1 SimGrid (3.3.5-svn) unstable; urgency=low
3 Java Bindings: Various Cleanups
4 * Remove put/get: no need to export deprecated interface in Java
5 Use send/receive instead.
6 * Cleanup the examples and add a README per directory
7 * Remove example autoDestination (that's the only way to go now)
8 * Remove example explicitDestination (was a plain copy of basic)
9 * Make JniException a runtime exception, so that there is no need to
10 declare the fact that you may encounter such a beast. I guess that
11 nobody will ever want to survive such error.
13 * Kill the useless "rate" argument of SD_task_get_execution_time()
14 Everyone used to provide -1 as a value, it was not used, and the
15 semantic of a possible use wasn't even clear.
16 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
17 either as comm costs or compute costs to mean that there is no
18 such thing for that specific task.
20 * In trace replay, allow to have one trace file per process.
21 Give the specific trace file as argument of each process,
22 and call MSG_action_trace_run(NULL)
23 You can still have one merged file for each processes.
24 * Kill the MSG_paje_output() function. It's a noop since 2 years.
26 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
29 SimGrid (3.3.4) stable; urgency=low
31 The "Desktop Grid needs love too" release (also called Xmas release).
34 * Major speedup in the maxmin system solving by using lazy evaluation
35 Instead of solving completely the maxmin system at each iteration,
36 only invalidate (and recompute) the modified parts.
37 This new feature is enabled in default models but you can try to
38 turn it on with "--cfg:maxmin-selective-update=1" for other models.
39 * Cas01 IMproved as default CPU model
40 This CPU model is the same Cas01 model, but it uses the
41 maxmin-selective-update flag and a heap structure to manage
42 actions on SURF kernel.
43 It reduces the complexity to find the next action to finish and,
44 consequently, it's faster than the old Cas01.
45 This is the new default CPU model (Cas01).
46 * Rename the old Cas01 model to Cas01_fullupdate
47 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
48 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
49 * CpuTI (CPU Trace Integration)
50 A new CPU model whose objective is simulate faster when using
51 availability trace files.
52 Instead of using a full featured, over engineered maxmin system for
53 CPU modeling, this model does the pre-integration of traces files
54 to calculate the amount of CPU power available, and so, executes
55 faster than the old CPU models.
56 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
57 * Use LV08 as default network model since it gives better accuracy
58 for small messages and shouldn't change things for big ones.
59 Use --cfg=network_model:CM02 to get the previous behavior.
62 ******************************************
63 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
64 ******************************************
65 * The new CPU model may changes simulations!
66 The point is that events occurring at the exact same timestamp
67 are not scheduled in the same order with the old and new
68 version. This may be enough to completely change the execution
69 of simulations in some cases.
70 * The new network model will change simulations!
71 This new model is more realistic than the previous one, so you
72 should consider redoing your old experiments with this model.
73 Sorry for the inconvenience.
76 * Introduce the supernovae compilation mode
77 When compiled that way, the whole SimGrid (or almost) is put in a
78 single compilation unit and compiled in one shoot.
79 This is to help gcc which has difficulties to inline stuff from one
81 The speedup seem to be above 15%, althrough more tests are needed on
82 amd64 to confirm that gain.
85 * Port of MSG's mailbox on top of SIMIX network
86 The put/get mechanism was greatly simplified on the way.
89 * New SIMIX network module. Provides:
90 - Mailbox: rendez-vous mecanism to find with who you want to speak
91 - Synchronous send/recv: easier and hopefully faster since the
92 logic is handled in the maestro process directly now
93 - Asynchronous send/recv: you dreamt of it? It's here now
94 Too bad that nobody cared enough to propagate the change to MSG.
95 * Add semaphores as SIMIX synchronization mechanism.
98 * new function SD_daxload(char*) to load a DAX file
99 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
100 * Introduce typed tasks. Specify its kind and cost at creation.
101 At scheduling, just give where it should be placed, and the cost
102 for each involved resource is automatically computed.
103 Existing constructors so far (more to come of course):
104 - SD_task_create_comm_e2e() for end-to-end communication
105 - SD_task_create_comp_seq() for sequential computation
106 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
107 * new function SD_task_dump() for debuging display
108 * new function SD_task_dotty(task,FILE*) writing to file the info
109 about the task in dotty format
110 * SD_task_dependency_exists() can now cope with having one of its
111 arguments NULL. If so, it tests whether the other argument has any
113 * Add getters on list of preceding/following tasks:
114 SD_task_get_parents(task) and SD_task_get_children(task)
115 * Add getters on amount of workstations and list:
116 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
117 * Add getter on task kind: SD_task_get_kind(task)
118 * Update the start_time and finish_time of tasks on completion/failure
119 * Bugfix: Remove task from state swags when destroyed
122 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
125 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
126 dynar with one separate thread per value of the dynar.
127 * Change the prototype of xbt_thread_create(), sorry.
128 Added a boolean parameter indicating whether we want to join this
129 thread (used in SG only for now)
130 * Implement xbt_thread_join and xbt_thread_yield in SG also.
133 * GTNetS wrappers should now be usable again (and betterly tested too)
134 * Fix a major regression from 3.2 where the timeout provided to
135 MSG_task_put_with_timeout() was used as absolute time before which
136 the comm should be done.
137 * Start to fix the <cluster> tag.
138 - Internal links should be good now (beside of the loopback, which
139 use the private link instead)
140 - paths to the external world is still rather broken
141 - the <route:multi> tag is just broken. Actually that's brain-dead.
142 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
143 to make it less stupid
144 ** Check your platform with teshsuite/simdag/platforms/flatifier **
145 * Fix a source-level compatibility glitch from 3.2: after defining
146 MSG_USE_DEPRECATED, you can use the old name
147 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
148 * Allow to compile from the SVN with automake 1.11
149 * Fix some problems when using the "start_time" tag in deployment XMLs.
150 * Fix #8569: XBT/synchro.h has redundant declarations
151 * Fix #8563: MSG return values and exceptions
152 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
153 * Integrate patch #8636: Obey DESTDIR when installing documentation.
154 Thanks to Robson Peixoto.
155 * Fix a vicious bug in dictionaries inducing that some elements were
156 not freed on xbt_dict_free()
158 Portability report of this version:
159 * Main portability targets:
160 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
162 - mac leopard on i386
163 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
164 but nothing critical.
165 * Other platforms: windows, AIX and others were not tested for this release
167 Timing report of this version:
168 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
169 scenario parameters). From 8h to a few seconds in desktop grid settings.
170 * Supernovae brings about 25% speedup on i386.
172 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
174 SimGrid (3.3.3) stable; urgency=low
176 The "Need for Speed" release.
178 The timings done to validate the 3.3.2 were faulty.
179 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
181 The problem was a conversion from a manually handled vector to
182 xbt_dynar_t on the critical path.
183 xbt_dynar_foreach calls functions, inducing stack management crap.
185 We inlined these functions and xbt_dynar_foreach is now breath taking.
186 We also inlined xbt_swag_belong on the way.
188 Here are some approximate speedup measurements (on master/slaves
189 simulations lasting between 10s and 20s each):
190 3.3.1 -> 3.3.2: about same performance
191 3.3.2 -> 3.3.3: 40% speedup
192 3.3.1 -> 3.3.3: 40% speedup
193 3.3.1 with inline patch -> 3.3.3: 30% speedup
195 Our reading is that the refactoring which occurred in 3.3.2 made us
196 suffer much more from the xbt_dynar_foreach low performance, but
197 once we solved this, this refactoring proved to be very performance
198 effective. From the 40% speedup, somehow, 10% are due to the
199 inlining and 30% to the refactoring.
201 That's a pitty that gcc cannot inline functions placed in other files
202 alone. We have to choose between:
203 - break the encapsulation (by putting private data structures and
204 accessors in headers files to help gcc)
205 - live with low performance
206 - switch to a decent compiler such as icc (not quite possible).
208 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
210 SimGrid (3.3.2) stable; urgency=low
212 The "Simplicity does not preceed complexity, but follows it" release.
214 The main contributors of this release were (lexical order):
215 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
218 * Extract the routing logic into its own object.
219 (was dupplicated in network.c and workstation_LV07.c;
220 Allows to implement other ways of storing that info)
221 => kill now useless network_card concept
222 - Use dynar to represent routes (instead of void** + int*)
223 - kill link_set (use surf_network_model->resource_set instead)
224 - Add a command-line option to choose the routing schema to use
225 - Add three new models:
226 * Floyd (shortest path computed at initialization)
227 * Dijikstra (shortest path recomputed all the time)
228 * Cached Dijikstra (shortest path computed on need)
229 All these models where contributed by Silas De Munck, and are
230 described in his ICCS09 paper.
232 * Simplify model declaration
233 (less redirections, less function to write when defining a model)
234 - Factorize stuff between models:
237 surf_model_resource_set(model)
238 surf_model_resource_by_name(model, name)
239 - Unify the types of models in s_surf_model_t (using an union)
240 - Embeed fields of common_public directly into s_surf_model_t
241 - Rename model methods:
242 action_free ~> action_unref
243 action_change_state ~> action_state_set
244 action_get_state ~> action_state_get
245 - Change model methods into functions :
246 (model)->common_public->action_use ~> surf_action_ref
248 * Implement a generic resource; use it as ancestor to specific ones
249 (allows to kill duplicated code in models)
250 Drawback: timer command don't need no name nor properties;
251 workstation_CLM03 don't need no properties
252 (but I guess we can live with those few bytes wasted)
254 * Improve the action object model
255 - implement a constructor avoiding dupplicated code about field
256 initialization in generic_action part.
258 * Kill the SDP model: it has an external dependency, is deprecated
259 in flavor of modern lmm models, and didn't compile since a while
262 * Relocation of the context module from XBT to SIMIX.
263 (the context were decoupled from the simix processes, duplicating a lot of code)
264 => a lot of code was factorized
265 - less overhead is introduced during scheduling
266 - simpler API for the context factory
267 - the logic for process creation,destruction and manipulation was simplified
268 * Simplification of the s_smx_process_t data structure.
269 => accesing the simix level data associated to a process is faster now,
270 and the code is a lot more readable.
273 * Implement some more MPI primitives:
274 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
275 -implementation: Bcast: flat or 2-ary tree (default),
278 Allreduce: Reduce then Bcast
279 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
280 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
281 Alltoallv: flat tree, like ompi
283 * Add support for optimized collectives (Bcast is now binomial by default)
284 * Port smpirun and smpicc to OS X
287 * Kill SD_link_get_properties: hard to maintain and makes very little sense
288 Shout out if you used it.
291 * Display the list of still queued messages in SG mode when existing
295 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
296 * Add xbt_graph_node_get_outedges() [Silas De Munck]
297 * Add xbt_str_from_file(FILE*)
298 * Add xbt_dict_get_key achieving a linear reverse search
299 * Remove the context module
301 Portability report of this version:
302 * Main portability targets:
303 - Linux(debian)/x86/context
304 - Linux(debian)/x86/pthreads
305 - Linux(debian)/amd64/context
306 - Linux(debian)/amd64/pthreads
307 On these, we still have the eratic breakages of gras/pmm and
308 amok/saturate_sg reported in previous version. We still think
309 that the tests are the cause of the fault, not the tested code.
311 - Mac OSX Leopard/x86/context
312 Still false negative in tesh autotesting.
313 Smpi still fails, but this time because readlink does not accept -f
314 Everything seems to work properly beside of that.
317 - AIX version 5.3 (only tested contexts this time)
318 Smpi still fails there because mktemp is not installed.
319 Everything seems to work properly beside of that.
321 I managed to compile it for the first time, but several breakages.
322 Won't delay the release for this exotic platform.
324 * Windows: it's still lagging behind. If you want to help, please
327 Timing report of this version:
328 This version seem to be more than 5% faster than 3.3.1 (on linux
329 64bits with contextes). The gain is less than expected, we are
330 investigating this for next release.
332 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
334 SimGrid (3.3.1) stable; urgency=low
337 * Implement a --cfg-help to show existing configuration variables
338 * Build chain do not require doxygen in maintainer mode
341 * fix a bug on struct sizeof computation, which prevented the
342 exchange of arrays of structs in some conditions
343 - added a regression test about this in datadesc_usage
344 * Allow the exchange of 0-long dynamic vectors.
345 - for that, use -1 as indicator of dynamic size instead of 0
346 - This implied to change any size from unsigned long to long,
347 reducing a bit communication abilities, but I guess that with
348 64bits being quite common, this is more than enough.
349 - This also induce a protocol change, thus bumping network protocol
350 version from 0 to 1 (if we have external users, we have to get
351 clean on that point too ;)
352 - added two regression tests about this in datadesc_usage
353 * Be more verbose when propagating local exceptions
354 This helps debugging.
355 * Display the status of simulated processes when receiving SIGINT in
359 * Allow to control the simulation from a trace file.
360 New functions MSG_action_register() and MSG_action_trace_run()
361 The first one allows to associate a function execution to each
362 kind of action while the second one parses a trace file and
363 triggers the corresponding actions within the system.
364 For now, only a toy example is provided in examples/msg/actions
365 * Add an exemple of process migration in examples/msg/migration
366 * Fix a bug in task exchange which broke MSG_task_get_sender()
367 Add a teshsuite regression test for that.
368 [Bug: if MSG_task_get_sender() is called after sender exit,
370 * Fix a bug which prevented suspend/resume to work properly
371 * Display the status of simulated processes when receiving SIGINT
372 This fixes a regression of v3.3. due to the introduction of SIMIX
373 * Bug fixing in failure management:
374 - trace could not start by a failure at time 0
375 - failure during communications were not working
378 * Add SIMIX_process_set_name() to change the name of the current
379 process in the log messages.
380 * Store smx_hosts in a dict since we only retrieve them by name
381 * Move the configuration infrastructure to surf
384 * Move the configuration infrastructure to surf
387 * Massive internal cleanups:
388 - Store internal structures on processes instead of hosts (allows
389 to have more than one process per host, in addition of being more
391 - Cleanup the initialization/finalization process
392 - Kill a whole bunch of unneeded synchronization:
393 processes run in exclusive manner within the simulator
394 - Move queues from global tables to process data fields
396 - now accept -platform and -hostfile arguments
397 - Pass the right rank value to processes according to the hostfile
398 * Compile the examples by default, and use them as regression tests
399 * Implement MPI_Wtime()
400 * Change the reference speed to a command line option
403 * TCP_gamma can now be specified as command line option using
404 --cfg=TCP_gamma:10000000.0
405 * Change the --surf-path cmd line option into --cfg=path:
408 * Also include strbuff from xbt.h public header
409 * xbt_ex_display(): do not free the exception after displaying
410 This allows to do more with the given exception afterward.
411 Users should call xbt_ex_free() themselves.
415 Portability report of this version:
416 * Main portability targets:
417 - Linux(debian)/x86/context
418 - Linux(debian)/x86/pthreads
419 - Linux(debian)/amd64/context
420 - Linux(debian)/amd64/pthreads
421 These targets fail about 1/10 of times on gras/pmm, but we believe
422 that this is because of the test, not because of simgrid.
423 amok/saturate_sg fails even more rarely, and the test may not be
426 - Mac OSX Leopard/x86/context
427 The test suite still spits tons of errors because some obscure
428 force prevents us from removing the temporary directories
429 arguing that they still contain some metadata I've never heard of.
430 Smpi fails because seq is not installed.
431 Everything seems to work properly beside of that.
434 - AIX version 5.3 (both contexts and pthread)
435 Smpi still fails there because mktemp is not installed.
436 XML inclusions seems rosty on AIX.
438 * Windows: it's still lagging behind. If you want to help, please
441 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
443 SimGrid (3.3) stable; urgency=high
447 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
450 * Introduce the SIMIX module: factorize code between MSG and GRAS.
453 Until now, GRAS were using MSG as an interface to SURF. It was
454 quite difficult because both interface have several differences
455 (MSG channels vs GRAS sockets were the most notable point).
457 This also opens the gate to SMPI (which should occur soon) and speed
458 up simulations by to 40% (even if it were not the main goal).
460 **************************************
461 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
462 **************************************
463 The point is that events occuring at the exact same timestamp are
464 not scheduled in the same order with the old and new version. This
465 may be enough to completely change the execution of simulations in
466 some cases. Sorry for the inconvenience.
468 * Cleanup and upgrade the XML format to push further scalability
469 issues (check http://hal.inria.fr/inria-00256883/ for more info)
471 * Improve the testing infrastructure with tesh. Now a very large part of
472 the code is tested not only by being run but also by checking that the
473 output match an expected output [Mt].
475 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
476 is really less memory-demanding, which should allow you to use
477 larger files in SimGrid [AL].
479 * Inform valgrind about our contextes, so that it becomes usable
480 with the default (and more effecient) version of SimGrid
481 [contributed by Sékou Diakite, many thanks]
484 * Introduce a listener thread in charge of receiving incomming
485 messages from the network. It allows to overlap communication and
486 computation but most notably, it removes some stupid deadlocks due
487 to the fact that so far, a process could not send and receive at
488 the same time. This made most non trivial communication schema
490 * Convert the PIDs from long int to int to match the MSG ones (and
492 * New function: gras_agent_spawn() to launch a new process on
493 current host. Only working in simulation for now. [Mt]
494 * New function: gras_os_hostport() returning a constant form (ie,
495 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
498 * Make the backtrace of exceptions more human readable [Mt]
499 * New module: xbt/str [Mt]
500 a ton of string utility functions (split, join, printf to a newly
501 allocated buffer, trim, etc)
502 * New module: xbt/hash [Mt]
503 SHA1 hashing algorithm (more to come if needed)
504 * New module: xbt/synchro [Mt]
505 synchronization tools (mutex and conditions) working the same way
506 in simulation and in real life (mainly useful for GRAS, but not
508 * New module: xbt/queue [Mt]
509 classical producer/consumer synchronization scheme
510 * xbt_dynar_new_sync() creates a synchronized dynar. All access
511 (using the classical functions will get serialized) [Mt]
512 * Make dictionary internal table dynamic. No need to specify its size
513 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
515 * Make sure the log channels are organized as a tree under windows
516 (because of ANSI C compatibility issue, any channel were child of
520 * Cleaned many thing in surf and fixed a few bugs [AL].
521 * Add a nice command line configuration mechanism to compose models [AL].
522 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
523 the previous one (KCCFLN05). It relies on something that looks like
524 a max-min sharing mechanism but cannot be written as such. A new solver
525 was thus designed [AL].
526 * Add a new solver to lmm. Based on Lagrange optimization and
527 gradient-based descent, it enables to efficiently maximise systems s.a
529 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
531 This solver enables to propose two new network models for TCP Reno and
532 TCP Vegas based on Low's work. These models still need to be fully
533 tested though [Pedro Velho].
536 * Bug fix in SD_simulate. Now the time bound given as argument is
538 * Use the new parallel task model (ptask_L07) as default.
539 * Use the SURF command line configuration mechanism.
540 * 0-size tasks (for synchronization) should now work.
542 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
544 SimGrid (3.2) stable; urgency=high
548 We still experience issues on this platform, but we believe that at
551 GRAS API BREAKAGE (for simplification purpose, sorry):
552 * the gras_msgtype_by_name is not used anymore. Instead of
553 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
554 you can write (and must)
555 gras_msg_send(toserver, "request", &request);
556 - If you still want to pass a gras_msgtype_t to the function (to cache
557 the type and avoid the lookup time), use the gras_msg_send_() variant.
558 - Impacted functions:
559 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
560 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
561 * The callbacks are now expected to return 0 when everything went well
562 (just like the main() function)
564 GRAS new features and improvements:
565 * New module mecanism where user code can use per process globals [Mt]
566 This is similar to gras_userdata_*() functions, but for libraries. It
567 factorize some code developped over and over in the examples and AMOK.
568 It has still to be documented and used (only amok/peermanagement is
570 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
571 loss when they were small enough to fit into the buffer and sent quickly
572 enough so that they can all get received in one shoot.
573 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
574 if not found. Use the *_or_null() variant for the old semantic.
575 * In gras_msg_handle, do not discard messages without callback.
576 They are probably messages to be explicitly awaited later (ie, proofs of
577 mis-synchronization in userland since they are sent before being awaited)
579 * gras_socket_meas_send/recv: semantic changed!
580 The numerical arguments used to be (1) the total amount of data to send
581 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
582 messages. This was need for the fool willing to send more than MAXINT
583 bytes on quite fat pipes.
586 * Do really rename the hostmanagement module to peermanagement. [Mt]
587 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
588 but this is rather new and this was documented in the module
589 documentation (poor excuses, I admit)
590 * Bandwidth measurement semantic changed! This follows the changes to
591 gras_socket_meas_send/recv explained above.
594 * A sequential mode has been added to the workstations. When a workstation
595 is in sequential mode, it can execute only one task, and the other tasks
596 are waiting in a FIFO. [Christophe Thiery]
599 * The KCCFLN05 workstation model now handles parallel tasks. It is the
600 model for SIMDAG. [Christophe Thiery]
601 * Bug fix in the maxmin solver: Some values were close to 0 instead of
602 equal to 0, which caused some bad behaviors in
603 saturated_constraint_set_update. I now use a threshold mechanism like in
607 * When running manually src/testall, you select specific units [Mt]
608 testall is the result of our cunit mecanism, and should replace all
609 the scripty thingy around since bash don't run easily on billware.
611 * A mallocator system has been added. [Christophe Thiery]
612 Mallocators allow you to recycle your unused objects instead of freeing them
613 and allocating new ones.
615 Documentation update:
616 * FAQ reworking + New FAQs:
617 - "Valgrind spits tons of errors!" [Mt]
618 - "How to repport bugs" [Mt]
619 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
620 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
621 - Communication time measurement within MSG [AL]
622 - I experience weird communication times when I change the latency [AL]
625 - an introduction to the framework and to the used communication model
626 - an initiatic tour introducing the most proheminent features:
628 . Lesson 0: Installing GRAS
629 . Lesson 1: Setting up your own project
630 o Part 2: Message passing
631 . Lesson 2: Exchanging simple messages
632 . Lesson 3: Passing arguments to the processes (in SG)
633 . Lesson 4: Attaching callbacks to messages
634 . Lesson 5: Using globals in processes
635 . Lesson 6: Logging informations properly
636 . Lesson 7: Using internal timers
637 . Lesson 8: Handling errors through exceptions
638 . Lesson 9: Exchanging simple data
639 . Lesson 10: Remote Procedure Calling (RPC)
640 . Lesson 11: Explicitely waiting for messages
641 . Recapping of message passing features in GRAS
642 - A HOWTO section containing:
643 o HOWTO design a GRAS application
644 More are due, of course. They will come latter. In the meanwhile, you can
645 check the examples which are still here.
647 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
649 SimGrid (3.1) stable; urgency=high
653 There was a stack corruption somewhere, visible only when optimizing
654 with these versions. [Vince]
657 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
658 module that enabled to play with Directed Acyclic Graphs. It is built
659 directly on top of SURF and provides an API rather close to the old
660 SG. Some old codes using SG are currently under rewrite to check that
661 all needful functions are provided. [Christophe Thiery]
664 * Complete rewrite of the KCCFLN05 workstation model. It is now an
665 extension of the classical CLM03 model that gracefully handles
666 failures. This is now the default model for MSG and GRAS. It doesn't
667 handle parallel tasks yet though. [AL]
668 * Bug fix: Weights were not correctly set in the network part.
669 WARNING: This may have resulted in incorrect results with simulations
670 where there are more than one flow on a given link. [AL]
673 * After a (long ?) discussion on simgrid-devel, we have decided that the
674 convention we had on units was stupid. That is why it has been decided
675 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
676 WARNING : This means that all previous platform files will not work as
677 such with this version! A warning is issued to ask users to update
679 A conversion script can be found in the contrib module of the CVS, under
680 the name contrib/platform_generation/surfxml_update.pl [MQ]
683 * Bug fix: Processes were started in reverse order, wrt deployment file.
684 WARNING: if your code relies on this bug, please fix it. [AL]
685 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
686 being executed on two different locations. [AL]
687 * Bug fix: Failures are now better supported thanks to Derrick's tests
688 (there was many failure situations I hadn't thought of and that weren't
689 correctly handled). [AL]
690 * New function: MSG_host_is_avail indicates you whether a given m_host_t
694 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
695 Exception occurring on server-side are propagated back to client (!).
697 API CHANGE: the callback changed their prototype. Change:
698 int my_handler(gras_socket_t expeditor, void *payload_data) {
700 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
701 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
703 * New! function: gras_msg_handleall to deal with all messages arriving
704 within a given period.
705 * New! function: gras_socket_server_range to get a server socket in a
706 range of port numbers (ease to avoid port number conflicts) [MQ]
707 * New! gras processes display their backtrace when they get a SIGUSR1
708 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
709 Sweet to debug RL processes [MQ]
713 - Do not force experiment sizes to be expressed in kb, or it becomes
714 impossible to measure the latency this way (needs one byte-long tests)
715 WARNING: this changes the amok_bw_* function semantic. [MQ]
716 - Implements the link saturation stuff. [MQ]
717 * Peer management module:
718 New! module factorizing code that we wrote over and over [MQ].
721 * New module: cunit (my jUnit implementation in ansi C) [MQ]
722 - Test units are placed directly into the library code, they get extracted
723 automatically and placed into the src/testall binary.
724 - Convert most of the XBT tests to this system.
725 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
726 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
727 live in the same C code. This required to change a little bit the API
728 of surfxml parsing but shouldn't be an issue for end-users. [AL]
729 * New module: sparse graph structure with basic algorithms (this is work
730 in progress and the API is not considered to be frozen yet). [AL]
731 * Display more information on backtraces: source line & function names are
732 now displayed just like valgrind does (rely on addr2line tool) [MQ]
733 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
734 * Reworked a little bit some #include statements to load only required
735 headers. Some user code that relied on SimGrid to include stdlib or
736 stdio may need to include it by themselves. [AL]
737 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
739 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
740 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
741 but I guess that almost nobody used those parts. [MQ]
743 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
745 SimGrid (3.0.1) stable; urgency=low
748 * Unfortunately, I had missed 5 misnamed functions:
749 xbt_fifo_item_t xbt_fifo_newitem(void);
750 void xbt_fifo_freeitem(xbt_fifo_item_t);
751 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
752 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
753 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
754 They're now deprecated. Please use their new versions:
755 xbt_fifo_item_t xbt_fifo_new_item(void);
756 void xbt_fifo_free_item(xbt_fifo_item_t);
757 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
758 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
759 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
761 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
762 * Documentation: xbt_log module unmercifully reworked [MQ]
763 * Bugfix: there was a problem with the ending of contexts with
764 the pthread backend. It caused some weird deadlock or behavior
765 depending on the pthread implementation. [AL]
766 * Bugfix: get the exceptions raised in the simulator repport where
767 and why they come from when they are not catched in time [AL, MQ]
770 * Bugfix: Do repport the error when two non-connected hosts try to
771 exchange data (Thanks to Flavien for stumbling into this one) [AL]
774 * Add additionnal checkings on communications. Assert that two
775 communicating hosts are connected by a set of links... [AL]
778 * Add additionnal checkings on channel values in communication [AL]
779 * New: MSG_task_get_source to see on which host a task was generated [HC]
780 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
781 the number of tasks waiting to be received on channel and sent
783 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
784 waits for the first task coming from a given host.. [AL]
786 GRAS new functionnalities: [MQ]
787 * Enhance the parsing macro to allow the size of multidimentional objects
788 to be given thru annotations.
789 * New example (and documentation): Matrix Multiplication a la RPC
790 (as when I was young!) and fix a bunch of bugs found on the way.
792 GRAS performance improvements: [MQ]
794 * Reduce the amount of cbps creation/destruction by making it static to
795 datadesc_send/recv() and using a (newly created) cbps_reset (based on
798 * Change libdata to a set so that we can search for stuff by ID (and thus
799 reduce the insane amount of dict lookups)
802 * Actually implement gras_datadesc_copy() so that we don't have to mimick
803 RL communication on top of SG since it's so uneffective.
804 It may also be used for inter-thread communication in RL, one day.
805 * Use gras_datadesc_copy() to exchange messages on top of SG
807 - improve message exchange performance on top of SG
808 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
809 * Don't exchange on the network the size of the used part of buffer,
810 instead, specify the possible buffer size to read().
812 - reduces the amount of read/write calls (one pair per exchange)
813 - reduces the amount of exchanged data (the size)
814 - allows to retrieve all arrived data on receiver side, if we don't need
815 it right now (subsequent read will peek the buffer)
816 - allows the receiver to proceed with the begining of the stream before
817 everything is arrived
818 - make it possible to build an iov transport (using readv/writev)
820 - take care of the data with non-stable storage (like stacked data),
822 * If possible, TCP send uses vector I/O (when writev() is here)
823 - Don't use it for receive since we send data sizes and data on the
824 same stream, so we wouldn't be able to chain large amount of chunks
825 before having to flush the stuff to read the size.
826 * Rework the transport plugin mecanism to simplify it and reduce the
827 amount of pointer dereferencement when searching for the right function
830 * I guess that now, we do almost as few system calls as possible while
831 doing as few data copy as possible.
833 To improve it further, we could try to send all the sizes first and then
834 all the data (to use iov on receiving size), but it's only a partial
835 solution: when you have 2 dimensional data, the sizes of the second
836 dimension is data of the first dimension, so you need 3 streams.
838 I'm not sure the potential performance gains justify the coding burden.
840 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
842 SimGrid (3.00) stable; urgency=high
845 * New! Give the possibility to hijack the surf parser and thus bypass
846 MSG_create_environment and MSG_launch_application. Have a look at
847 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
849 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
851 SimGrid (2.96) unstable; urgency=low
856 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
857 This deprecates the xbt_error_t mecanisms.
858 It modifies (simplifies) all XBT and GRAS API.
859 MSG API keeps unchanged (exceptions raised by XBT are catched from
860 within MSG and masked with existing error handling facilities)
863 * New! Add a FATPIPE model. [AL]
864 * New! Add a parallel task model. [AL]
865 * New! Add automatically a loopback interface (in the default
866 network model) if none was precised.
869 * Bugfix: MSG_process_resume now works with the current running process.
871 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
872 * Modification of MSG_task_get_compute_duration. Once a task has been
873 processed, the value returned by this function is now equal to 0. [AL]
874 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
875 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
876 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
877 of a m_task. For now, MSG_task_cancel only works with computation
879 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
880 of the processor (in Mflop/s) regardless of the current load on the
882 * API Change: use proper naming convention for MSG_getClock and
883 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
885 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
886 This function changes the priority of a computation task. This priority
887 doesn't affect the transfer rate. A priority of 2 will make a task
888 receive two times more cpu power than the other ones. This function
889 has been added to suit the needs of Nguyen The Loc and hasn't been that
890 much tested yet. So if it fails, please report it and send me your code.
892 * API Change: removed all functions and types that were marked "deprecated"
893 since many months. Renamed MSG_global_init_args to MSG_global_init.
895 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
897 SimGrid (2.95) unstable; urgency=low
900 * Steal some nice code to GNU pth to fix context detection and usage [AL]
901 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
902 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
905 * Bugfix: Allow absolute paths for platform description files [MQ]
906 * Bugfix: do free the variables after use. Leads to drastic performance
908 * Implement max_duration (ie, timeouts) on resources [AL]
911 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
913 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
914 the needed support (timeouts) [AL]
917 * Implement measurement sockets. You can now get the bandwidth between two
918 hosts thanks to AMOK (see below). [MQ]
919 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
920 dynar over the network (yeah) [MQ]
921 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
922 * simple/static token-ring example. [Alexandre Colucci and MQ]
923 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
927 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
928 type parsing. "short" and "long" modifiers were also ignored; other
929 modifier (such as reference level) are still ignored. [MQ]
930 * Embeed the buffer size within the buffer itself on SG. [MQ]
931 That way, send() are atomic and cannot get intermixed anymore (at least
932 the ones which are less than 100k; bigger messages still have the issue)
933 * Array size pushed by the field, not by the field type (or each
934 and every long int will push stuff to the cbps) [MQ]
935 * use select() to sleep since it allows to portably sleep less than one
938 GRAS (minor cleanups)
939 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
942 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
943 * type category 'ignored' killed as it was never used and were difficult
945 * whether a type can cycle or not is now a flag, leaving room for more
946 flags (as "ignored", if we feel the need one day ;)
947 * Rename raw sockets to measurement sockets since "raw" has another
948 meaning in networking community.
951 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
952 and offers features not belonging to GRAS but that most applications
953 need. One day, it may be a set of plugins loadable at runtime.
954 * New module: bandwidth
955 bandwidth measurement between arbitrary nodes running this module. [MQ]
957 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
959 SimGrid (2.94) unstable; urgency=low
961 The first beta release of SimGrid 3 !
965 * Update the main page and the FAQ. Adding references to gforge.
968 * Add a gras_os_getpid function.
971 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
972 * Extend the logs so that they also print PID, hostname, date, ... if
974 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
975 and kill the old version which were in testsuite/
976 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
980 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
981 for users. Instead, each programming environment (GRAS, MSG,...) use it
982 when needed to provide such a feature to users.
983 Don't shortcut the mecanism or you will also shortcut the virtualization
984 you need on the simulator.
988 * Cleanups in configury with regard to compile optimization/warning flags.
989 Also add -fno-loop-optimize to any powerpc since it's the optimization
990 killing gcc (< 3.4.0).
991 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
992 needed by MSG examples complications
993 * Borrow configury beautifications from PHP
996 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
997 warning (thanks loris for stumbling into this one).
998 * Bugfix: stop loading private headers (gras_config.h) from the public
1002 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1003 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1004 RL would imply a DNS resolver, which is *hard* to do in a portable way
1005 (and therefore delayed).
1006 * Implement a real timer mecanism and use it in timing macros. This allows
1007 to avoid rounding errors and get a 0.000005 sec precision in timing
1008 macros. While I was at it, various cleanups:
1009 - allow to declare more than one timed section per file (fix a stupid bug)
1010 - move some private declaration to the right place
1011 - merge conditional execution and timing macros into emulation module
1012 - document the module
1013 - make sure the module cleanups its mess on gras_exit
1014 * Documentation improvements:
1015 - (new) how to compile applications using GRAS
1016 - (new) emulation support (timing macros)
1018 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1020 SimGrid (2.93) unstable; urgency=low
1022 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1025 - Use Paje properly where used. Still to be sanitized properly.
1026 - Portability fix: Add an implementation of the contexts using pthread
1030 - Add xbt_procname(): returns the name of the current process.
1031 Use it to show the current process's name in all logging.
1033 - fix detection of older flex version and the reaction, since we do
1034 depend on modern ones (we use lex_destroy)
1035 - Better separation of SG and RL in the libs: remove all simulation code
1036 from libgras. As a result, this lib is now only 200k when stripped.
1037 Some of the xbt modules may also be duplicated (two sets and such) and
1038 should be cleaned/killed before SG3.
1039 - Insist on using xlC on AIX because of weird problems involving gcc there.
1040 - Cleanup the make remote stuff. This is now done by scripts
1041 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1042 mainly for our private use, but we're working on changing them to user
1045 - Bugfix: flush the socket on close only if there is some *output*.
1046 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1048 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1051 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1053 SimGrid (2.92) unstable; urgency=low
1055 Alpha 3 on the path to SimGrid 3
1059 - New! First try of benchmarking macros.
1060 - New! First try so that gras_stub_generator generate deployment and
1061 remote compilation helpers.
1063 - Bugfix: Initialization fix in msg_test.
1067 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1069 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1071 - API Change: gras_os_sleep to take the amount of seconds as a double.
1072 Accepting an int was error prone since it was the only location where
1073 seconds were coded as such. It leaded to damn rounding errors.
1074 - Bugfix: Hard to belive that timers ever worked before this.
1076 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1078 SimGrid (2.91) unstable; urgency=low
1080 Alpha 2 on the path to SimGrid 3
1084 - Bug fix in the lmm_solver.
1086 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1087 through the function MSG_paje_output.
1088 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1089 - It is possible to bound the rate of a communication in MSG with
1090 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1091 to write it in the changelog).
1092 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1097 - Various cleanups to the autotools stuff
1098 - Begin to move Gras examples to examples/gras/
1099 - Let make distcheck work again (yeah!)
1101 - documentation overhauled using doxygen.
1102 gtk-doc-tools is dead in SimGrid now.
1103 - Automatically extract all existing logging categories, and add the list
1104 to the documentation (long standing one, to say the less)
1106 - Cleanup the known architecture table. Reorder the entries to group what
1107 should be, and use a more consistent naming scheme.
1108 (some of the test dataset are still to be regenerated)
1109 - New! Allow library to register globals on each process just as userdata
1111 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1112 lookup time (for now).
1113 Use it in msg and trp.
1114 This cleans a lot the internals and helps enforcing privacy of the
1115 headers between the gras components.
1116 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1117 - Bugfix: gras_os_time was delirious in RL.
1118 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1120 - Reenable GRAS now that it works.
1122 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1124 SimGrid (2.90) unstable; urgency=low
1126 Alpha 1 on the path to SimGrid 3
1128 * It is a long time since the last release of SimGrid. I'm sorry about
1129 that but as I had told you, I was rewriting a lot of things. I apologize
1130 to those who had been reporting bugs to me and that I had not answered.
1131 If your bug is still in the new version, please tell me. Here is a
1132 summary of the main changes.
1134 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1135 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1136 portability, speed, and a lot more but you'll figure it out later.
1137 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1138 I wanted to release the new MSG as soon as possible and I have
1139 broken GRAS, which is the reason why, for now, only MSG is fully
1140 functional. A laconic description of these projects is available
1141 in the documentation.
1143 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1144 kernel optimized for our needs (called SURF but only the developers
1145 should use it). Hence, MSG is now roughly 30 times faster and I think
1146 that by rewriting a little bit MSG, I could event speed it up a little
1147 bit more. Beside the gain in speed, it is also much easier to encode a
1148 new platform model with SURF than it was with SG. More to come...
1150 * REVOLUTION 3: I have tried to change a little as possible the API of
1151 MSG but a few things really had to disappear. The main differences
1152 with the previous version are :
1153 1) no more m_links_t and the corresponding functions. Platforms are
1154 directly read from a XML description and cannot be hard-coded
1155 anymore. The same format is used for application deployment
1156 description. The new format is described in the documentation.
1157 Have a look in tools/platform_generation. There is a tiny script
1158 that converts from the old platform format to the new one. Concerning
1159 the application deployment format, parsing the old one is tricky.
1160 I think most of you should however be able to convert your files. If
1161 it is really an issue, I can write a C code that does the conversion.
1163 2) the toolbox tbx does not exist anymore. We now have a library
1164 with much more data-structures but without the hash-tables (we have
1165 dictionaries that are much faster).
1167 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1169 *****************************************************************************
1170 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1171 * don't feel like converting the oldies. (Mt) *
1172 *****************************************************************************
1175 Version 2.90: "the long awaited one"
1176 - Finished rewriting and debugging MSG. Rewrote the documentation.
1177 - disable GRAS for now since it needs to be ported to the newest SG
1180 - Finish the port to windows (using mingw32 for cross-compile)
1183 - Main loop and datastructures of SURF. A cpu resource object is
1184 functional. Surf can thus be used to create cpu's with variable
1185 performance on which you can execute some actions.
1187 2004-11-15 Martin Quinson
1188 - Port to ARM. Simply added the alignment and size descriptions. Should
1189 work, but the ARM machines are so slow that I didn't had the opportunity
1190 to 'make check' over there yet.
1192 2004-11-15 Arnaud Legrand
1193 - Trace manager now written. It uses a heap structure and is therefore
1194 expected to be efficient. It may however be speeded up (particularly
1195 when many events occur at the same date) by using red and black
1196 trees. One day maybe...
1197 - Max-min linear system solver written. It uses a sparse matrix
1198 structure taking advantage of its expected use. Most operations are
1199 O(1) and free/calloc are called as few as possible. The computation of
1200 the minimum could however be improved by using a red and black tree
1203 2004-11-03 Arnaud Legrand
1204 - Rename every gras_* function that was in xbt/ to its xbt_
1206 - Add a heap and a doubly-linked list to xbt
1207 - Added a dichotomy to the dictionaries. make check works as well before
1208 so I assume that the patch is correct. I do not know however if things
1209 run effectively faster than before now. :)
1211 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1212 SimGrid, and the version number is bumped to 2.x
1214 2004-10-29 Martin Quinson
1215 - Introduction of the remote errors.
1216 They are the result of a RMI/RPC on the remote machine.
1217 ErrCodes being scalar values, you can't get the host on which those
1218 errors did happen. Extending the error mechanism as in Gnome is possible.
1219 No idea yet whether it is a good idea.
1221 2004-10-28 Martin Quinson
1222 - Interface revolution: the Starred Structure Eradication.
1223 I used to do typedef struct {} toto_t; and then handle *toto_t.
1224 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1225 - ???_t is a valid type (builded with typedef)
1226 - s_toto_t is a structure (access to fields with .)
1227 - s_toto is a structure needing 'struct' keyword to be used
1228 - e_toto_t is an enum
1229 - toto_t is an 'object' (struct*)
1231 typedef struct s_toto {} s_toto_t, *toto_t;
1232 typedef enum {} e_toto_t;
1233 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1234 s_toto_t) is private.
1236 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1237 so that it can set the variable to NULL. It was so for dicts and sets,
1238 it changed for dynars.
1240 - Fix a bunch of memleaks in dict_remove
1241 - Fix a bug in sg/server_socket opening: it failed all the time.
1243 2004-10-07 Martin Quinson
1244 - Speed up dynar lookup operation a bit.
1246 gras_dynar_get is dead.
1248 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1249 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1250 (which gives you the address of the stored data).
1252 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1253 copy of the data using an affectation to do the job and not a memcpy.
1255 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1256 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1258 It does not really speedup the dynar test because they are
1259 setting elements all the time (and look them seldom). But the dict does
1260 far more lookup than setting.
1262 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1264 2004-10-05 Martin Quinson
1265 - Allow to (en/dis)able the cycle detection at run time.
1267 Whether we should check for cycle or not is now a property of each
1268 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1269 datadesc_cycle_unset allow to remove this property when previously set.
1271 Note that the cycle detection is off by default since it impacts the
1272 performance. Watch the data you feed GRAS with ;)
1274 This property is hereditary. Any element embedded in a structure having it
1275 set have it set for the time of this data exchange.
1277 You should set it both on sender and receiver side. If you don't set it on
1278 sender side, it will enter an endless loop. If you forget on receiver
1279 side, the cycles won't be recreated after communication.
1281 - Header reorganization.
1282 Kill gras_private.h, each submodule must load the headers it needs.
1284 2004-10-04 Martin Quinson
1285 - Interface revolution: do not try to survive to malloc failure.
1287 Now, gras_malloc and friends call gras_abort() on failure.
1288 As a conclusion, malloc_error is not a valid error anymore, and all
1289 functions for which it was the only gras_error_t return value are
1290 changed. They now return void, or there result directly.
1291 This simplify the API a lot.
1293 2004-09-29 Martin Quinson
1294 - Re-enable raw sockets.
1295 Created by gras_socket_{client,server}_ext;
1296 Used with gras_raw_{send,recv}
1299 It should allow to kill the last bits of gras first version soon.
1301 This is not completely satisfactory yet (duplicate code with
1302 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1305 - Simplify transport plugin (internal) interface by not passing any
1306 argument to _server and _client, but embedding them in the socket
1309 2004-09-28 Martin Quinson
1310 - Finish the port to AIX.
1311 autoconf was my problem (segfault within the malloc replacement
1312 function. No idea why)
1314 2004-09-16 Martin Quinson
1315 - Fix some size_t madness on 64bit architectures.
1317 2004-09-08 Martin Quinson
1318 - Reduce the number of system headers loaded, overload some more system
1319 calls (such as malloc to cast the result of the system one, and work
1321 - Fix and reintroduce the config support
1323 2004-09-07 Martin Quinson
1324 - Source code reorganization to allow Arnaud to surf all over there.
1325 - Allow to document the logging categories.
1326 - Remove all uppercase from logging categories and useless cleanup in names.
1328 2004-08-18 Martin Quinson
1329 Version 0.6.2 (protocol not changed; API changed)
1330 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1331 gras_error_t), and NULL when not found. Functions expecting a msgtype
1332 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1333 hopefully usefull message.
1334 - Portability to prehistoric sparcs again
1336 2004-08-17 Martin Quinson
1337 Version 0.6.1 (protocol not changed; ABI not changed)
1338 - prealloc some buffers to speed things up
1340 2004-08-11 Martin Quinson
1341 Version 0.6 (protocol not changed; ABI expended)
1342 - The parsing macro can deal with the references, provided that you add
1343 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1345 2004-08-09 Martin Quinson
1346 Version 0.5 (protocol not changed; ABI changed)
1347 - Allow to off turn the cycle detection code in data exchange at
1348 compilation time. It should be at run time, but I'm short of time (and
1349 the config stuff is still broken). That way, we keep dict out of the
1350 critical path, which is good because the performance is poor:
1351 - search not dichotomial yet
1352 - dynar give no way to access their content and memcpy everytime
1353 - In composed data description (struct, ref and so on), stop foolness of
1354 keeping the subtype's ID, but store the type itself. This keeps sets out
1355 of the critical path, which is good since they rely on dynar and
1356 dictionnaries. The only loose of that is that we cannot detect the
1357 redeclaration of a structure/union with another content (but I'm not sure
1358 the code detected well this error before anyway). We still can detect
1359 the redefinition discrepancy for the other types.
1360 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1361 it breaks the code because of type-punning used all over the place).
1362 This breaks on all non-gcc architectures (for now).
1364 All those changes (plus the buffer of last time) allow me to gain 2 order
1365 of magnitude on cruel tests consisting of 800000 array of integers on two
1366 level of a hierarchical structure (200 secondes -> 4 secondes)
1369 - the selector of reference must now return the type it points to, not
1370 the ID of this type.
1372 2004-08-06 Martin Quinson
1373 Version 0.4 (protocol changed; ABI not changed)
1374 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1375 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1376 - Implement a buffer transport, and use it by default (it relies on tcp in
1377 real life and on sg in simulation).
1378 That's a bit hackish since I had a new field to the structure to store
1379 its data without interfering with the subtype ones. Inheritance
1380 is tricky in C. And that's a kind of reverse inheritance with one class
1381 derivating two classes. Or maybe a game with java interfaces. Anyway,
1382 that's damn hard in C (at least).
1383 Moreover, I got tired while trying to ensure plugin separation and
1384 genericity in SG mode. MSG wants me to do weird things, so let's go for
1385 cruel hacks (temporarily of course ;).
1386 See comment in transport_private.h:71
1387 - do not include all the _interface headers in private but in the files
1388 which really need them (to cut the compilation time when they are
1391 2004-07-26 Martin Quinson
1392 Version 0.3 (protocol not changed; ABI changed)
1393 - Major overhault of the datadesc interface to simplify it:
1394 - shorted the function names:
1395 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1396 - add a trivial way to push/pop integers into the cbps without malloc.
1397 This allows to make really generic sub_type description, which simply
1398 pop their size of the stack.
1399 - add a function gras_datadesc_ref_pop_arr() which does what users want
1400 most of the time: Declare a dynamic array (which pops its size of the
1401 stack) and declare a reference to it. Poor name, but anyway.
1402 - kill the post-send callback, add a post-receive one
1404 2004-07-23 Martin Quinson
1405 Version 0.2 (protocol changed; ABI changed)
1406 - add some testing for cpbs in the test cases, and fix some more bugs.
1407 This invalidate again the little64 data file, since I cannot regenerate
1409 - remove an awfull optimization in the logging stuff, allowing me to:
1410 - understand it again
1411 - learn gcc how to check that the argument match the provided format
1412 - fix all errors revealed by gcc after that
1413 - internal keys of dict are not \0 terminated. Deal with it properly in
1414 loggings instead of segfaulting when the user want to see the logs :-/
1416 2004-07-22 Martin Quinson
1417 - Fix some stupid bug preventing cbps (callback postit) from working
1419 2004-07-21 Martin Quinson
1420 - Some documentation cleanups
1421 - remove the useless last argument of msgtype_declare
1422 - rename the Virtu functions to fit into the 'os' namespace
1423 - move headers src/include -> src/include/gras/ and stop fooling with
1424 gras -> . symbolic link
1425 - make distcheck is now successful
1427 2004-07-19 Martin Quinson
1429 - Build shared library also
1430 - Install html doc to the right location
1431 - stop removing maintainer files in make clean
1432 - build tests only on make check
1434 2004-07-13 Martin Quinson
1436 - No major issue in previous version => change versionning schema
1437 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1438 corresponding dataset.
1440 2004-07-11 Martin Quinson
1442 - Get it working with any kind of structure (we can compute the padding
1443 bytes remotely for all the architectures I have access to)
1444 - Implement the structure parsing macro (still not quite robust/complete)
1445 - Improvement to the remote testing toysuite
1447 2004-07-10 Martin Quinson
1448 [autoconf mechanism]
1449 - get ride of a bunch of deprecated macros
1450 - actually run the test for two-compliment, not only compile it :-/
1451 - test whether the structures get packed (and bail out if yes. Damn.
1452 Alignment is a serious matter)
1453 - test whether the structures get compacted (but respecting the alignment
1454 constraints of each types)
1455 - test whether the array fields of structures can straddle alignment boundaries
1457 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1458 (took me 2 hours to find that bug, looking at the wrong place)
1459 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1460 before sending/receiving and are used to compute the offsets of fields
1461 - Given that padding size depend even on compiler options, keep track of
1462 alignment and aligned_size only for the current architecture. Not a big
1463 deal since we send structure fields one after the other (seems
1465 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1466 program, let it work on linux/gcc/little32. portability todo.
1468 2004-07-08 Martin Quinson
1469 - import and improve remote compilation support from FAST
1470 - make sure make check works on half a dozen of machines out there
1472 2004-07-07 Martin Quinson
1473 Let's say it's version 0.0.3 ;)
1474 - Implement conversions (yuhu!)
1475 - Let it work on solaris (beside conversion, of course)
1476 - Stupid me, using rand() to generate the conversion datatests in not wise.
1478 2004-07-06 Martin Quinson
1479 - Let make dist work, since I'm gonna need it to compile on remote hosts
1480 - Let Tests/datadesc_usage write the architecture on which the file was
1481 generated as first byte.
1482 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1483 architecture descriptions.
1484 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1485 on those architectures.
1486 - Optimization: send/recv array of scalar in one shoot
1488 2004-07-05 Martin Quinson
1489 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1491 - Plug a whole bunch of memleaks
1492 - each process now have to call gras_{init,exit}. One day, their log
1493 settings will be separated
1494 - Continue the code factorisation between SG, RL and common in Transport.
1496 2004-07-04 Martin Quinson
1498 - Redistribution between SG and RL.
1499 We wanna have to accept in SG, so move accepted related parts of RL in
1500 the common part. (more precisely, the dynar of all known sockets is no
1501 more a static in transport.c, but part of the process_data)
1503 [gras_stub_generator]
1504 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1506 2004-07-03 Martin Quinson
1507 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1508 them all up in one shot)
1510 - Fix the ugly case of reference to dynamic array.
1511 - New (semi-public) function gras_datadesc_size to allow the messaging
1512 layer to malloc the needed space for the buffer.
1514 - gras_socket_close now expect the socket to close (and not its address to
1515 put NULL in it after it). This is because the socket passed to handlers
1516 is one of their argument (=> not writable).
1518 - propagate the interface cleanup from last week in datadesc, ie remove a
1519 superfluous level of indirection. User pass adress of variable
1520 containing data (both when sending and receiving), and not of a variable
1521 being a pointer to the data. Let's say that I like it better ;)
1522 The price for that is constructs like "int msg=*(int*)payload" in
1523 handlers, but it's a fine price, IMHO.
1525 - Let it work in RL (yuhu)
1527 2004-06-21 Martin Quinson
1529 - porting SG plugin and SG select to new standards (works almost).
1530 - plug memleaks and fix bugs around.
1533 - cleanup the prototype of data recv and force users to specify when they
1534 want to handle references to objects. Test case working even for cycles.
1535 - plug memleaks. Valgrind is perfectly ok with this.
1537 2004-06-12 Martin Quinson
1539 - cleanup the separation between plugin and main code in plugin creation
1541 2004-06-11 Martin Quinson
1543 - Reput hook for raw sockets, needed for BW experiments
1544 - kill a few lines of dead code
1545 [Data description] Interface cleanup
1546 - gras_datadesc_by_name returns the searched type or NULL.
1547 That way, no variable is needed to use a type desc once, which makes
1549 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1550 two parameters were strings (field name, type name), leading to
1552 [Dicos] Interface cleanup
1553 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1554 This is consistant with the dynar API.
1556 2004-04-21 Martin Quinson
1558 - Porting to new standards.
1560 - interface cleanup.
1561 There is no bag anymore, no need to take extra provision to mask the
1562 pointers behind "ID".
1563 Better splitup of functions between files create/exchange/convert.
1564 This is still a bit artificial since convert and receive are so
1565 interleaved, but anyway.
1567 - add a queued message list to procdata (the ones not matching criteria
1569 - factorize some more code between SG and RL wrt procdata
1571 - use gras_exit in example to track memleaks
1572 - get rid of gs_example now that GS is properly integrated into gras
1573 - update run_test to integrate the lastest tests (datadesc)
1575 - rename WARNINGn macros to WARNn since it prooved error-prone
1577 2004-04-19 Martin Quinson
1579 - register init/exit functions within gras module mechanism
1580 - send/receive function.
1581 Convertion is not implemented, but short-cutted if not needed.
1582 struct/array elements are sent one by one (instead of block-wise), but
1583 nobody really cares (yet). Get a prototype before optimizing.
1584 - tests (using a file socket) for DD send/receive on:
1585 - base types: int, float
1586 - array: fixed size, string (ie ref to dynamic string)
1587 - structure: homogeneous, heterogeneous
1588 - chained list, graph with cycle
1589 Believe it or not, valgrind is not too unhappy with the results. The
1590 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1594 - Bugfix when using a filename explicitely (instead of '-')
1596 2004-04-09 Martin Quinson
1598 - factorize more code between RL and SG in socket creation
1599 - Complete the implementation and tests of:
1601 o file (only in RL, and mainly for debugging)
1603 I lost 3 days to design a portable address resolver, and then decided
1604 that the prototype mainly have to run on my box.
1605 Addressing portability too early may be like optimizing too early :-/
1607 - use gras_init in the Tests instead of the crappy parse_log_opt
1608 (the latter function is removed)
1609 [Conditional execution]
1610 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1611 [Code reorganisation]
1612 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1613 Build examples against the RL library, since there is no way to disable
1614 its creation for now.
1616 For information, the beginning of coding on GRAS was back in june
1617 2003. I guess that every line has been rewritten at least twice since