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.
17 * In trace replay, allow to have one trace file per process.
18 Give the specific trace file as argument of each process,
19 and call MSG_action_trace_run(NULL)
20 You can still have one merged file for each processes.
22 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
25 SimGrid (3.3.4) stable; urgency=low
27 The "Desktop Grid needs love too" release (also called Xmas release).
30 * Major speedup in the maxmin system solving by using lazy evaluation
31 Instead of solving completely the maxmin system at each iteration,
32 only invalidate (and recompute) the modified parts.
33 This new feature is enabled in default models but you can try to
34 turn it on with "--cfg:maxmin-selective-update=1" for other models.
35 * Cas01 IMproved as default CPU model
36 This CPU model is the same Cas01 model, but it uses the
37 maxmin-selective-update flag and a heap structure to manage
38 actions on SURF kernel.
39 It reduces the complexity to find the next action to finish and,
40 consequently, it's faster than the old Cas01.
41 This is the new default CPU model (Cas01).
42 * Rename the old Cas01 model to Cas01_fullupdate
43 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
44 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
45 * CpuTI (CPU Trace Integration)
46 A new CPU model whose objective is simulate faster when using
47 availability trace files.
48 Instead of using a full featured, over engineered maxmin system for
49 CPU modeling, this model does the pre-integration of traces files
50 to calculate the amount of CPU power available, and so, executes
51 faster than the old CPU models.
52 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
53 * Use LV08 as default network model since it gives better accuracy
54 for small messages and shouldn't change things for big ones.
55 Use --cfg=network_model:CM02 to get the previous behavior.
58 ******************************************
59 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
60 ******************************************
61 * The new CPU model may changes simulations!
62 The point is that events occurring at the exact same timestamp
63 are not scheduled in the same order with the old and new
64 version. This may be enough to completely change the execution
65 of simulations in some cases.
66 * The new network model will change simulations!
67 This new model is more realistic than the previous one, so you
68 should consider redoing your old experiments with this model.
69 Sorry for the inconvenience.
72 * Introduce the supernovae compilation mode
73 When compiled that way, the whole SimGrid (or almost) is put in a
74 single compilation unit and compiled in one shoot.
75 This is to help gcc which has difficulties to inline stuff from one
77 The speedup seem to be above 15%, althrough more tests are needed on
78 amd64 to confirm that gain.
81 * Port of MSG's mailbox on top of SIMIX network
82 The put/get mechanism was greatly simplified on the way.
85 * New SIMIX network module. Provides:
86 - Mailbox: rendez-vous mecanism to find with who you want to speak
87 - Synchronous send/recv: easier and hopefully faster since the
88 logic is handled in the maestro process directly now
89 - Asynchronous send/recv: you dreamt of it? It's here now
90 Too bad that nobody cared enough to propagate the change to MSG.
91 * Add semaphores as SIMIX synchronization mechanism.
94 * new function SD_daxload(char*) to load a DAX file
95 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
96 * Introduce typed tasks. Specify its kind and cost at creation.
97 At scheduling, just give where it should be placed, and the cost
98 for each involved resource is automatically computed.
99 Existing constructors so far (more to come of course):
100 - SD_task_create_comm_e2e() for end-to-end communication
101 - SD_task_create_comp_seq() for sequential computation
102 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
103 * new function SD_task_dump() for debuging display
104 * new function SD_task_dotty(task,FILE*) writing to file the info
105 about the task in dotty format
106 * SD_task_dependency_exists() can now cope with having one of its
107 arguments NULL. If so, it tests whether the other argument has any
109 * Add getters on list of preceding/following tasks:
110 SD_task_get_parents(task) and SD_task_get_children(task)
111 * Add getters on amount of workstations and list:
112 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
113 * Add getter on task kind: SD_task_get_kind(task)
114 * Update the start_time and finish_time of tasks on completion/failure
115 * Bugfix: Remove task from state swags when destroyed
118 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
121 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
122 dynar with one separate thread per value of the dynar.
123 * Change the prototype of xbt_thread_create(), sorry.
124 Added a boolean parameter indicating whether we want to join this
125 thread (used in SG only for now)
126 * Implement xbt_thread_join and xbt_thread_yield in SG also.
129 * GTNetS wrappers should now be usable again (and betterly tested too)
130 * Fix a major regression from 3.2 where the timeout provided to
131 MSG_task_put_with_timeout() was used as absolute time before which
132 the comm should be done.
133 * Start to fix the <cluster> tag.
134 - Internal links should be good now (beside of the loopback, which
135 use the private link instead)
136 - paths to the external world is still rather broken
137 - the <route:multi> tag is just broken. Actually that's brain-dead.
138 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
139 to make it less stupid
140 ** Check your platform with teshsuite/simdag/platforms/flatifier **
141 * Fix a source-level compatibility glitch from 3.2: after defining
142 MSG_USE_DEPRECATED, you can use the old name
143 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
144 * Allow to compile from the SVN with automake 1.11
145 * Fix some problems when using the "start_time" tag in deployment XMLs.
146 * Fix #8569: XBT/synchro.h has redundant declarations
147 * Fix #8563: MSG return values and exceptions
148 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
149 * Integrate patch #8636: Obey DESTDIR when installing documentation.
150 Thanks to Robson Peixoto.
151 * Fix a vicious bug in dictionaries inducing that some elements were
152 not freed on xbt_dict_free()
154 Portability report of this version:
155 * Main portability targets:
156 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
158 - mac leopard on i386
159 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
160 but nothing critical.
161 * Other platforms: windows, AIX and others were not tested for this release
163 Timing report of this version:
164 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
165 scenario parameters). From 8h to a few seconds in desktop grid settings.
166 * Supernovae brings about 25% speedup on i386.
168 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
170 SimGrid (3.3.3) stable; urgency=low
172 The "Need for Speed" release.
174 The timings done to validate the 3.3.2 were faulty.
175 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
177 The problem was a conversion from a manually handled vector to
178 xbt_dynar_t on the critical path.
179 xbt_dynar_foreach calls functions, inducing stack management crap.
181 We inlined these functions and xbt_dynar_foreach is now breath taking.
182 We also inlined xbt_swag_belong on the way.
184 Here are some approximate speedup measurements (on master/slaves
185 simulations lasting between 10s and 20s each):
186 3.3.1 -> 3.3.2: about same performance
187 3.3.2 -> 3.3.3: 40% speedup
188 3.3.1 -> 3.3.3: 40% speedup
189 3.3.1 with inline patch -> 3.3.3: 30% speedup
191 Our reading is that the refactoring which occurred in 3.3.2 made us
192 suffer much more from the xbt_dynar_foreach low performance, but
193 once we solved this, this refactoring proved to be very performance
194 effective. From the 40% speedup, somehow, 10% are due to the
195 inlining and 30% to the refactoring.
197 That's a pitty that gcc cannot inline functions placed in other files
198 alone. We have to choose between:
199 - break the encapsulation (by putting private data structures and
200 accessors in headers files to help gcc)
201 - live with low performance
202 - switch to a decent compiler such as icc (not quite possible).
204 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
206 SimGrid (3.3.2) stable; urgency=low
208 The "Simplicity does not preceed complexity, but follows it" release.
210 The main contributors of this release were (lexical order):
211 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
214 * Extract the routing logic into its own object.
215 (was dupplicated in network.c and workstation_LV07.c;
216 Allows to implement other ways of storing that info)
217 => kill now useless network_card concept
218 - Use dynar to represent routes (instead of void** + int*)
219 - kill link_set (use surf_network_model->resource_set instead)
220 - Add a command-line option to choose the routing schema to use
221 - Add three new models:
222 * Floyd (shortest path computed at initialization)
223 * Dijikstra (shortest path recomputed all the time)
224 * Cached Dijikstra (shortest path computed on need)
225 All these models where contributed by Silas De Munck, and are
226 described in his ICCS09 paper.
228 * Simplify model declaration
229 (less redirections, less function to write when defining a model)
230 - Factorize stuff between models:
233 surf_model_resource_set(model)
234 surf_model_resource_by_name(model, name)
235 - Unify the types of models in s_surf_model_t (using an union)
236 - Embeed fields of common_public directly into s_surf_model_t
237 - Rename model methods:
238 action_free ~> action_unref
239 action_change_state ~> action_state_set
240 action_get_state ~> action_state_get
241 - Change model methods into functions :
242 (model)->common_public->action_use ~> surf_action_ref
244 * Implement a generic resource; use it as ancestor to specific ones
245 (allows to kill duplicated code in models)
246 Drawback: timer command don't need no name nor properties;
247 workstation_CLM03 don't need no properties
248 (but I guess we can live with those few bytes wasted)
250 * Improve the action object model
251 - implement a constructor avoiding dupplicated code about field
252 initialization in generic_action part.
254 * Kill the SDP model: it has an external dependency, is deprecated
255 in flavor of modern lmm models, and didn't compile since a while
258 * Relocation of the context module from XBT to SIMIX.
259 (the context were decoupled from the simix processes, duplicating a lot of code)
260 => a lot of code was factorized
261 - less overhead is introduced during scheduling
262 - simpler API for the context factory
263 - the logic for process creation,destruction and manipulation was simplified
264 * Simplification of the s_smx_process_t data structure.
265 => accesing the simix level data associated to a process is faster now,
266 and the code is a lot more readable.
269 * Implement some more MPI primitives:
270 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
271 -implementation: Bcast: flat or 2-ary tree (default),
274 Allreduce: Reduce then Bcast
275 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
276 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
277 Alltoallv: flat tree, like ompi
279 * Add support for optimized collectives (Bcast is now binomial by default)
280 * Port smpirun and smpicc to OS X
283 * Kill SD_link_get_properties: hard to maintain and makes very little sense
284 Shout out if you used it.
287 * Display the list of still queued messages in SG mode when existing
291 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
292 * Add xbt_graph_node_get_outedges() [Silas De Munck]
293 * Add xbt_str_from_file(FILE*)
294 * Add xbt_dict_get_key achieving a linear reverse search
295 * Remove the context module
297 Portability report of this version:
298 * Main portability targets:
299 - Linux(debian)/x86/context
300 - Linux(debian)/x86/pthreads
301 - Linux(debian)/amd64/context
302 - Linux(debian)/amd64/pthreads
303 On these, we still have the eratic breakages of gras/pmm and
304 amok/saturate_sg reported in previous version. We still think
305 that the tests are the cause of the fault, not the tested code.
307 - Mac OSX Leopard/x86/context
308 Still false negative in tesh autotesting.
309 Smpi still fails, but this time because readlink does not accept -f
310 Everything seems to work properly beside of that.
313 - AIX version 5.3 (only tested contexts this time)
314 Smpi still fails there because mktemp is not installed.
315 Everything seems to work properly beside of that.
317 I managed to compile it for the first time, but several breakages.
318 Won't delay the release for this exotic platform.
320 * Windows: it's still lagging behind. If you want to help, please
323 Timing report of this version:
324 This version seem to be more than 5% faster than 3.3.1 (on linux
325 64bits with contextes). The gain is less than expected, we are
326 investigating this for next release.
328 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
330 SimGrid (3.3.1) stable; urgency=low
333 * Implement a --cfg-help to show existing configuration variables
334 * Build chain do not require doxygen in maintainer mode
337 * fix a bug on struct sizeof computation, which prevented the
338 exchange of arrays of structs in some conditions
339 - added a regression test about this in datadesc_usage
340 * Allow the exchange of 0-long dynamic vectors.
341 - for that, use -1 as indicator of dynamic size instead of 0
342 - This implied to change any size from unsigned long to long,
343 reducing a bit communication abilities, but I guess that with
344 64bits being quite common, this is more than enough.
345 - This also induce a protocol change, thus bumping network protocol
346 version from 0 to 1 (if we have external users, we have to get
347 clean on that point too ;)
348 - added two regression tests about this in datadesc_usage
349 * Be more verbose when propagating local exceptions
350 This helps debugging.
351 * Display the status of simulated processes when receiving SIGINT in
355 * Allow to control the simulation from a trace file.
356 New functions MSG_action_register() and MSG_action_trace_run()
357 The first one allows to associate a function execution to each
358 kind of action while the second one parses a trace file and
359 triggers the corresponding actions within the system.
360 For now, only a toy example is provided in examples/msg/actions
361 * Add an exemple of process migration in examples/msg/migration
362 * Fix a bug in task exchange which broke MSG_task_get_sender()
363 Add a teshsuite regression test for that.
364 [Bug: if MSG_task_get_sender() is called after sender exit,
366 * Fix a bug which prevented suspend/resume to work properly
367 * Display the status of simulated processes when receiving SIGINT
368 This fixes a regression of v3.3. due to the introduction of SIMIX
369 * Bug fixing in failure management:
370 - trace could not start by a failure at time 0
371 - failure during communications were not working
374 * Add SIMIX_process_set_name() to change the name of the current
375 process in the log messages.
376 * Store smx_hosts in a dict since we only retrieve them by name
377 * Move the configuration infrastructure to surf
380 * Move the configuration infrastructure to surf
383 * Massive internal cleanups:
384 - Store internal structures on processes instead of hosts (allows
385 to have more than one process per host, in addition of being more
387 - Cleanup the initialization/finalization process
388 - Kill a whole bunch of unneeded synchronization:
389 processes run in exclusive manner within the simulator
390 - Move queues from global tables to process data fields
392 - now accept -platform and -hostfile arguments
393 - Pass the right rank value to processes according to the hostfile
394 * Compile the examples by default, and use them as regression tests
395 * Implement MPI_Wtime()
396 * Change the reference speed to a command line option
399 * TCP_gamma can now be specified as command line option using
400 --cfg=TCP_gamma:10000000.0
401 * Change the --surf-path cmd line option into --cfg=path:
404 * Also include strbuff from xbt.h public header
405 * xbt_ex_display(): do not free the exception after displaying
406 This allows to do more with the given exception afterward.
407 Users should call xbt_ex_free() themselves.
411 Portability report of this version:
412 * Main portability targets:
413 - Linux(debian)/x86/context
414 - Linux(debian)/x86/pthreads
415 - Linux(debian)/amd64/context
416 - Linux(debian)/amd64/pthreads
417 These targets fail about 1/10 of times on gras/pmm, but we believe
418 that this is because of the test, not because of simgrid.
419 amok/saturate_sg fails even more rarely, and the test may not be
422 - Mac OSX Leopard/x86/context
423 The test suite still spits tons of errors because some obscure
424 force prevents us from removing the temporary directories
425 arguing that they still contain some metadata I've never heard of.
426 Smpi fails because seq is not installed.
427 Everything seems to work properly beside of that.
430 - AIX version 5.3 (both contexts and pthread)
431 Smpi still fails there because mktemp is not installed.
432 XML inclusions seems rosty on AIX.
434 * Windows: it's still lagging behind. If you want to help, please
437 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
439 SimGrid (3.3) stable; urgency=high
443 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
446 * Introduce the SIMIX module: factorize code between MSG and GRAS.
449 Until now, GRAS were using MSG as an interface to SURF. It was
450 quite difficult because both interface have several differences
451 (MSG channels vs GRAS sockets were the most notable point).
453 This also opens the gate to SMPI (which should occur soon) and speed
454 up simulations by to 40% (even if it were not the main goal).
456 **************************************
457 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
458 **************************************
459 The point is that events occuring at the exact same timestamp are
460 not scheduled in the same order with the old and new version. This
461 may be enough to completely change the execution of simulations in
462 some cases. Sorry for the inconvenience.
464 * Cleanup and upgrade the XML format to push further scalability
465 issues (check http://hal.inria.fr/inria-00256883/ for more info)
467 * Improve the testing infrastructure with tesh. Now a very large part of
468 the code is tested not only by being run but also by checking that the
469 output match an expected output [Mt].
471 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
472 is really less memory-demanding, which should allow you to use
473 larger files in SimGrid [AL].
475 * Inform valgrind about our contextes, so that it becomes usable
476 with the default (and more effecient) version of SimGrid
477 [contributed by Sékou Diakite, many thanks]
480 * Introduce a listener thread in charge of receiving incomming
481 messages from the network. It allows to overlap communication and
482 computation but most notably, it removes some stupid deadlocks due
483 to the fact that so far, a process could not send and receive at
484 the same time. This made most non trivial communication schema
486 * Convert the PIDs from long int to int to match the MSG ones (and
488 * New function: gras_agent_spawn() to launch a new process on
489 current host. Only working in simulation for now. [Mt]
490 * New function: gras_os_hostport() returning a constant form (ie,
491 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
494 * Make the backtrace of exceptions more human readable [Mt]
495 * New module: xbt/str [Mt]
496 a ton of string utility functions (split, join, printf to a newly
497 allocated buffer, trim, etc)
498 * New module: xbt/hash [Mt]
499 SHA1 hashing algorithm (more to come if needed)
500 * New module: xbt/synchro [Mt]
501 synchronization tools (mutex and conditions) working the same way
502 in simulation and in real life (mainly useful for GRAS, but not
504 * New module: xbt/queue [Mt]
505 classical producer/consumer synchronization scheme
506 * xbt_dynar_new_sync() creates a synchronized dynar. All access
507 (using the classical functions will get serialized) [Mt]
508 * Make dictionary internal table dynamic. No need to specify its size
509 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
511 * Make sure the log channels are organized as a tree under windows
512 (because of ANSI C compatibility issue, any channel were child of
516 * Cleaned many thing in surf and fixed a few bugs [AL].
517 * Add a nice command line configuration mechanism to compose models [AL].
518 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
519 the previous one (KCCFLN05). It relies on something that looks like
520 a max-min sharing mechanism but cannot be written as such. A new solver
521 was thus designed [AL].
522 * Add a new solver to lmm. Based on Lagrange optimization and
523 gradient-based descent, it enables to efficiently maximise systems s.a
525 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
527 This solver enables to propose two new network models for TCP Reno and
528 TCP Vegas based on Low's work. These models still need to be fully
529 tested though [Pedro Velho].
532 * Bug fix in SD_simulate. Now the time bound given as argument is
534 * Use the new parallel task model (ptask_L07) as default.
535 * Use the SURF command line configuration mechanism.
536 * 0-size tasks (for synchronization) should now work.
538 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
540 SimGrid (3.2) stable; urgency=high
544 We still experience issues on this platform, but we believe that at
547 GRAS API BREAKAGE (for simplification purpose, sorry):
548 * the gras_msgtype_by_name is not used anymore. Instead of
549 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
550 you can write (and must)
551 gras_msg_send(toserver, "request", &request);
552 - If you still want to pass a gras_msgtype_t to the function (to cache
553 the type and avoid the lookup time), use the gras_msg_send_() variant.
554 - Impacted functions:
555 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
556 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
557 * The callbacks are now expected to return 0 when everything went well
558 (just like the main() function)
560 GRAS new features and improvements:
561 * New module mecanism where user code can use per process globals [Mt]
562 This is similar to gras_userdata_*() functions, but for libraries. It
563 factorize some code developped over and over in the examples and AMOK.
564 It has still to be documented and used (only amok/peermanagement is
566 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
567 loss when they were small enough to fit into the buffer and sent quickly
568 enough so that they can all get received in one shoot.
569 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
570 if not found. Use the *_or_null() variant for the old semantic.
571 * In gras_msg_handle, do not discard messages without callback.
572 They are probably messages to be explicitly awaited later (ie, proofs of
573 mis-synchronization in userland since they are sent before being awaited)
575 * gras_socket_meas_send/recv: semantic changed!
576 The numerical arguments used to be (1) the total amount of data to send
577 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
578 messages. This was need for the fool willing to send more than MAXINT
579 bytes on quite fat pipes.
582 * Do really rename the hostmanagement module to peermanagement. [Mt]
583 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
584 but this is rather new and this was documented in the module
585 documentation (poor excuses, I admit)
586 * Bandwidth measurement semantic changed! This follows the changes to
587 gras_socket_meas_send/recv explained above.
590 * A sequential mode has been added to the workstations. When a workstation
591 is in sequential mode, it can execute only one task, and the other tasks
592 are waiting in a FIFO. [Christophe Thiery]
595 * The KCCFLN05 workstation model now handles parallel tasks. It is the
596 model for SIMDAG. [Christophe Thiery]
597 * Bug fix in the maxmin solver: Some values were close to 0 instead of
598 equal to 0, which caused some bad behaviors in
599 saturated_constraint_set_update. I now use a threshold mechanism like in
603 * When running manually src/testall, you select specific units [Mt]
604 testall is the result of our cunit mecanism, and should replace all
605 the scripty thingy around since bash don't run easily on billware.
607 * A mallocator system has been added. [Christophe Thiery]
608 Mallocators allow you to recycle your unused objects instead of freeing them
609 and allocating new ones.
611 Documentation update:
612 * FAQ reworking + New FAQs:
613 - "Valgrind spits tons of errors!" [Mt]
614 - "How to repport bugs" [Mt]
615 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
616 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
617 - Communication time measurement within MSG [AL]
618 - I experience weird communication times when I change the latency [AL]
621 - an introduction to the framework and to the used communication model
622 - an initiatic tour introducing the most proheminent features:
624 . Lesson 0: Installing GRAS
625 . Lesson 1: Setting up your own project
626 o Part 2: Message passing
627 . Lesson 2: Exchanging simple messages
628 . Lesson 3: Passing arguments to the processes (in SG)
629 . Lesson 4: Attaching callbacks to messages
630 . Lesson 5: Using globals in processes
631 . Lesson 6: Logging informations properly
632 . Lesson 7: Using internal timers
633 . Lesson 8: Handling errors through exceptions
634 . Lesson 9: Exchanging simple data
635 . Lesson 10: Remote Procedure Calling (RPC)
636 . Lesson 11: Explicitely waiting for messages
637 . Recapping of message passing features in GRAS
638 - A HOWTO section containing:
639 o HOWTO design a GRAS application
640 More are due, of course. They will come latter. In the meanwhile, you can
641 check the examples which are still here.
643 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
645 SimGrid (3.1) stable; urgency=high
649 There was a stack corruption somewhere, visible only when optimizing
650 with these versions. [Vince]
653 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
654 module that enabled to play with Directed Acyclic Graphs. It is built
655 directly on top of SURF and provides an API rather close to the old
656 SG. Some old codes using SG are currently under rewrite to check that
657 all needful functions are provided. [Christophe Thiery]
660 * Complete rewrite of the KCCFLN05 workstation model. It is now an
661 extension of the classical CLM03 model that gracefully handles
662 failures. This is now the default model for MSG and GRAS. It doesn't
663 handle parallel tasks yet though. [AL]
664 * Bug fix: Weights were not correctly set in the network part.
665 WARNING: This may have resulted in incorrect results with simulations
666 where there are more than one flow on a given link. [AL]
669 * After a (long ?) discussion on simgrid-devel, we have decided that the
670 convention we had on units was stupid. That is why it has been decided
671 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
672 WARNING : This means that all previous platform files will not work as
673 such with this version! A warning is issued to ask users to update
675 A conversion script can be found in the contrib module of the CVS, under
676 the name contrib/platform_generation/surfxml_update.pl [MQ]
679 * Bug fix: Processes were started in reverse order, wrt deployment file.
680 WARNING: if your code relies on this bug, please fix it. [AL]
681 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
682 being executed on two different locations. [AL]
683 * Bug fix: Failures are now better supported thanks to Derrick's tests
684 (there was many failure situations I hadn't thought of and that weren't
685 correctly handled). [AL]
686 * New function: MSG_host_is_avail indicates you whether a given m_host_t
690 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
691 Exception occurring on server-side are propagated back to client (!).
693 API CHANGE: the callback changed their prototype. Change:
694 int my_handler(gras_socket_t expeditor, void *payload_data) {
696 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
697 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
699 * New! function: gras_msg_handleall to deal with all messages arriving
700 within a given period.
701 * New! function: gras_socket_server_range to get a server socket in a
702 range of port numbers (ease to avoid port number conflicts) [MQ]
703 * New! gras processes display their backtrace when they get a SIGUSR1
704 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
705 Sweet to debug RL processes [MQ]
709 - Do not force experiment sizes to be expressed in kb, or it becomes
710 impossible to measure the latency this way (needs one byte-long tests)
711 WARNING: this changes the amok_bw_* function semantic. [MQ]
712 - Implements the link saturation stuff. [MQ]
713 * Peer management module:
714 New! module factorizing code that we wrote over and over [MQ].
717 * New module: cunit (my jUnit implementation in ansi C) [MQ]
718 - Test units are placed directly into the library code, they get extracted
719 automatically and placed into the src/testall binary.
720 - Convert most of the XBT tests to this system.
721 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
722 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
723 live in the same C code. This required to change a little bit the API
724 of surfxml parsing but shouldn't be an issue for end-users. [AL]
725 * New module: sparse graph structure with basic algorithms (this is work
726 in progress and the API is not considered to be frozen yet). [AL]
727 * Display more information on backtraces: source line & function names are
728 now displayed just like valgrind does (rely on addr2line tool) [MQ]
729 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
730 * Reworked a little bit some #include statements to load only required
731 headers. Some user code that relied on SimGrid to include stdlib or
732 stdio may need to include it by themselves. [AL]
733 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
735 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
736 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
737 but I guess that almost nobody used those parts. [MQ]
739 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
741 SimGrid (3.0.1) stable; urgency=low
744 * Unfortunately, I had missed 5 misnamed functions:
745 xbt_fifo_item_t xbt_fifo_newitem(void);
746 void xbt_fifo_freeitem(xbt_fifo_item_t);
747 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
748 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
749 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
750 They're now deprecated. Please use their new versions:
751 xbt_fifo_item_t xbt_fifo_new_item(void);
752 void xbt_fifo_free_item(xbt_fifo_item_t);
753 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
754 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
755 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
757 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
758 * Documentation: xbt_log module unmercifully reworked [MQ]
759 * Bugfix: there was a problem with the ending of contexts with
760 the pthread backend. It caused some weird deadlock or behavior
761 depending on the pthread implementation. [AL]
762 * Bugfix: get the exceptions raised in the simulator repport where
763 and why they come from when they are not catched in time [AL, MQ]
766 * Bugfix: Do repport the error when two non-connected hosts try to
767 exchange data (Thanks to Flavien for stumbling into this one) [AL]
770 * Add additionnal checkings on communications. Assert that two
771 communicating hosts are connected by a set of links... [AL]
774 * Add additionnal checkings on channel values in communication [AL]
775 * New: MSG_task_get_source to see on which host a task was generated [HC]
776 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
777 the number of tasks waiting to be received on channel and sent
779 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
780 waits for the first task coming from a given host.. [AL]
782 GRAS new functionnalities: [MQ]
783 * Enhance the parsing macro to allow the size of multidimentional objects
784 to be given thru annotations.
785 * New example (and documentation): Matrix Multiplication a la RPC
786 (as when I was young!) and fix a bunch of bugs found on the way.
788 GRAS performance improvements: [MQ]
790 * Reduce the amount of cbps creation/destruction by making it static to
791 datadesc_send/recv() and using a (newly created) cbps_reset (based on
794 * Change libdata to a set so that we can search for stuff by ID (and thus
795 reduce the insane amount of dict lookups)
798 * Actually implement gras_datadesc_copy() so that we don't have to mimick
799 RL communication on top of SG since it's so uneffective.
800 It may also be used for inter-thread communication in RL, one day.
801 * Use gras_datadesc_copy() to exchange messages on top of SG
803 - improve message exchange performance on top of SG
804 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
805 * Don't exchange on the network the size of the used part of buffer,
806 instead, specify the possible buffer size to read().
808 - reduces the amount of read/write calls (one pair per exchange)
809 - reduces the amount of exchanged data (the size)
810 - allows to retrieve all arrived data on receiver side, if we don't need
811 it right now (subsequent read will peek the buffer)
812 - allows the receiver to proceed with the begining of the stream before
813 everything is arrived
814 - make it possible to build an iov transport (using readv/writev)
816 - take care of the data with non-stable storage (like stacked data),
818 * If possible, TCP send uses vector I/O (when writev() is here)
819 - Don't use it for receive since we send data sizes and data on the
820 same stream, so we wouldn't be able to chain large amount of chunks
821 before having to flush the stuff to read the size.
822 * Rework the transport plugin mecanism to simplify it and reduce the
823 amount of pointer dereferencement when searching for the right function
826 * I guess that now, we do almost as few system calls as possible while
827 doing as few data copy as possible.
829 To improve it further, we could try to send all the sizes first and then
830 all the data (to use iov on receiving size), but it's only a partial
831 solution: when you have 2 dimensional data, the sizes of the second
832 dimension is data of the first dimension, so you need 3 streams.
834 I'm not sure the potential performance gains justify the coding burden.
836 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
838 SimGrid (3.00) stable; urgency=high
841 * New! Give the possibility to hijack the surf parser and thus bypass
842 MSG_create_environment and MSG_launch_application. Have a look at
843 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
845 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
847 SimGrid (2.96) unstable; urgency=low
852 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
853 This deprecates the xbt_error_t mecanisms.
854 It modifies (simplifies) all XBT and GRAS API.
855 MSG API keeps unchanged (exceptions raised by XBT are catched from
856 within MSG and masked with existing error handling facilities)
859 * New! Add a FATPIPE model. [AL]
860 * New! Add a parallel task model. [AL]
861 * New! Add automatically a loopback interface (in the default
862 network model) if none was precised.
865 * Bugfix: MSG_process_resume now works with the current running process.
867 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
868 * Modification of MSG_task_get_compute_duration. Once a task has been
869 processed, the value returned by this function is now equal to 0. [AL]
870 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
871 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
872 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
873 of a m_task. For now, MSG_task_cancel only works with computation
875 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
876 of the processor (in Mflop/s) regardless of the current load on the
878 * API Change: use proper naming convention for MSG_getClock and
879 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
881 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
882 This function changes the priority of a computation task. This priority
883 doesn't affect the transfer rate. A priority of 2 will make a task
884 receive two times more cpu power than the other ones. This function
885 has been added to suit the needs of Nguyen The Loc and hasn't been that
886 much tested yet. So if it fails, please report it and send me your code.
888 * API Change: removed all functions and types that were marked "deprecated"
889 since many months. Renamed MSG_global_init_args to MSG_global_init.
891 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
893 SimGrid (2.95) unstable; urgency=low
896 * Steal some nice code to GNU pth to fix context detection and usage [AL]
897 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
898 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
901 * Bugfix: Allow absolute paths for platform description files [MQ]
902 * Bugfix: do free the variables after use. Leads to drastic performance
904 * Implement max_duration (ie, timeouts) on resources [AL]
907 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
909 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
910 the needed support (timeouts) [AL]
913 * Implement measurement sockets. You can now get the bandwidth between two
914 hosts thanks to AMOK (see below). [MQ]
915 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
916 dynar over the network (yeah) [MQ]
917 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
918 * simple/static token-ring example. [Alexandre Colucci and MQ]
919 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
923 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
924 type parsing. "short" and "long" modifiers were also ignored; other
925 modifier (such as reference level) are still ignored. [MQ]
926 * Embeed the buffer size within the buffer itself on SG. [MQ]
927 That way, send() are atomic and cannot get intermixed anymore (at least
928 the ones which are less than 100k; bigger messages still have the issue)
929 * Array size pushed by the field, not by the field type (or each
930 and every long int will push stuff to the cbps) [MQ]
931 * use select() to sleep since it allows to portably sleep less than one
934 GRAS (minor cleanups)
935 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
938 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
939 * type category 'ignored' killed as it was never used and were difficult
941 * whether a type can cycle or not is now a flag, leaving room for more
942 flags (as "ignored", if we feel the need one day ;)
943 * Rename raw sockets to measurement sockets since "raw" has another
944 meaning in networking community.
947 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
948 and offers features not belonging to GRAS but that most applications
949 need. One day, it may be a set of plugins loadable at runtime.
950 * New module: bandwidth
951 bandwidth measurement between arbitrary nodes running this module. [MQ]
953 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
955 SimGrid (2.94) unstable; urgency=low
957 The first beta release of SimGrid 3 !
961 * Update the main page and the FAQ. Adding references to gforge.
964 * Add a gras_os_getpid function.
967 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
968 * Extend the logs so that they also print PID, hostname, date, ... if
970 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
971 and kill the old version which were in testsuite/
972 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
976 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
977 for users. Instead, each programming environment (GRAS, MSG,...) use it
978 when needed to provide such a feature to users.
979 Don't shortcut the mecanism or you will also shortcut the virtualization
980 you need on the simulator.
984 * Cleanups in configury with regard to compile optimization/warning flags.
985 Also add -fno-loop-optimize to any powerpc since it's the optimization
986 killing gcc (< 3.4.0).
987 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
988 needed by MSG examples complications
989 * Borrow configury beautifications from PHP
992 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
993 warning (thanks loris for stumbling into this one).
994 * Bugfix: stop loading private headers (gras_config.h) from the public
998 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
999 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1000 RL would imply a DNS resolver, which is *hard* to do in a portable way
1001 (and therefore delayed).
1002 * Implement a real timer mecanism and use it in timing macros. This allows
1003 to avoid rounding errors and get a 0.000005 sec precision in timing
1004 macros. While I was at it, various cleanups:
1005 - allow to declare more than one timed section per file (fix a stupid bug)
1006 - move some private declaration to the right place
1007 - merge conditional execution and timing macros into emulation module
1008 - document the module
1009 - make sure the module cleanups its mess on gras_exit
1010 * Documentation improvements:
1011 - (new) how to compile applications using GRAS
1012 - (new) emulation support (timing macros)
1014 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1016 SimGrid (2.93) unstable; urgency=low
1018 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1021 - Use Paje properly where used. Still to be sanitized properly.
1022 - Portability fix: Add an implementation of the contexts using pthread
1026 - Add xbt_procname(): returns the name of the current process.
1027 Use it to show the current process's name in all logging.
1029 - fix detection of older flex version and the reaction, since we do
1030 depend on modern ones (we use lex_destroy)
1031 - Better separation of SG and RL in the libs: remove all simulation code
1032 from libgras. As a result, this lib is now only 200k when stripped.
1033 Some of the xbt modules may also be duplicated (two sets and such) and
1034 should be cleaned/killed before SG3.
1035 - Insist on using xlC on AIX because of weird problems involving gcc there.
1036 - Cleanup the make remote stuff. This is now done by scripts
1037 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1038 mainly for our private use, but we're working on changing them to user
1041 - Bugfix: flush the socket on close only if there is some *output*.
1042 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1044 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1047 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1049 SimGrid (2.92) unstable; urgency=low
1051 Alpha 3 on the path to SimGrid 3
1055 - New! First try of benchmarking macros.
1056 - New! First try so that gras_stub_generator generate deployment and
1057 remote compilation helpers.
1059 - Bugfix: Initialization fix in msg_test.
1063 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1065 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1067 - API Change: gras_os_sleep to take the amount of seconds as a double.
1068 Accepting an int was error prone since it was the only location where
1069 seconds were coded as such. It leaded to damn rounding errors.
1070 - Bugfix: Hard to belive that timers ever worked before this.
1072 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1074 SimGrid (2.91) unstable; urgency=low
1076 Alpha 2 on the path to SimGrid 3
1080 - Bug fix in the lmm_solver.
1082 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1083 through the function MSG_paje_output.
1084 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1085 - It is possible to bound the rate of a communication in MSG with
1086 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1087 to write it in the changelog).
1088 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1093 - Various cleanups to the autotools stuff
1094 - Begin to move Gras examples to examples/gras/
1095 - Let make distcheck work again (yeah!)
1097 - documentation overhauled using doxygen.
1098 gtk-doc-tools is dead in SimGrid now.
1099 - Automatically extract all existing logging categories, and add the list
1100 to the documentation (long standing one, to say the less)
1102 - Cleanup the known architecture table. Reorder the entries to group what
1103 should be, and use a more consistent naming scheme.
1104 (some of the test dataset are still to be regenerated)
1105 - New! Allow library to register globals on each process just as userdata
1107 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1108 lookup time (for now).
1109 Use it in msg and trp.
1110 This cleans a lot the internals and helps enforcing privacy of the
1111 headers between the gras components.
1112 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1113 - Bugfix: gras_os_time was delirious in RL.
1114 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1116 - Reenable GRAS now that it works.
1118 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1120 SimGrid (2.90) unstable; urgency=low
1122 Alpha 1 on the path to SimGrid 3
1124 * It is a long time since the last release of SimGrid. I'm sorry about
1125 that but as I had told you, I was rewriting a lot of things. I apologize
1126 to those who had been reporting bugs to me and that I had not answered.
1127 If your bug is still in the new version, please tell me. Here is a
1128 summary of the main changes.
1130 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1131 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1132 portability, speed, and a lot more but you'll figure it out later.
1133 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1134 I wanted to release the new MSG as soon as possible and I have
1135 broken GRAS, which is the reason why, for now, only MSG is fully
1136 functional. A laconic description of these projects is available
1137 in the documentation.
1139 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1140 kernel optimized for our needs (called SURF but only the developers
1141 should use it). Hence, MSG is now roughly 30 times faster and I think
1142 that by rewriting a little bit MSG, I could event speed it up a little
1143 bit more. Beside the gain in speed, it is also much easier to encode a
1144 new platform model with SURF than it was with SG. More to come...
1146 * REVOLUTION 3: I have tried to change a little as possible the API of
1147 MSG but a few things really had to disappear. The main differences
1148 with the previous version are :
1149 1) no more m_links_t and the corresponding functions. Platforms are
1150 directly read from a XML description and cannot be hard-coded
1151 anymore. The same format is used for application deployment
1152 description. The new format is described in the documentation.
1153 Have a look in tools/platform_generation. There is a tiny script
1154 that converts from the old platform format to the new one. Concerning
1155 the application deployment format, parsing the old one is tricky.
1156 I think most of you should however be able to convert your files. If
1157 it is really an issue, I can write a C code that does the conversion.
1159 2) the toolbox tbx does not exist anymore. We now have a library
1160 with much more data-structures but without the hash-tables (we have
1161 dictionaries that are much faster).
1163 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1165 *****************************************************************************
1166 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1167 * don't feel like converting the oldies. (Mt) *
1168 *****************************************************************************
1171 Version 2.90: "the long awaited one"
1172 - Finished rewriting and debugging MSG. Rewrote the documentation.
1173 - disable GRAS for now since it needs to be ported to the newest SG
1176 - Finish the port to windows (using mingw32 for cross-compile)
1179 - Main loop and datastructures of SURF. A cpu resource object is
1180 functional. Surf can thus be used to create cpu's with variable
1181 performance on which you can execute some actions.
1183 2004-11-15 Martin Quinson
1184 - Port to ARM. Simply added the alignment and size descriptions. Should
1185 work, but the ARM machines are so slow that I didn't had the opportunity
1186 to 'make check' over there yet.
1188 2004-11-15 Arnaud Legrand
1189 - Trace manager now written. It uses a heap structure and is therefore
1190 expected to be efficient. It may however be speeded up (particularly
1191 when many events occur at the same date) by using red and black
1192 trees. One day maybe...
1193 - Max-min linear system solver written. It uses a sparse matrix
1194 structure taking advantage of its expected use. Most operations are
1195 O(1) and free/calloc are called as few as possible. The computation of
1196 the minimum could however be improved by using a red and black tree
1199 2004-11-03 Arnaud Legrand
1200 - Rename every gras_* function that was in xbt/ to its xbt_
1202 - Add a heap and a doubly-linked list to xbt
1203 - Added a dichotomy to the dictionaries. make check works as well before
1204 so I assume that the patch is correct. I do not know however if things
1205 run effectively faster than before now. :)
1207 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1208 SimGrid, and the version number is bumped to 2.x
1210 2004-10-29 Martin Quinson
1211 - Introduction of the remote errors.
1212 They are the result of a RMI/RPC on the remote machine.
1213 ErrCodes being scalar values, you can't get the host on which those
1214 errors did happen. Extending the error mechanism as in Gnome is possible.
1215 No idea yet whether it is a good idea.
1217 2004-10-28 Martin Quinson
1218 - Interface revolution: the Starred Structure Eradication.
1219 I used to do typedef struct {} toto_t; and then handle *toto_t.
1220 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1221 - ???_t is a valid type (builded with typedef)
1222 - s_toto_t is a structure (access to fields with .)
1223 - s_toto is a structure needing 'struct' keyword to be used
1224 - e_toto_t is an enum
1225 - toto_t is an 'object' (struct*)
1227 typedef struct s_toto {} s_toto_t, *toto_t;
1228 typedef enum {} e_toto_t;
1229 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1230 s_toto_t) is private.
1232 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1233 so that it can set the variable to NULL. It was so for dicts and sets,
1234 it changed for dynars.
1236 - Fix a bunch of memleaks in dict_remove
1237 - Fix a bug in sg/server_socket opening: it failed all the time.
1239 2004-10-07 Martin Quinson
1240 - Speed up dynar lookup operation a bit.
1242 gras_dynar_get is dead.
1244 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1245 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1246 (which gives you the address of the stored data).
1248 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1249 copy of the data using an affectation to do the job and not a memcpy.
1251 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1252 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1254 It does not really speedup the dynar test because they are
1255 setting elements all the time (and look them seldom). But the dict does
1256 far more lookup than setting.
1258 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1260 2004-10-05 Martin Quinson
1261 - Allow to (en/dis)able the cycle detection at run time.
1263 Whether we should check for cycle or not is now a property of each
1264 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1265 datadesc_cycle_unset allow to remove this property when previously set.
1267 Note that the cycle detection is off by default since it impacts the
1268 performance. Watch the data you feed GRAS with ;)
1270 This property is hereditary. Any element embedded in a structure having it
1271 set have it set for the time of this data exchange.
1273 You should set it both on sender and receiver side. If you don't set it on
1274 sender side, it will enter an endless loop. If you forget on receiver
1275 side, the cycles won't be recreated after communication.
1277 - Header reorganization.
1278 Kill gras_private.h, each submodule must load the headers it needs.
1280 2004-10-04 Martin Quinson
1281 - Interface revolution: do not try to survive to malloc failure.
1283 Now, gras_malloc and friends call gras_abort() on failure.
1284 As a conclusion, malloc_error is not a valid error anymore, and all
1285 functions for which it was the only gras_error_t return value are
1286 changed. They now return void, or there result directly.
1287 This simplify the API a lot.
1289 2004-09-29 Martin Quinson
1290 - Re-enable raw sockets.
1291 Created by gras_socket_{client,server}_ext;
1292 Used with gras_raw_{send,recv}
1295 It should allow to kill the last bits of gras first version soon.
1297 This is not completely satisfactory yet (duplicate code with
1298 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1301 - Simplify transport plugin (internal) interface by not passing any
1302 argument to _server and _client, but embedding them in the socket
1305 2004-09-28 Martin Quinson
1306 - Finish the port to AIX.
1307 autoconf was my problem (segfault within the malloc replacement
1308 function. No idea why)
1310 2004-09-16 Martin Quinson
1311 - Fix some size_t madness on 64bit architectures.
1313 2004-09-08 Martin Quinson
1314 - Reduce the number of system headers loaded, overload some more system
1315 calls (such as malloc to cast the result of the system one, and work
1317 - Fix and reintroduce the config support
1319 2004-09-07 Martin Quinson
1320 - Source code reorganization to allow Arnaud to surf all over there.
1321 - Allow to document the logging categories.
1322 - Remove all uppercase from logging categories and useless cleanup in names.
1324 2004-08-18 Martin Quinson
1325 Version 0.6.2 (protocol not changed; API changed)
1326 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1327 gras_error_t), and NULL when not found. Functions expecting a msgtype
1328 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1329 hopefully usefull message.
1330 - Portability to prehistoric sparcs again
1332 2004-08-17 Martin Quinson
1333 Version 0.6.1 (protocol not changed; ABI not changed)
1334 - prealloc some buffers to speed things up
1336 2004-08-11 Martin Quinson
1337 Version 0.6 (protocol not changed; ABI expended)
1338 - The parsing macro can deal with the references, provided that you add
1339 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1341 2004-08-09 Martin Quinson
1342 Version 0.5 (protocol not changed; ABI changed)
1343 - Allow to off turn the cycle detection code in data exchange at
1344 compilation time. It should be at run time, but I'm short of time (and
1345 the config stuff is still broken). That way, we keep dict out of the
1346 critical path, which is good because the performance is poor:
1347 - search not dichotomial yet
1348 - dynar give no way to access their content and memcpy everytime
1349 - In composed data description (struct, ref and so on), stop foolness of
1350 keeping the subtype's ID, but store the type itself. This keeps sets out
1351 of the critical path, which is good since they rely on dynar and
1352 dictionnaries. The only loose of that is that we cannot detect the
1353 redeclaration of a structure/union with another content (but I'm not sure
1354 the code detected well this error before anyway). We still can detect
1355 the redefinition discrepancy for the other types.
1356 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1357 it breaks the code because of type-punning used all over the place).
1358 This breaks on all non-gcc architectures (for now).
1360 All those changes (plus the buffer of last time) allow me to gain 2 order
1361 of magnitude on cruel tests consisting of 800000 array of integers on two
1362 level of a hierarchical structure (200 secondes -> 4 secondes)
1365 - the selector of reference must now return the type it points to, not
1366 the ID of this type.
1368 2004-08-06 Martin Quinson
1369 Version 0.4 (protocol changed; ABI not changed)
1370 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1371 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1372 - Implement a buffer transport, and use it by default (it relies on tcp in
1373 real life and on sg in simulation).
1374 That's a bit hackish since I had a new field to the structure to store
1375 its data without interfering with the subtype ones. Inheritance
1376 is tricky in C. And that's a kind of reverse inheritance with one class
1377 derivating two classes. Or maybe a game with java interfaces. Anyway,
1378 that's damn hard in C (at least).
1379 Moreover, I got tired while trying to ensure plugin separation and
1380 genericity in SG mode. MSG wants me to do weird things, so let's go for
1381 cruel hacks (temporarily of course ;).
1382 See comment in transport_private.h:71
1383 - do not include all the _interface headers in private but in the files
1384 which really need them (to cut the compilation time when they are
1387 2004-07-26 Martin Quinson
1388 Version 0.3 (protocol not changed; ABI changed)
1389 - Major overhault of the datadesc interface to simplify it:
1390 - shorted the function names:
1391 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1392 - add a trivial way to push/pop integers into the cbps without malloc.
1393 This allows to make really generic sub_type description, which simply
1394 pop their size of the stack.
1395 - add a function gras_datadesc_ref_pop_arr() which does what users want
1396 most of the time: Declare a dynamic array (which pops its size of the
1397 stack) and declare a reference to it. Poor name, but anyway.
1398 - kill the post-send callback, add a post-receive one
1400 2004-07-23 Martin Quinson
1401 Version 0.2 (protocol changed; ABI changed)
1402 - add some testing for cpbs in the test cases, and fix some more bugs.
1403 This invalidate again the little64 data file, since I cannot regenerate
1405 - remove an awfull optimization in the logging stuff, allowing me to:
1406 - understand it again
1407 - learn gcc how to check that the argument match the provided format
1408 - fix all errors revealed by gcc after that
1409 - internal keys of dict are not \0 terminated. Deal with it properly in
1410 loggings instead of segfaulting when the user want to see the logs :-/
1412 2004-07-22 Martin Quinson
1413 - Fix some stupid bug preventing cbps (callback postit) from working
1415 2004-07-21 Martin Quinson
1416 - Some documentation cleanups
1417 - remove the useless last argument of msgtype_declare
1418 - rename the Virtu functions to fit into the 'os' namespace
1419 - move headers src/include -> src/include/gras/ and stop fooling with
1420 gras -> . symbolic link
1421 - make distcheck is now successful
1423 2004-07-19 Martin Quinson
1425 - Build shared library also
1426 - Install html doc to the right location
1427 - stop removing maintainer files in make clean
1428 - build tests only on make check
1430 2004-07-13 Martin Quinson
1432 - No major issue in previous version => change versionning schema
1433 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1434 corresponding dataset.
1436 2004-07-11 Martin Quinson
1438 - Get it working with any kind of structure (we can compute the padding
1439 bytes remotely for all the architectures I have access to)
1440 - Implement the structure parsing macro (still not quite robust/complete)
1441 - Improvement to the remote testing toysuite
1443 2004-07-10 Martin Quinson
1444 [autoconf mechanism]
1445 - get ride of a bunch of deprecated macros
1446 - actually run the test for two-compliment, not only compile it :-/
1447 - test whether the structures get packed (and bail out if yes. Damn.
1448 Alignment is a serious matter)
1449 - test whether the structures get compacted (but respecting the alignment
1450 constraints of each types)
1451 - test whether the array fields of structures can straddle alignment boundaries
1453 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1454 (took me 2 hours to find that bug, looking at the wrong place)
1455 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1456 before sending/receiving and are used to compute the offsets of fields
1457 - Given that padding size depend even on compiler options, keep track of
1458 alignment and aligned_size only for the current architecture. Not a big
1459 deal since we send structure fields one after the other (seems
1461 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1462 program, let it work on linux/gcc/little32. portability todo.
1464 2004-07-08 Martin Quinson
1465 - import and improve remote compilation support from FAST
1466 - make sure make check works on half a dozen of machines out there
1468 2004-07-07 Martin Quinson
1469 Let's say it's version 0.0.3 ;)
1470 - Implement conversions (yuhu!)
1471 - Let it work on solaris (beside conversion, of course)
1472 - Stupid me, using rand() to generate the conversion datatests in not wise.
1474 2004-07-06 Martin Quinson
1475 - Let make dist work, since I'm gonna need it to compile on remote hosts
1476 - Let Tests/datadesc_usage write the architecture on which the file was
1477 generated as first byte.
1478 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1479 architecture descriptions.
1480 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1481 on those architectures.
1482 - Optimization: send/recv array of scalar in one shoot
1484 2004-07-05 Martin Quinson
1485 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1487 - Plug a whole bunch of memleaks
1488 - each process now have to call gras_{init,exit}. One day, their log
1489 settings will be separated
1490 - Continue the code factorisation between SG, RL and common in Transport.
1492 2004-07-04 Martin Quinson
1494 - Redistribution between SG and RL.
1495 We wanna have to accept in SG, so move accepted related parts of RL in
1496 the common part. (more precisely, the dynar of all known sockets is no
1497 more a static in transport.c, but part of the process_data)
1499 [gras_stub_generator]
1500 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1502 2004-07-03 Martin Quinson
1503 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1504 them all up in one shot)
1506 - Fix the ugly case of reference to dynamic array.
1507 - New (semi-public) function gras_datadesc_size to allow the messaging
1508 layer to malloc the needed space for the buffer.
1510 - gras_socket_close now expect the socket to close (and not its address to
1511 put NULL in it after it). This is because the socket passed to handlers
1512 is one of their argument (=> not writable).
1514 - propagate the interface cleanup from last week in datadesc, ie remove a
1515 superfluous level of indirection. User pass adress of variable
1516 containing data (both when sending and receiving), and not of a variable
1517 being a pointer to the data. Let's say that I like it better ;)
1518 The price for that is constructs like "int msg=*(int*)payload" in
1519 handlers, but it's a fine price, IMHO.
1521 - Let it work in RL (yuhu)
1523 2004-06-21 Martin Quinson
1525 - porting SG plugin and SG select to new standards (works almost).
1526 - plug memleaks and fix bugs around.
1529 - cleanup the prototype of data recv and force users to specify when they
1530 want to handle references to objects. Test case working even for cycles.
1531 - plug memleaks. Valgrind is perfectly ok with this.
1533 2004-06-12 Martin Quinson
1535 - cleanup the separation between plugin and main code in plugin creation
1537 2004-06-11 Martin Quinson
1539 - Reput hook for raw sockets, needed for BW experiments
1540 - kill a few lines of dead code
1541 [Data description] Interface cleanup
1542 - gras_datadesc_by_name returns the searched type or NULL.
1543 That way, no variable is needed to use a type desc once, which makes
1545 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1546 two parameters were strings (field name, type name), leading to
1548 [Dicos] Interface cleanup
1549 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1550 This is consistant with the dynar API.
1552 2004-04-21 Martin Quinson
1554 - Porting to new standards.
1556 - interface cleanup.
1557 There is no bag anymore, no need to take extra provision to mask the
1558 pointers behind "ID".
1559 Better splitup of functions between files create/exchange/convert.
1560 This is still a bit artificial since convert and receive are so
1561 interleaved, but anyway.
1563 - add a queued message list to procdata (the ones not matching criteria
1565 - factorize some more code between SG and RL wrt procdata
1567 - use gras_exit in example to track memleaks
1568 - get rid of gs_example now that GS is properly integrated into gras
1569 - update run_test to integrate the lastest tests (datadesc)
1571 - rename WARNINGn macros to WARNn since it prooved error-prone
1573 2004-04-19 Martin Quinson
1575 - register init/exit functions within gras module mechanism
1576 - send/receive function.
1577 Convertion is not implemented, but short-cutted if not needed.
1578 struct/array elements are sent one by one (instead of block-wise), but
1579 nobody really cares (yet). Get a prototype before optimizing.
1580 - tests (using a file socket) for DD send/receive on:
1581 - base types: int, float
1582 - array: fixed size, string (ie ref to dynamic string)
1583 - structure: homogeneous, heterogeneous
1584 - chained list, graph with cycle
1585 Believe it or not, valgrind is not too unhappy with the results. The
1586 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1590 - Bugfix when using a filename explicitely (instead of '-')
1592 2004-04-09 Martin Quinson
1594 - factorize more code between RL and SG in socket creation
1595 - Complete the implementation and tests of:
1597 o file (only in RL, and mainly for debugging)
1599 I lost 3 days to design a portable address resolver, and then decided
1600 that the prototype mainly have to run on my box.
1601 Addressing portability too early may be like optimizing too early :-/
1603 - use gras_init in the Tests instead of the crappy parse_log_opt
1604 (the latter function is removed)
1605 [Conditional execution]
1606 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1607 [Code reorganisation]
1608 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1609 Build examples against the RL library, since there is no way to disable
1610 its creation for now.
1612 For information, the beginning of coding on GRAS was back in june
1613 2003. I guess that every line has been rewritten at least twice since