the refactoring which occurred in 3.3.2 made us
247 suffer much more from the xbt_dynar_foreach low performance, but
248 once we solved this, this refactoring proved to be very performance
249 effective. From the 40% speedup, somehow, 10% are due to the
250 inlining and 30% to the refactoring.
252 That's a pitty that gcc cannot inline functions placed in other files
253 alone. We have to choose between:
254 - break the encapsulation (by putting private data structures and
255 accessors in headers files to help gcc)
256 - live with low performance
257 - switch to a decent compiler such as icc (not quite possible).
259 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
261 SimGrid (3.3.2) stable; urgency=low
263 The "Simplicity does not preceed complexity, but follows it" release.
265 The main contributors of this release were (lexical order):
266 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
269 * Extract the routing logic into its own object.
270 (was dupplicated in network.c and workstation_LV07.c;
271 Allows to implement other ways of storing that info)
272 => kill now useless network_card concept
273 - Use dynar to represent routes (instead of void** + int*)
274 - kill link_set (use surf_network_model->resource_set instead)
275 - Add a command-line option to choose the routing schema to use
276 - Add three new models:
277 * Floyd (shortest path computed at initialization)
278 * Dijikstra (shortest path recomputed all the time)
279 * Cached Dijikstra (shortest path computed on need)
280 All these models where contributed by Silas De Munck, and are
281 described in his ICCS09 paper.
283 * Simplify model declaration
284 (less redirections, less function to write when defining a model)
285 - Factorize stuff between models:
288 surf_model_resource_set(model)
289 surf_model_resource_by_name(model, name)
290 - Unify the types of models in s_surf_model_t (using an union)
291 - Embeed fields of common_public directly into s_surf_model_t
292 - Rename model methods:
293 action_free ~> action_unref
294 action_change_state ~> action_state_set
295 action_get_state ~> action_state_get
296 - Change model methods into functions :
297 (model)->common_public->action_use ~> surf_action_ref
299 * Implement a generic resource; use it as ancestor to specific ones
300 (allows to kill duplicated code in models)
301 Drawback: timer command don't need no name nor properties;
302 workstation_CLM03 don't need no properties
303 (but I guess we can live with those few bytes wasted)
305 * Improve the action object model
306 - implement a constructor avoiding dupplicated code about field
307 initialization in generic_action part.
309 * Kill the SDP model: it has an external dependency, is deprecated
310 in flavor of modern lmm models, and didn't compile since a while
313 * Relocation of the context module from XBT to SIMIX.
314 (the context were decoupled from the simix processes, duplicating a lot of code)
315 => a lot of code was factorized
316 - less overhead is introduced during scheduling
317 - simpler API for the context factory
318 - the logic for process creation,destruction and manipulation was simplified
319 * Simplification of the s_smx_process_t data structure.
320 => accesing the simix level data associated to a process is faster now,
321 and the code is a lot more readable.
324 * Implement some more MPI primitives:
325 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
326 -implementation: Bcast: flat or 2-ary tree (default),
329 Allreduce: Reduce then Bcast
330 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
331 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
332 Alltoallv: flat tree, like ompi
334 * Add support for optimized collectives (Bcast is now binomial by default)
335 * Port smpirun and smpicc to OS X
338 * Kill SD_link_get_properties: hard to maintain and makes very little sense
339 Shout out if you used it.
342 * Display the list of still queued messages in SG mode when existing
346 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
347 * Add xbt_graph_node_get_outedges() [Silas De Munck]
348 * Add xbt_str_from_file(FILE*)
349 * Add xbt_dict_get_key achieving a linear reverse search
350 * Remove the context module
352 Portability report of this version:
353 * Main portability targets:
354 - Linux(debian)/x86/context
355 - Linux(debian)/x86/pthreads
356 - Linux(debian)/amd64/context
357 - Linux(debian)/amd64/pthreads
358 On these, we still have the eratic breakages of gras/pmm and
359 amok/saturate_sg reported in previous version. We still think
360 that the tests are the cause of the fault, not the tested code.
362 - Mac OSX Leopard/x86/context
363 Still false negative in tesh autotesting.
364 Smpi still fails, but this time because readlink does not accept -f
365 Everything seems to work properly beside of that.
368 - AIX version 5.3 (only tested contexts this time)
369 Smpi still fails there because mktemp is not installed.
370 Everything seems to work properly beside of that.
372 I managed to compile it for the first time, but several breakages.
373 Won't delay the release for this exotic platform.
375 * Windows: it's still lagging behind. If you want to help, please
378 Timing report of this version:
379 This version seem to be more than 5% faster than 3.3.1 (on linux
380 64bits with contextes). The gain is less than expected, we are
381 investigating this for next release.
383 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
385 SimGrid (3.3.1) stable; urgency=low
388 * Implement a --cfg-help to show existing configuration variables
389 * Build chain do not require doxygen in maintainer mode
392 * fix a bug on struct sizeof computation, which prevented the
393 exchange of arrays of structs in some conditions
394 - added a regression test about this in datadesc_usage
395 * Allow the exchange of 0-long dynamic vectors.
396 - for that, use -1 as indicator of dynamic size instead of 0
397 - This implied to change any size from unsigned long to long,
398 reducing a bit communication abilities, but I guess that with
399 64bits being quite common, this is more than enough.
400 - This also induce a protocol change, thus bumping network protocol
401 version from 0 to 1 (if we have external users, we have to get
402 clean on that point too ;)
403 - added two regression tests about this in datadesc_usage
404 * Be more verbose when propagating local exceptions
405 This helps debugging.
406 * Display the status of simulated processes when receiving SIGINT in
410 * Allow to control the simulation from a trace file.
411 New functions MSG_action_register() and MSG_action_trace_run()
412 The first one allows to associate a function execution to each
413 kind of action while the second one parses a trace file and
414 triggers the corresponding actions within the system.
415 For now, only a toy example is provided in examples/msg/actions
416 * Add an exemple of process migration in examples/msg/migration
417 * Fix a bug in task exchange which broke MSG_task_get_sender()
418 Add a teshsuite regression test for that.
419 [Bug: if MSG_task_get_sender() is called after sender exit,
421 * Fix a bug which prevented suspend/resume to work properly
422 * Display the status of simulated processes when receiving SIGINT
423 This fixes a regression of v3.3. due to the introduction of SIMIX
424 * Bug fixing in failure management:
425 - trace could not start by a failure at time 0
426 - failure during communications were not working
429 * Add SIMIX_process_set_name() to change the name of the current
430 process in the log messages.
431 * Store smx_hosts in a dict since we only retrieve them by name
432 * Move the configuration infrastructure to surf
435 * Move the configuration infrastructure to surf
438 * Massive internal cleanups:
439 - Store internal structures on processes instead of hosts (allows
440 to have more than one process per host, in addition of being more
442 - Cleanup the initialization/finalization process
443 - Kill a whole bunch of unneeded synchronization:
444 processes run in exclusive manner within the simulator
445 - Move queues from global tables to process data fields
447 - now accept -platform and -hostfile arguments
448 - Pass the right rank value to processes according to the hostfile
449 * Compile the examples by default, and use them as regression tests
450 * Implement MPI_Wtime()
451 * Change the reference speed to a command line option
454 * TCP_gamma can now be specified as command line option using
455 --cfg=TCP_gamma:10000000.0
456 * Change the --surf-path cmd line option into --cfg=path:
459 * Also include strbuff from xbt.h public header
460 * xbt_ex_display(): do not free the exception after displaying
461 This allows to do more with the given exception afterward.
462 Users should call xbt_ex_free() themselves.
466 Portability report of this version:
467 * Main portability targets:
468 - Linux(debian)/x86/context
469 - Linux(debian)/x86/pthreads
470 - Linux(debian)/amd64/context
471 - Linux(debian)/amd64/pthreads
472 These targets fail about 1/10 of times on gras/pmm, but we believe
473 that this is because of the test, not because of simgrid.
474 amok/saturate_sg fails even more rarely, and the test may not be
477 - Mac OSX Leopard/x86/context
478 The test suite still spits tons of errors because some obscure
479 force prevents us from removing the temporary directories
480 arguing that they still contain some metadata I've never heard of.
481 Smpi fails because seq is not installed.
482 Everything seems to work properly beside of that.
485 - AIX version 5.3 (both contexts and pthread)
486 Smpi still fails there because mktemp is not installed.
487 XML inclusions seems rosty on AIX.
489 * Windows: it's still lagging behind. If you want to help, please
492 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
494 SimGrid (3.3) stable; urgency=high
498 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
501 * Introduce the SIMIX module: factorize code between MSG and GRAS.
504 Until now, GRAS were using MSG as an interface to SURF. It was
505 quite difficult because both interface have several differences
506 (MSG channels vs GRAS sockets were the most notable point).
508 This also opens the gate to SMPI (which should occur soon) and speed
509 up simulations by to 40% (even if it were not the main goal).
511 **************************************
512 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
513 **************************************
514 The point is that events occuring at the exact same timestamp are
515 not scheduled in the same order with the old and new version. This
516 may be enough to completely change the execution of simulations in
517 some cases. Sorry for the inconvenience.
519 * Cleanup and upgrade the XML format to push further scalability
520 issues (check http://hal.inria.fr/inria-00256883/ for more info)
522 * Improve the testing infrastructure with tesh. Now a very large part of
523 the code is tested not only by being run but also by checking that the
524 output match an expected output [Mt].
526 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
527 is really less memory-demanding, which should allow you to use
528 larger files in SimGrid [AL].
530 * Inform valgrind about our contextes, so that it becomes usable
531 with the default (and more effecient) version of SimGrid
532 [contributed by Sékou Diakite, many thanks]
535 * Introduce a listener thread in charge of receiving incomming
536 messages from the network. It allows to overlap communication and
537 computation but most notably, it removes some stupid deadlocks due
538 to the fact that so far, a process could not send and receive at
539 the same time. This made most non trivial communication schema
541 * Convert the PIDs from long int to int to match the MSG ones (and
543 * New function: gras_agent_spawn() to launch a new process on
544 current host. Only working in simulation for now. [Mt]
545 * New function: gras_os_hostport() returning a constant form (ie,
546 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
549 * Make the backtrace of exceptions more human readable [Mt]
550 * New module: xbt/str [Mt]
551 a ton of string utility functions (split, join, printf to a newly
552 allocated buffer, trim, etc)
553 * New module: xbt/hash [Mt]
554 SHA1 hashing algorithm (more to come if needed)
555 * New module: xbt/synchro [Mt]
556 synchronization tools (mutex and conditions) working the same way
557 in simulation and in real life (mainly useful for GRAS, but not
559 * New module: xbt/queue [Mt]
560 classical producer/consumer synchronization scheme
561 * xbt_dynar_new_sync() creates a synchronized dynar. All access
562 (using the classical functions will get serialized) [Mt]
563 * Make dictionary internal table dynamic. No need to specify its size
564 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
566 * Make sure the log channels are organized as a tree under windows
567 (because of ANSI C compatibility issue, any channel were child of
571 * Cleaned many thing in surf and fixed a few bugs [AL].
572 * Add a nice command line configuration mechanism to compose models [AL].
573 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
574 the previous one (KCCFLN05). It relies on something that looks like
575 a max-min sharing mechanism but cannot be written as such. A new solver
576 was thus designed [AL].
577 * Add a new solver to lmm. Based on Lagrange optimization and
578 gradient-based descent, it enables to efficiently maximise systems s.a
580 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
582 This solver enables to propose two new network models for TCP Reno and
583 TCP Vegas based on Low's work. These models still need to be fully
584 tested though [Pedro Velho].
587 * Bug fix in SD_simulate. Now the time bound given as argument is
589 * Use the new parallel task model (ptask_L07) as default.
590 * Use the SURF command line configuration mechanism.
591 * 0-size tasks (for synchronization) should now work.
593 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
595 SimGrid (3.2) stable; urgency=high
599 We still experience issues on this platform, but we believe that at
602 GRAS API BREAKAGE (for simplification purpose, sorry):
603 * the gras_msgtype_by_name is not used anymore. Instead of
604 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
605 you can write (and must)
606 gras_msg_send(toserver, "request", &request);
607 - If you still want to pass a gras_msgtype_t to the function (to cache
608 the type and avoid the lookup time), use the gras_msg_send_() variant.
609 - Impacted functions:
610 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
611 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
612 * The callbacks are now expected to return 0 when everything went well
613 (just like the main() function)
615 GRAS new features and improvements:
616 * New module mecanism where user code can use per process globals [Mt]
617 This is similar to gras_userdata_*() functions, but for libraries. It
618 factorize some code developped over and over in the examples and AMOK.
619 It has still to be documented and used (only amok/peermanagement is
621 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
622 loss when they were small enough to fit into the buffer and sent quickly
623 enough so that they can all get received in one shoot.
624 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
625 if not found. Use the *_or_null() variant for the old semantic.
626 * In gras_msg_handle, do not discard messages without callback.
627 They are probably messages to be explicitly awaited later (ie, proofs of
628 mis-synchronization in userland since they are sent before being awaited)
630 * gras_socket_meas_send/recv: semantic changed!
631 The numerical arguments used to be (1) the total amount of data to send
632 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
633 messages. This was need for the fool willing to send more than MAXINT
634 bytes on quite fat pipes.
637 * Do really rename the hostmanagement module to peermanagement. [Mt]
638 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
639 but this is rather new and this was documented in the module
640 documentation (poor excuses, I admit)
641 * Bandwidth measurement semantic changed! This follows the changes to
642 gras_socket_meas_send/recv explained above.
645 * A sequential mode has been added to the workstations. When a workstation
646 is in sequential mode, it can execute only one task, and the other tasks
647 are waiting in a FIFO. [Christophe Thiery]
650 * The KCCFLN05 workstation model now handles parallel tasks. It is the
651 model for SIMDAG. [Christophe Thiery]
652 * Bug fix in the maxmin solver: Some values were close to 0 instead of
653 equal to 0, which caused some bad behaviors in
654 saturated_constraint_set_update. I now use a threshold mechanism like in
658 * When running manually src/testall, you select specific units [Mt]
659 testall is the result of our cunit mecanism, and should replace all
660 the scripty thingy around since bash don't run easily on billware.
662 * A mallocator system has been added. [Christophe Thiery]
663 Mallocators allow you to recycle your unused objects instead of freeing them
664 and allocating new ones.
666 Documentation update:
667 * FAQ reworking + New FAQs:
668 - "Valgrind spits tons of errors!" [Mt]
669 - "How to repport bugs" [Mt]
670 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
671 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
672 - Communication time measurement within MSG [AL]
673 - I experience weird communication times when I change the latency [AL]
676 - an introduction to the framework and to the used communication model
677 - an initiatic tour introducing the most proheminent features:
679 . Lesson 0: Installing GRAS
680 . Lesson 1: Setting up your own project
681 o Part 2: Message passing
682 . Lesson 2: Exchanging simple messages
683 . Lesson 3: Passing arguments to the processes (in SG)
684 . Lesson 4: Attaching callbacks to messages
685 . Lesson 5: Using globals in processes
686 . Lesson 6: Logging informations properly
687 . Lesson 7: Using internal timers
688 . Lesson 8: Handling errors through exceptions
689 . Lesson 9: Exchanging simple data
690 . Lesson 10: Remote Procedure Calling (RPC)
691 . Lesson 11: Explicitely waiting for messages
692 . Recapping of message passing features in GRAS
693 - A HOWTO section containing:
694 o HOWTO design a GRAS application
695 More are due, of course. They will come latter. In the meanwhile, you can
696 check the examples which are still here.
698 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
700 SimGrid (3.1) stable; urgency=high
704 There was a stack corruption somewhere, visible only when optimizing
705 with these versions. [Vince]
708 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
709 module that enabled to play with Directed Acyclic Graphs. It is built
710 directly on top of SURF and provides an API rather close to the old
711 SG. Some old codes using SG are currently under rewrite to check that
712 all needful functions are provided. [Christophe Thiery]
715 * Complete rewrite of the KCCFLN05 workstation model. It is now an
716 extension of the classical CLM03 model that gracefully handles
717 failures. This is now the default model for MSG and GRAS. It doesn't
718 handle parallel tasks yet though. [AL]
719 * Bug fix: Weights were not correctly set in the network part.
720 WARNING: This may have resulted in incorrect results with simulations
721 where there are more than one flow on a given link. [AL]
724 * After a (long ?) discussion on simgrid-devel, we have decided that the
725 convention we had on units was stupid. That is why it has been decided
726 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
727 WARNING : This means that all previous platform files will not work as
728 such with this version! A warning is issued to ask users to update
730 A conversion script can be found in the contrib module of the CVS, under
731 the name contrib/platform_generation/surfxml_update.pl [MQ]
734 * Bug fix: Processes were started in reverse order, wrt deployment file.
735 WARNING: if your code relies on this bug, please fix it. [AL]
736 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
737 being executed on two different locations. [AL]
738 * Bug fix: Failures are now better supported thanks to Derrick's tests
739 (there was many failure situations I hadn't thought of and that weren't
740 correctly handled). [AL]
741 * New function: MSG_host_is_avail indicates you whether a given m_host_t
745 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
746 Exception occurring on server-side are propagated back to client (!).
748 API CHANGE: the callback changed their prototype. Change:
749 int my_handler(gras_socket_t expeditor, void *payload_data) {
751 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
752 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
754 * New! function: gras_msg_handleall to deal with all messages arriving
755 within a given period.
756 * New! function: gras_socket_server_range to get a server socket in a
757 range of port numbers (ease to avoid port number conflicts) [MQ]
758 * New! gras processes display their backtrace when they get a SIGUSR1
759 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
760 Sweet to debug RL processes [MQ]
764 - Do not force experiment sizes to be expressed in kb, or it becomes
765 impossible to measure the latency this way (needs one byte-long tests)
766 WARNING: this changes the amok_bw_* function semantic. [MQ]
767 - Implements the link saturation stuff. [MQ]
768 * Peer management module:
769 New! module factorizing code that we wrote over and over [MQ].
772 * New module: cunit (my jUnit implementation in ansi C) [MQ]
773 - Test units are placed directly into the library code, they get extracted
774 automatically and placed into the src/testall binary.
775 - Convert most of the XBT tests to this system.
776 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
777 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
778 live in the same C code. This required to change a little bit the API
779 of surfxml parsing but shouldn't be an issue for end-users. [AL]
780 * New module: sparse graph structure with basic algorithms (this is work
781 in progress and the API is not considered to be frozen yet). [AL]
782 * Display more information on backtraces: source line & function names are
783 now displayed just like valgrind does (rely on addr2line tool) [MQ]
784 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
785 * Reworked a little bit some #include statements to load only required
786 headers. Some user code that relied on SimGrid to include stdlib or
787 stdio may need to include it by themselves. [AL]
788 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
790 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
791 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
792 but I guess that almost nobody used those parts. [MQ]
794 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
796 SimGrid (3.0.1) stable; urgency=low
799 * Unfortunately, I had missed 5 misnamed functions:
800 xbt_fifo_item_t xbt_fifo_newitem(void);
801 void xbt_fifo_freeitem(xbt_fifo_item_t);
802 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
803 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
804 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
805 They're now deprecated. Please use their new versions:
806 xbt_fifo_item_t xbt_fifo_new_item(void);
807 void xbt_fifo_free_item(xbt_fifo_item_t);
808 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
809 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
810 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
812 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
813 * Documentation: xbt_log module unmercifully reworked [MQ]
814 * Bugfix: there was a problem with the ending of contexts with
815 the pthread backend. It caused some weird deadlock or behavior
816 depending on the pthread implementation. [AL]
817 * Bugfix: get the exceptions raised in the simulator repport where
818 and why they come from when they are not catched in time [AL, MQ]
821 * Bugfix: Do repport the error when two non-connected hosts try to
822 exchange data (Thanks to Flavien for stumbling into this one) [AL]
825 * Add additionnal checkings on communications. Assert that two
826 communicating hosts are connected by a set of links... [AL]
829 * Add additionnal checkings on channel values in communication [AL]
830 * New: MSG_task_get_source to see on which host a task was generated [HC]
831 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
832 the number of tasks waiting to be received on channel and sent
834 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
835 waits for the first task coming from a given host.. [AL]
837 GRAS new functionnalities: [MQ]
838 * Enhance the parsing macro to allow the size of multidimentional objects
839 to be given thru annotations.
840 * New example (and documentation): Matrix Multiplication a la RPC
841 (as when I was young!) and fix a bunch of bugs found on the way.
843 GRAS performance improvements: [MQ]
845 * Reduce the amount of cbps creation/destruction by making it static to
846 datadesc_send/recv() and using a (newly created) cbps_reset (based on
849 * Change libdata to a set so that we can search for stuff by ID (and thus
850 reduce the insane amount of dict lookups)
853 * Actually implement gras_datadesc_copy() so that we don't have to mimick
854 RL communication on top of SG since it's so uneffective.
855 It may also be used for inter-thread communication in RL, one day.
856 * Use gras_datadesc_copy() to exchange messages on top of SG
858 - improve message exchange performance on top of SG
859 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
860 * Don't exchange on the network the size of the used part of buffer,
861 instead, specify the possible buffer size to read().
863 - reduces the amount of read/write calls (one pair per exchange)
864 - reduces the amount of exchanged data (the size)
865 - allows to retrieve all arrived data on receiver side, if we don't need
866 it right now (subsequent read will peek the buffer)
867 - allows the receiver to proceed with the begining of the stream before
868 everything is arrived
869 - make it possible to build an iov transport (using readv/writev)
871 - take care of the data with non-stable storage (like stacked data),
873 * If possible, TCP send uses vector I/O (when writev() is here)
874 - Don't use it for receive since we send data sizes and data on the
875 same stream, so we wouldn't be able to chain large amount of chunks
876 before having to flush the stuff to read the size.
877 * Rework the transport plugin mecanism to simplify it and reduce the
878 amount of pointer dereferencement when searching for the right function
881 * I guess that now, we do almost as few system calls as possible while
882 doing as few data copy as possible.
884 To improve it further, we could try to send all the sizes first and then
885 all the data (to use iov on receiving size), but it's only a partial
886 solution: when you have 2 dimensional data, the sizes of the second
887 dimension is data of the first dimension, so you need 3 streams.
889 I'm not sure the potential performance gains justify the coding burden.
891 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
893 SimGrid (3.00) stable; urgency=high
896 * New! Give the possibility to hijack the surf parser and thus bypass
897 MSG_create_environment and MSG_launch_application. Have a look at
898 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
900 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
902 SimGrid (2.96) unstable; urgency=low
907 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
908 This deprecates the xbt_error_t mecanisms.
909 It modifies (simplifies) all XBT and GRAS API.
910 MSG API keeps unchanged (exceptions raised by XBT are catched from
911 within MSG and masked with existing error handling facilities)
914 * New! Add a FATPIPE model. [AL]
915 * New! Add a parallel task model. [AL]
916 * New! Add automatically a loopback interface (in the default
917 network model) if none was precised.
920 * Bugfix: MSG_process_resume now works with the current running process.
922 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
923 * Modification of MSG_task_get_compute_duration. Once a task has been
924 processed, the value returned by this function is now equal to 0. [AL]
925 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
926 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
927 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
928 of a m_task. For now, MSG_task_cancel only works with computation
930 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
931 of the processor (in Mflop/s) regardless of the current load on the
933 * API Change: use proper naming convention for MSG_getClock and
934 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
936 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
937 This function changes the priority of a computation task. This priority
938 doesn't affect the transfer rate. A priority of 2 will make a task
939 receive two times more cpu power than the other ones. This function
940 has been added to suit the needs of Nguyen The Loc and hasn't been that
941 much tested yet. So if it fails, please report it and send me your code.
943 * API Change: removed all functions and types that were marked "deprecated"
944 since many months. Renamed MSG_global_init_args to MSG_global_init.
946 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
948 SimGrid (2.95) unstable; urgency=low
951 * Steal some nice code to GNU pth to fix context detection and usage [AL]
952 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
953 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
956 * Bugfix: Allow absolute paths for platform description files [MQ]
957 * Bugfix: do free the variables after use. Leads to drastic performance
959 * Implement max_duration (ie, timeouts) on resources [AL]
962 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
964 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
965 the needed support (timeouts) [AL]
968 * Implement measurement sockets. You can now get the bandwidth between two
969 hosts thanks to AMOK (see below). [MQ]
970 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
971 dynar over the network (yeah) [MQ]
972 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
973 * simple/static token-ring example. [Alexandre Colucci and MQ]
974 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
978 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
979 type parsing. "short" and "long" modifiers were also ignored; other
980 modifier (such as reference level) are still ignored. [MQ]
981 * Embeed the buffer size within the buffer itself on SG. [MQ]
982 That way, send() are atomic and cannot get intermixed anymore (at least
983 the ones which are less than 100k; bigger messages still have the issue)
984 * Array size pushed by the field, not by the field type (or each
985 and every long int will push stuff to the cbps) [MQ]
986 * use select() to sleep since it allows to portably sleep less than one
989 GRAS (minor cleanups)
990 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
993 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
994 * type category 'ignored' killed as it was never used and were difficult
996 * whether a type can cycle or not is now a flag, leaving room for more
997 flags (as "ignored", if we feel the need one day ;)
998 * Rename raw sockets to measurement sockets since "raw" has another
999 meaning in networking community.
1002 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1003 and offers features not belonging to GRAS but that most applications
1004 need. One day, it may be a set of plugins loadable at runtime.
1005 * New module: bandwidth
1006 bandwidth measurement between arbitrary nodes running this module. [MQ]
1008 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1010 SimGrid (2.94) unstable; urgency=low
1012 The first beta release of SimGrid 3 !
1016 * Update the main page and the FAQ. Adding references to gforge.
1019 * Add a gras_os_getpid function.
1022 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1023 * Extend the logs so that they also print PID, hostname, date, ... if
1025 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1026 and kill the old version which were in testsuite/
1027 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1031 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1032 for users. Instead, each programming environment (GRAS, MSG,...) use it
1033 when needed to provide such a feature to users.
1034 Don't shortcut the mecanism or you will also shortcut the virtualization
1035 you need on the simulator.
1039 * Cleanups in configury with regard to compile optimization/warning flags.
1040 Also add -fno-loop-optimize to any powerpc since it's the optimization
1041 killing gcc (< 3.4.0).
1042 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1043 needed by MSG examples complications
1044 * Borrow configury beautifications from PHP
1047 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1048 warning (thanks loris for stumbling into this one).
1049 * Bugfix: stop loading private headers (gras_config.h) from the public
1053 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1054 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1055 RL would imply a DNS resolver, which is *hard* to do in a portable way
1056 (and therefore delayed).
1057 * Implement a real timer mecanism and use it in timing macros. This allows
1058 to avoid rounding errors and get a 0.000005 sec precision in timing
1059 macros. While I was at it, various cleanups:
1060 - allow to declare more than one timed section per file (fix a stupid bug)
1061 - move some private declaration to the right place
1062 - merge conditional execution and timing macros into emulation module
1063 - document the module
1064 - make sure the module cleanups its mess on gras_exit
1065 * Documentation improvements:
1066 - (new) how to compile applications using GRAS
1067 - (new) emulation support (timing macros)
1069 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1071 SimGrid (2.93) unstable; urgency=low
1073 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1076 - Use Paje properly where used. Still to be sanitized properly.
1077 - Portability fix: Add an implementation of the contexts using pthread
1081 - Add xbt_procname(): returns the name of the current process.
1082 Use it to show the current process's name in all logging.
1084 - fix detection of older flex version and the reaction, since we do
1085 depend on modern ones (we use lex_destroy)
1086 - Better separation of SG and RL in the libs: remove all simulation code
1087 from libgras. As a result, this lib is now only 200k when stripped.
1088 Some of the xbt modules may also be duplicated (two sets and such) and
1089 should be cleaned/killed before SG3.
1090 - Insist on using xlC on AIX because of weird problems involving gcc there.
1091 - Cleanup the make remote stuff. This is now done by scripts
1092 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1093 mainly for our private use, but we're working on changing them to user
1096 - Bugfix: flush the socket on close only if there is some *output*.
1097 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1099 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1102 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1104 SimGrid (2.92) unstable; urgency=low
1106 Alpha 3 on the path to SimGrid 3
1110 - New! First try of benchmarking macros.
1111 - New! First try so that gras_stub_generator generate deployment and
1112 remote compilation helpers.
1114 - Bugfix: Initialization fix in msg_test.
1118 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1120 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1122 - API Change: gras_os_sleep to take the amount of seconds as a double.
1123 Accepting an int was error prone since it was the only location where
1124 seconds were coded as such. It leaded to damn rounding errors.
1125 - Bugfix: Hard to belive that timers ever worked before this.
1127 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1129 SimGrid (2.91) unstable; urgency=low
1131 Alpha 2 on the path to SimGrid 3
1135 - Bug fix in the lmm_solver.
1137 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1138 through the function MSG_paje_output.
1139 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1140 - It is possible to bound the rate of a communication in MSG with
1141 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1142 to write it in the changelog).
1143 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1148 - Various cleanups to the autotools stuff
1149 - Begin to move Gras examples to examples/gras/
1150 - Let make distcheck work again (yeah!)
1152 - documentation overhauled using doxygen.
1153 gtk-doc-tools is dead in SimGrid now.
1154 - Automatically extract all existing logging categories, and add the list
1155 to the documentation (long standing one, to say the less)
1157 - Cleanup the known architecture table. Reorder the entries to group what
1158 should be, and use a more consistent naming scheme.
1159 (some of the test dataset are still to be regenerated)
1160 - New! Allow library to register globals on each process just as userdata
1162 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1163 lookup time (for now).
1164 Use it in msg and trp.
1165 This cleans a lot the internals and helps enforcing privacy of the
1166 headers between the gras components.
1167 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1168 - Bugfix: gras_os_time was delirious in RL.
1169 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1171 - Reenable GRAS now that it works.
1173 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1175 SimGrid (2.90) unstable; urgency=low
1177 Alpha 1 on the path to SimGrid 3
1179 * It is a long time since the last release of SimGrid. I'm sorry about
1180 that but as I had told you, I was rewriting a lot of things. I apologize
1181 to those who had been reporting bugs to me and that I had not answered.
1182 If your bug is still in the new version, please tell me. Here is a
1183 summary of the main changes.
1185 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1186 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1187 portability, speed, and a lot more but you'll figure it out later.
1188 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1189 I wanted to release the new MSG as soon as possible and I have
1190 broken GRAS, which is the reason why, for now, only MSG is fully
1191 functional. A laconic description of these projects is available
1192 in the documentation.
1194 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1195 kernel optimized for our needs (called SURF but only the developers
1196 should use it). Hence, MSG is now roughly 30 times faster and I think
1197 that by rewriting a little bit MSG, I could event speed it up a little
1198 bit more. Beside the gain in speed, it is also much easier to encode a
1199 new platform model with SURF than it was with SG. More to come...
1201 * REVOLUTION 3: I have tried to change a little as possible the API of
1202 MSG but a few things really had to disappear. The main differences
1203 with the previous version are :
1204 1) no more m_links_t and the corresponding functions. Platforms are
1205 directly read from a XML description and cannot be hard-coded
1206 anymore. The same format is used for application deployment
1207 description. The new format is described in the documentation.
1208 Have a look in tools/platform_generation. There is a tiny script
1209 that converts from the old platform format to the new one. Concerning
1210 the application deployment format, parsing the old one is tricky.
1211 I think most of you should however be able to convert your files. If
1212 it is really an issue, I can write a C code that does the conversion.
1214 2) the toolbox tbx does not exist anymore. We now have a library
1215 with much more data-structures but without the hash-tables (we have
1216 dictionaries that are much faster).
1218 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1220 *****************************************************************************
1221 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1222 * don't feel like converting the oldies. (Mt) *
1223 *****************************************************************************
1226 Version 2.90: "the long awaited one"
1227 - Finished rewriting and debugging MSG. Rewrote the documentation.
1228 - disable GRAS for now since it needs to be ported to the newest SG
1231 - Finish the port to windows (using mingw32 for cross-compile)
1234 - Main loop and datastructures of SURF. A cpu resource object is
1235 functional. Surf can thus be used to create cpu's with variable
1236 performance on which you can execute some actions.
1238 2004-11-15 Martin Quinson
1239 - Port to ARM. Simply added the alignment and size descriptions. Should
1240 work, but the ARM machines are so slow that I didn't had the opportunity
1241 to 'make check' over there yet.
1243 2004-11-15 Arnaud Legrand
1244 - Trace manager now written. It uses a heap structure and is therefore
1245 expected to be efficient. It may however be speeded up (particularly
1246 when many events occur at the same date) by using red and black
1247 trees. One day maybe...
1248 - Max-min linear system solver written. It uses a sparse matrix
1249 structure taking advantage of its expected use. Most operations are
1250 O(1) and free/calloc are called as few as possible. The computation of
1251 the minimum could however be improved by using a red and black tree
1254 2004-11-03 Arnaud Legrand
1255 - Rename every gras_* function that was in xbt/ to its xbt_
1257 - Add a heap and a doubly-linked list to xbt
1258 - Added a dichotomy to the dictionaries. make check works as well before
1259 so I assume that the patch is correct. I do not know however if things
1260 run effectively faster than before now. :)
1262 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1263 SimGrid, and the version number is bumped to 2.x
1265 2004-10-29 Martin Quinson
1266 - Introduction of the remote errors.
1267 They are the result of a RMI/RPC on the remote machine.
1268 ErrCodes being scalar values, you can't get the host on which those
1269 errors did happen. Extending the error mechanism as in Gnome is possible.
1270 No idea yet whether it is a good idea.
1272 2004-10-28 Martin Quinson
1273 - Interface revolution: the Starred Structure Eradication.
1274 I used to do typedef struct {} toto_t; and then handle *toto_t.
1275 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1276 - ???_t is a valid type (builded with typedef)
1277 - s_toto_t is a structure (access to fields with .)
1278 - s_toto is a structure needing 'struct' keyword to be used
1279 - e_toto_t is an enum
1280 - toto_t is an 'object' (struct*)
1282 typedef struct s_toto {} s_toto_t, *toto_t;
1283 typedef enum {} e_toto_t;
1284 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1285 s_toto_t) is private.
1287 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1288 so that it can set the variable to NULL. It was so for dicts and sets,
1289 it changed for dynars.
1291 - Fix a bunch of memleaks in dict_remove
1292 - Fix a bug in sg/server_socket opening: it failed all the time.
1294 2004-10-07 Martin Quinson
1295 - Speed up dynar lookup operation a bit.
1297 gras_dynar_get is dead.
1299 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1300 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1301 (which gives you the address of the stored data).
1303 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1304 copy of the data using an affectation to do the job and not a memcpy.
1306 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1307 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1309 It does not really speedup the dynar test because they are
1310 setting elements all the time (and look them seldom). But the dict does
1311 far more lookup than setting.
1313 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1315 2004-10-05 Martin Quinson
1316 - Allow to (en/dis)able the cycle detection at run time.
1318 Whether we should check for cycle or not is now a property of each
1319 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1320 datadesc_cycle_unset allow to remove this property when previously set.
1322 Note that the cycle detection is off by default since it impacts the
1323 performance. Watch the data you feed GRAS with ;)
1325 This property is hereditary. Any element embedded in a structure having it
1326 set have it set for the time of this data exchange.
1328 You should set it both on sender and receiver side. If you don't set it on
1329 sender side, it will enter an endless loop. If you forget on receiver
1330 side, the cycles won't be recreated after communication.
1332 - Header reorganization.
1333 Kill gras_private.h, each submodule must load the headers it needs.
1335 2004-10-04 Martin Quinson
1336 - Interface revolution: do not try to survive to malloc failure.
1338 Now, gras_malloc and friends call gras_abort() on failure.
1339 As a conclusion, malloc_error is not a valid error anymore, and all
1340 functions for which it was the only gras_error_t return value are
1341 changed. They now return void, or there result directly.
1342 This simplify the API a lot.
1344 2004-09-29 Martin Quinson
1345 - Re-enable raw sockets.
1346 Created by gras_socket_{client,server}_ext;
1347 Used with gras_raw_{send,recv}
1350 It should allow to kill the last bits of gras first version soon.
1352 This is not completely satisfactory yet (duplicate code with
1353 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1356 - Simplify transport plugin (internal) interface by not passing any
1357 argument to _server and _client, but embedding them in the socket
1360 2004-09-28 Martin Quinson
1361 - Finish the port to AIX.
1362 autoconf was my problem (segfault within the malloc replacement
1363 function. No idea why)
1365 2004-09-16 Martin Quinson
1366 - Fix some size_t madness on 64bit architectures.
1368 2004-09-08 Martin Quinson
1369 - Reduce the number of system headers loaded, overload some more system
1370 calls (such as malloc to cast the result of the system one, and work
1372 - Fix and reintroduce the config support
1374 2004-09-07 Martin Quinson
1375 - Source code reorganization to allow Arnaud to surf all over there.
1376 - Allow to document the logging categories.
1377 - Remove all uppercase from logging categories and useless cleanup in names.
1379 2004-08-18 Martin Quinson
1380 Version 0.6.2 (protocol not changed; API changed)
1381 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1382 gras_error_t), and NULL when not found. Functions expecting a msgtype
1383 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1384 hopefully usefull message.
1385 - Portability to prehistoric sparcs again
1387 2004-08-17 Martin Quinson
1388 Version 0.6.1 (protocol not changed; ABI not changed)
1389 - prealloc some buffers to speed things up
1391 2004-08-11 Martin Quinson
1392 Version 0.6 (protocol not changed; ABI expended)
1393 - The parsing macro can deal with the references, provided that you add
1394 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1396 2004-08-09 Martin Quinson
1397 Version 0.5 (protocol not changed; ABI changed)
1398 - Allow to off turn the cycle detection code in data exchange at
1399 compilation time. It should be at run time, but I'm short of time (and
1400 the config stuff is still broken). That way, we keep dict out of the
1401 critical path, which is good because the performance is poor:
1402 - search not dichotomial yet
1403 - dynar give no way to access their content and memcpy everytime
1404 - In composed data description (struct, ref and so on), stop foolness of
1405 keeping the subtype's ID, but store the type itself. This keeps sets out
1406 of the critical path, which is good since they rely on dynar and
1407 dictionnaries. The only loose of that is that we cannot detect the
1408 redeclaration of a structure/union with another content (but I'm not sure
1409 the code detected well this error before anyway). We still can detect
1410 the redefinition discrepancy for the other types.
1411 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1412 it breaks the code because of type-punning used all over the place).
1413 This breaks on all non-gcc architectures (for now).
1415 All those changes (plus the buffer of last time) allow me to gain 2 order
1416 of magnitude on cruel tests consisting of 800000 array of integers on two
1417 level of a hierarchical structure (200 secondes -> 4 secondes)
1420 - the selector of reference must now return the type it points to, not
1421 the ID of this type.
1423 2004-08-06 Martin Quinson
1424 Version 0.4 (protocol changed; ABI not changed)
1425 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1426 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1427 - Implement a buffer transport, and use it by default (it relies on tcp in
1428 real life and on sg in simulation).
1429 That's a bit hackish since I had a new field to the structure to store
1430 its data without interfering with the subtype ones. Inheritance
1431 is tricky in C. And that's a kind of reverse inheritance with one class
1432 derivating two classes. Or maybe a game with java interfaces. Anyway,
1433 that's damn hard in C (at least).
1434 Moreover, I got tired while trying to ensure plugin separation and
1435 genericity in SG mode. MSG wants me to do weird things, so let's go for
1436 cruel hacks (temporarily of course ;).
1437 See comment in transport_private.h:71
1438 - do not include all the _interface headers in private but in the files
1439 which really need them (to cut the compilation time when they are
1442 2004-07-26 Martin Quinson
1443 Version 0.3 (protocol not changed; ABI changed)
1444 - Major overhault of the datadesc interface to simplify it:
1445 - shorted the function names:
1446 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1447 - add a trivial way to push/pop integers into the cbps without malloc.
1448 This allows to make really generic sub_type description, which simply
1449 pop their size of the stack.
1450 - add a function gras_datadesc_ref_pop_arr() which does what users want
1451 most of the time: Declare a dynamic array (which pops its size of the
1452 stack) and declare a reference to it. Poor name, but anyway.
1453 - kill the post-send callback, add a post-receive one
1455 2004-07-23 Martin Quinson
1456 Version 0.2 (protocol changed; ABI changed)
1457 - add some testing for cpbs in the test cases, and fix some more bugs.
1458 This invalidate again the little64 data file, since I cannot regenerate
1460 - remove an awfull optimization in the logging stuff, allowing me to:
1461 - understand it again
1462 - learn gcc how to check that the argument match the provided format
1463 - fix all errors revealed by gcc after that
1464 - internal keys of dict are not \0 terminated. Deal with it properly in
1465 loggings instead of segfaulting when the user want to see the logs :-/
1467 2004-07-22 Martin Quinson
1468 - Fix some stupid bug preventing cbps (callback postit) from working
1470 2004-07-21 Martin Quinson
1471 - Some documentation cleanups
1472 - remove the useless last argument of msgtype_declare
1473 - rename the Virtu functions to fit into the 'os' namespace
1474 - move headers src/include -> src/include/gras/ and stop fooling with
1475 gras -> . symbolic link
1476 - make distcheck is now successful
1478 2004-07-19 Martin Quinson
1480 - Build shared library also
1481 - Install html doc to the right location
1482 - stop removing maintainer files in make clean
1483 - build tests only on make check
1485 2004-07-13 Martin Quinson
1487 - No major issue in previous version => change versionning schema
1488 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1489 corresponding dataset.
1491 2004-07-11 Martin Quinson
1493 - Get it working with any kind of structure (we can compute the padding
1494 bytes remotely for all the architectures I have access to)
1495 - Implement the structure parsing macro (still not quite robust/complete)
1496 - Improvement to the remote testing toysuite
1498 2004-07-10 Martin Quinson
1499 [autoconf mechanism]
1500 - get ride of a bunch of deprecated macros
1501 - actually run the test for two-compliment, not only compile it :-/
1502 - test whether the structures get packed (and bail out if yes. Damn.
1503 Alignment is a serious matter)
1504 - test whether the structures get compacted (but respecting the alignment
1505 constraints of each types)
1506 - test whether the array fields of structures can straddle alignment boundaries
1508 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1509 (took me 2 hours to find that bug, looking at the wrong place)
1510 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1511 before sending/receiving and are used to compute the offsets of fields
1512 - Given that padding size depend even on compiler options, keep track of
1513 alignment and aligned_size only for the current architecture. Not a big
1514 deal since we send structure fields one after the other (seems
1516 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1517 program, let it work on linux/gcc/little32. portability todo.
1519 2004-07-08 Martin Quinson
1520 - import and improve remote compilation support from FAST
1521 - make sure make check works on half a dozen of machines out there
1523 2004-07-07 Martin Quinson
1524 Let's say it's version 0.0.3 ;)
1525 - Implement conversions (yuhu!)
1526 - Let it work on solaris (beside conversion, of course)
1527 - Stupid me, using rand() to generate the conversion datatests in not wise.
1529 2004-07-06 Martin Quinson
1530 - Let make dist work, since I'm gonna need it to compile on remote hosts
1531 - Let Tests/datadesc_usage write the architecture on which the file was
1532 generated as first byte.
1533 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1534 architecture descriptions.
1535 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1536 on those architectures.
1537 - Optimization: send/recv array of scalar in one shoot
1539 2004-07-05 Martin Quinson
1540 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1542 - Plug a whole bunch of memleaks
1543 - each process now have to call gras_{init,exit}. One day, their log
1544 settings will be separated
1545 - Continue the code factorisation between SG, RL and common in Transport.
1547 2004-07-04 Martin Quinson
1549 - Redistribution between SG and RL.
1550 We wanna have to accept in SG, so move accepted related parts of RL in
1551 the common part. (more precisely, the dynar of all known sockets is no
1552 more a static in transport.c, but part of the process_data)
1554 [gras_stub_generator]
1555 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1557 2004-07-03 Martin Quinson
1558 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1559 them all up in one shot)
1561 - Fix the ugly case of reference to dynamic array.
1562 - New (semi-public) function gras_datadesc_size to allow the messaging
1563 layer to malloc the needed space for the buffer.
1565 - gras_socket_close now expect the socket to close (and not its address to
1566 put NULL in it after it). This is because the socket passed to handlers
1567 is one of their argument (=> not writable).
1569 - propagate the interface cleanup from last week in datadesc, ie remove a
1570 superfluous level of indirection. User pass adress of variable
1571 containing data (both when sending and receiving), and not of a variable
1572 being a pointer to the data. Let's say that I like it better ;)
1573 The price for that is constructs like "int msg=*(int*)payload" in
1574 handlers, but it's a fine price, IMHO.
1576 - Let it work in RL (yuhu)
1578 2004-06-21 Martin Quinson
1580 - porting SG plugin and SG select to new standards (works almost).
1581 - plug memleaks and fix bugs around.
1584 - cleanup the prototype of data recv and force users to specify when they
1585 want to handle references to objects. Test case working even for cycles.
1586 - plug memleaks. Valgrind is perfectly ok with this.
1588 2004-06-12 Martin Quinson
1590 - cleanup the separation between plugin and main code in plugin creation
1592 2004-06-11 Martin Quinson
1594 - Reput hook for raw sockets, needed for BW experiments
1595 - kill a few lines of dead code
1596 [Data description] Interface cleanup
1597 - gras_datadesc_by_name returns the searched type or NULL.
1598 That way, no variable is needed to use a type desc once, which makes
1600 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1601 two parameters were strings (field name, type name), leading to
1603 [Dicos] Interface cleanup
1604 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1605 This is consistant with the dynar API.
1607 2004-04-21 Martin Quinson
1609 - Porting to new standards.
1611 - interface cleanup.
1612 There is no bag anymore, no need to take extra provision to mask the
1613 pointers behind "ID".
1614 Better splitup of functions between files create/exchange/convert.
1615 This is still a bit artificial since convert and receive are so
1616 interleaved, but anyway.
1618 - add a queued message list to procdata (the ones not matching criteria
1620 - factorize some more code between SG and RL wrt procdata
1622 - use gras_exit in example to track memleaks
1623 - get rid of gs_example now that GS is properly integrated into gras
1624 - update run_test to integrate the lastest tests (datadesc)
1626 - rename WARNINGn macros to WARNn since it prooved error-prone
1628 2004-04-19 Martin Quinson
1630 - register init/exit functions within gras module mechanism
1631 - send/receive function.
1632 Convertion is not implemented, but short-cutted if not needed.
1633 struct/array elements are sent one by one (instead of block-wise), but
1634 nobody really cares (yet). Get a prototype before optimizing.
1635 - tests (using a file socket) for DD send/receive on:
1636 - base types: int, float
1637 - array: fixed size, string (ie ref to dynamic string)
1638 - structure: homogeneous, heterogeneous
1639 - chained list, graph with cycle
1640 Believe it or not, valgrind is not too unhappy with the results. The
1641 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1645 - Bugfix when using a filename explicitely (instead of '-')
1647 2004-04-09 Martin Quinson
1649 - factorize more code between RL and SG in socket creation
1650 - Complete the implementation and tests of:
1652 o file (only in RL, and mainly for debugging)
1654 I lost 3 days to design a portable address resolver, and then decided
1655 that the prototype mainly have to run on my box.
1656 Addressing portability too early may be like optimizing too early :-/
1658 - use gras_init in the Tests instead of the crappy parse_log_opt
1659 (the latter function is removed)
1660 [Conditional execution]
1661 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1662 [Code reorganisation]
1663 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1664 Build examples against the RL library, since there is no way to disable
1665 its creation for now.
1667 For information, the beginning of coding on GRAS was back in june
1668 2003. I guess that every line has been rewritten at least twice since