1 SimGrid (3.6) unstable; urgency=low
4 * New model for multi-core CPUs. You can now use the core attribute to
5 precise the number of cores of a host. This is a basic model. Every
6 process running on the host receive at most the power provided in
7 the DTD (throughput<=power). Total throughput of process cannot exceed
9 * New peer tag. This peer tag creates a tiny AS comprising a host and a
10 router linked by an up-link and a down-link (possibly asymmetrical).
11 This kind of pattern allows to easily build last-mile model style platforms.
12 Aggregating such patterns in a rule-based AS is thus the technique of
13 choice for modeling large peer-to-peer/volunteer computing/cloud platforms.
14 * New model for Vivaldi routing. We transformed the Vivaldi network model
15 into a Vivaldi routing model (based on the rule-based model). This allows to
16 combine Vivaldi based latencies with last-mile platforms.
19 * Completely rewrote this module to allow parallel execution.
22 * Add a new function xbt_dynar_to_array that transforms a dynar into a
23 NULL-terminated array. This may solve backward compatibility issues
24 due to the change to return type of SD_Simulate. See also:
25 http://lists.gforge.inria.fr/pipermail/simgrid-user/2010-December/002206.html
28 * New configuration options
29 Options triva/categorized and triva/uncategorized can be used to generate
30 graph configuration files for Triva visualization tool.
31 * Configuration option tracing/platform is renamed to tracing/categorized
32 * XBT logging makes tracing error checks easier, new root log hierarchy: instr
34 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
36 SimGrid (3.5) stable; urgency=medium
39 * New feature to any SimGrid-based simulator: Model-Checking
40 Check SIN#1 for more details.
43 * New Model SMPI (three-interval linear regression for correction factors)
44 See RR-7426, available at http://hal.inria.fr/inria-00527150
45 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
46 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
47 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
48 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
49 * See also src/smpi/README
54 - Tracing API changes: TRACE_start and TRACE_end should not be called
55 by user-code. They are automatically called by simulators created
56 with SimDAG, MSG and SMPI if the toolkit is compiled with
57 tracing_enabled=ON. Categories declaration and utilization remain the
58 same for MSG and SimDag.
59 - A function was added to the tracing API to declare categories with
61 - TRACE_category_with_color (char *category, char *color)
62 where color must be in the following format
63 "%f %f %f", red, green, blue
64 and red, green, blue are float values in the interval [0, 1]
65 - User can specify NULL as color parameter, or continue calling
67 On that case, the tracing system will define random colors
68 - The following command-line options are supported:
69 --cfg=tracing/filename:msg.trace
70 --cfg=tracing:1 (activate tracing, needed to use others)
71 --cfg=tracing/platform:1 (categorized resource use)
72 --cfg=tracing/uncategorized:1 (uncategorized resource use)
73 --cfg=tracing/msg/task:1 (task creation)
74 --cfg=tracing/msg/process:1 (process creation, migration)
75 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
76 --cfg=tracing/smpi:1 (SMPI interface tracing)
77 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
78 - examples of examples/msg/tracing updated
80 - DAXLoader and DOTLoader functions can generate tasks with categories
81 - A new function to attribute a category to SD tasks:
82 TRACE_sd_set_task_category (SD_task_t task, char *category)
83 * Tracing the MPI interface implemented by SMPI
84 - Collective operations are traced with states
85 - Point-to-Point operations are traced with states/links
86 - Tracing activated by a parameter "-trace filename" passed
87 to smpirun during execution (considering that simgrid
88 is compiled with tracing enabled)
89 - To run the simulation with gdb, the simulator
90 accepts --cfg=tracing/smpi:1 to trace SMPI
91 - tesh files to check if smpi tracing is ok
92 - See examples/smpi/NAS/DT-trace
93 * GTNetS tracing re-worked
94 - adaptation to the tracing system of GTNets to cope
95 with modifications regarding the fullduplex mode
96 - new tesh files to check if gtnets tracing is ok
99 * Asynchronous communications through the functions:
100 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
101 * New function: MSG_load_platform_script()
102 to make possible using a lua script instead of XML files to set up platforms
103 * New function: MSG_set_function
104 to associate functions to processes, used when bypassing the parser
105 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
107 Platforms: Add some more examples in examples/platforms
108 * Grid'5000: see www.grid5000.fr
109 * *_30000_hosts.xml: various huge files [mainly scalability testing]
112 * Change the XML format. This is a very important modification. SimGrid
113 3.5 introduces a new hierarchical format based on the notion of
114 Autonomous Systems. Compatibility with old format is ensured
115 through the perl script provided in the install bin directory
116 bin/simgrid_update_xml.
117 It is now possible to build platforms with specific routing
118 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
119 connect several platforms together. We will try to provide soon
120 set of realistic platforms exploiting these properties (have a
121 look at examples/platforms/ for the moment).
122 * Take the opportunity of the XML format change to be a good XML citizen:
123 rename link:ctn to link_ctn and similar changes (also dealed with by
125 * Add a new routing scheme (rule-based) using regular expressions. It
126 enables to have an extremely low memory footprint when the
127 underlying routing is simple and can be compactly described. You
128 need to have libpcre4-dev (perl regular expressions) installed if
129 you want to use this routing scheme.
130 * Revive the cluster TAG and allow to easily and efficiently (both in
131 term of memory and speed) connect clusters together. Have a look
132 at teshsuite/simdag/platforms/ to see how this can be done. With
133 this tag, you can create clusters with thousands of tasks at no
134 cost (have a look at examples/platforms/).
135 Note: clusters are implemented as ASes, so there is no need for an
136 enclosing AS tag if you have only one cluster in your platform.
137 * Add new generic functions in the public interface that allows the user
138 to call SURF 'create_resource' methods from your code (same
139 functionality as the XML bypass mechanism but with a much lighter
141 * Add a new model (enabled through command line --cfg=network/model:SMPI)
142 that uses a piecewise linear approximation to produce better
143 results when exchanging small messages.
144 * Add a new parameter to handle correctly full duplex link and account
145 for interferences between uplink and downlink communications
146 (activate with --cfg=fullduplex:1).
149 * Rename the SD_READY (all dependencies are satisfied and task is
150 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
151 dependencies are satisfied) state.
152 This prevents a confusion between the notion of "ready to schedule"
153 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
154 simulated" (SD_RUNNABLE) used by the simulation kernel.
155 * Change the way a task is considered as ready. Instead of removing
156 dependencies when a task is done, a counter is decreased. This way,
157 it is always possible to reach ancestors thanks to the
158 SD_taks_get_parents function (even after the end of the simulation.)
159 * Change the return type of SD_Simulate from (SD_task_t*) into
160 xbt_dynar_t. This function was in handling a dynar internally and
161 converted it into a NULL terminated array for historical reasons.
162 * New function SD_dotload(char*) to load a DAG described in dot
163 format. This loader and the corresponding examples require the
164 installation of the graphviz library.
165 * Fix a bug in the management of tasks of size 0 in the surf network
166 models. This problem was only visible with SIMDAG and you should
167 thus disregard results produced with earlier versions if you
168 relied on this feature (some tasks were blocked because of this).
169 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
170 even though your applications were doing only point-to-point
171 communications and sequential computations. Now you can really use any
172 model you want (of course, if you create real parallel tasks, which are
173 not implemented in most models beside ptaskL07, this will abort).
174 * Add an example that schedules a DAX on an heterogeneous platform
175 using a Min-Min strategy.
176 * New function SD_workstation_get_current_task() that returns the kind
177 of task currently running on a workstation in the sequential access
179 * Raise some warnings when unexecuted tasks remains at the end of the
180 simulation. This is usually caused by cycles in the DAG.
183 * New function: SIMIX_process_set_function() called by MSG_set_function
184 * Change the underlying waiting queue in semaphores so that a process
185 can wait on several of them simultaneously (as in waitany).
186 * Fix the way to handle tokens in semaphores so that all access patterns
187 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
188 * kill the dirty pimple SIMIX_message_sizes_output()
189 Please use (proper) visualization instead
192 * New data container: setset (set of sets of elements)
193 * New module: mmalloc (mapped malloc, allowing to have several
194 independent segments of malloc)
195 * New function: xbt_dict_cursor_set_data()
196 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
197 * New function: xbt_dynar_is_empty()
198 * New function: xbt_fifo_get_last_item()
199 * Fix xbt_dynar_shrink(): use the right element size.
200 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
201 uninitialized areas during expand.
202 * Fix semaphores: previous implementation was severely broken.
203 * Use library init/fini functions for our initialization.
204 - you can use logs and other feature as soon as you want in your
205 code (even before the xbt_init / MSG_init)
206 - xbt_exit is now a no-op and produce a warning when used.
209 * Port GRAS to new SIMIX mechanisms. This allows gras users to
210 benefit from the latest improvement to the simulation kernel.
211 * Kill measurement sockets for now. If you rely on them, sorry. This
212 release is not for you. This feature will be reintroduced in the
213 future, but we cannot delay the release any further.
214 * New function: gras_msgtype_get_name().
215 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
216 * Fix (at last) the pmm example: it should not randomly fail anymore.
218 Build chain: bug fixes and overall polishing
219 * Cmake is now stable enough. Hence, we killed the autotools.
220 * Port to windows ( TM :)
221 * Fix the 'make install' target.
222 No need to use 'make install-simgrid' anymore
223 * Introduce a 'make dist' target compiling a *source* archive
224 'make package' compiles a binary archive
225 * Compile java files only on need
226 * Add --cd and --setenv command line options to tesh
227 * Out of source builds are not fully supported yet, but we are close
228 * Enable supernovae and optimization flags by default for our users
231 * Add layer to set up environment directly from lua, without XML.
232 * The effect of gras_stub_generator can be achieved through
233 lua too (check examples/gras/console/ping_generator.lua)
235 -- Wed, 01 Dec 2010 22:09:23 +0100 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
237 SimGrid (3.4.1) stable; urgency=low
239 The "Polishing easter eggs is probably a good idea" release.
240 This is a bug fixes release only.
244 * Fix a bug preventing the tasks from begin garbage collected.
247 * Fix a bug occuring when a host involved in a communication fails.
248 This was not detected properly by the other peer involved in the
249 communication. Now, it's reported as a network error.
252 * Warn the user about loop dependencies in data flow of DAX files
253 * Obey the control-flow dependencies of DAX files
256 * Add option "enable_smpi" allowing to not compile SMPI.
257 Probably useful for the (Mac) users experiencing a build error here
258 * Improve the detection of lua5.1 and ruby1.8
260 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
262 SimGrid (3.4) stable; urgency=low
264 The "Easter in Cargese" release. Also known as (major changes):
266 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
267 ~> bindings were greatly improved
268 ~> new tracing infrastructure for better visualization introduced
270 * the "Welcome to configury modernity" release.
271 ~> we switched from autotools to cmake, and improved our cdash
274 A more detailled list of changes follow (full detail in svn log).
276 Java Bindings: Various Cleanups
277 * (install java-gcj-compat-dev on debian-like to use them)
278 * Remove put/get: no need to export deprecated interface in Java
279 Use send/receive instead.
280 * Cleanup the examples and add a README per directory
281 * Remove example autoDestination (that's the only way to go now)
282 * Remove example explicitDestination (was a plain copy of basic)
283 * Make JniException a runtime exception, so that there is no need to
284 declare the fact that you may encounter such a beast. I guess that
285 nobody will ever want to survive such error.
286 * Create specific errors for each MSG case of failure:
287 host failure, transfer failure, timeout, task cancelled
288 * Cleanup the exceptions that may get thrown by each function
289 * Other internal cleanups in Java bindings. Performance still bad :/
290 Ruby and Lua Bindings: create them
291 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
292 * That's new and great, you should try them out.
293 Same functionalities than Java bindings, only even less polished
295 * Kill the useless "rate" argument of SD_task_get_execution_time()
296 Everyone used to provide -1 as a value, it was not used, and the
297 semantic of a possible use wasn't even clear.
298 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
299 either as comm costs or compute costs to mean that there is no
300 such thing for that specific task.
301 * Add a SD_task_set_name() function
302 * Fix SD_task_unschedule() on typed tasks
303 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
304 * In DAX loader, accept useless 'level' attributes to <job> since
305 LIGO DAGs have them (seem to be to ease graphical representation).
307 * Add an example masterslave_mailbox.c using send/receive and not
308 the deprecated put/get interface.
309 * Kill the MSG_paje_output() function. It's a noop since 2 years.
310 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
312 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
313 (declare MSG_USE_DEPRECATED to still have the old name)
314 * Add a MSG_task_set_data() function
315 * About trace replay (see examples/msg/actions):
317 - Allow to work with splitted trace files for each process
318 Give the specific trace file as argument of each process,
319 and call MSG_action_trace_run(NULL)
320 You can still have one merged file for all processes.
321 - Fix implementation of collective operations
322 * Allow task_execute() on 0-sized tasks (closes #10063)
324 * This is the first release of SimGrid where SMPI is not considered
325 beta anymore (even if some corners should still be improved)
326 * Port over the new SIMIX_network submodule (internal refactoring)
327 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
328 * Implement more missing elements of the standard:
330 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
332 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
333 MPI_Scan() MPI_Get_processor_name()
334 - Added implementation of missing case for Alltoall (warning: it's
335 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
336 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
337 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
338 SMPI_MPI_Allgatherv()
340 - MPI_Waitsome() was broken
341 - Allow relative includes in smpicc
342 - Command line cfg argument 'reference_speed' was ignored...
343 - Some functions did not properly lead to auto-benching of user code
344 - smpicc passes -O2 by default (just like openmpi one)
346 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
347 * Bug fixes about timeouts during communications
348 * add SIMIX_message_sizes_output() as a pimple to write to file the
349 amount of messages per size. Use gnuplot to get histogram.
350 Pimple because that's the only user-visible function of simix,
351 defined directly in xbt.h (irk, sorry)
353 - Add a SIMIX_sem_get_capacity() function
354 - Fix interactions with processe resume/suspende
355 - release_forever() was stupidly broken
356 - Fix SIMIX_display_process_status() for processes in a semaphore
357 - Make SIMIX_sem_block_onto() user-visible
358 * Refactoring context stuff:
359 - Use pseudo-OOP for better modularity
360 - reimplement SIMIX_process_kill() without process_schedule() so
361 that the latter can take as invariant that it is called from
363 - Merge context_start into context_new for sake of simplicity
365 * Add a Vivaldi network model, coded live during SUD'10 ;)
366 * Rename configuration variables to start a hierarchy:
367 o cpu_model -> cpu/model
368 o network_model -> network/model
369 o workstation_model -> workstation/model
370 * New configuration variables:
371 o network/bandwidth_factor: correction to bandwith
372 o network/latency_factor: correction to latency
373 o netwotk/weight_S: correction to the weight of competing streams
374 * Add a long description to the models, that users can see with such
375 argument on the command line: --cfg=cpu/model:help
376 * --help-models display the long description of all known models
378 * config: add the ability to set a default value after registration
379 Does not override any previously set value (e.g. from cmd line)
380 * dict: allow to have integer key and data.
381 When so, you need to use the following functions
382 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
383 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
384 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
385 In contrary to regular dicts, the key is not malloced before copy.
386 Mixing scalar and regular elements in the same dict is not tested
388 * Allow to use xbt_dynar_shrink() to expend the dynar instead
389 Tracing for Visualization:
390 * SimGrid is now instrumented in order to generate a trace file for
391 visualization analysis: to use it, need to compile SimGrid with the
392 "tracing" option enabled, and instrument the program using SimGrid with
393 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
394 (among other functions).
395 * The instrumentation only traces the platform utilization for now
396 * Documentation to use the tracing functions and how to analyze the
397 traces with the Triva tool is written.
398 * More information about: SimGrid FAQ (in the section Tracing Simulations
401 * We moved to cmake as default build system. Autotools support will
402 be dropped soon. Check the FAQ for more info about how to use it.
403 * Greatly improved our cdash/ctest interactions
404 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
405 * Added memory checking tests with valgrind; lot of memleak fixing.
406 This may be the first release of simgrid with so few memory issues
407 * Added code coverage tests.
408 Our coverage is still improvable, but at least we see it on cdash.
410 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
412 SimGrid (3.3.4) stable; urgency=low
414 The "Desktop Grid needs love too" release (also called Xmas release).
417 * Major speedup in the maxmin system solving by using lazy evaluation
418 Instead of solving completely the maxmin system at each iteration,
419 only invalidate (and recompute) the modified parts.
420 This new feature is enabled in default models but you can try to
421 turn it on with "--cfg:maxmin-selective-update=1" for other models.
422 * Cas01 IMproved as default CPU model
423 This CPU model is the same Cas01 model, but it uses the
424 maxmin-selective-update flag and a heap structure to manage
425 actions on SURF kernel.
426 It reduces the complexity to find the next action to finish and,
427 consequently, it's faster than the old Cas01.
428 This is the new default CPU model (Cas01).
429 * Rename the old Cas01 model to Cas01_fullupdate
430 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
431 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
432 * CpuTI (CPU Trace Integration)
433 A new CPU model whose objective is simulate faster when using
434 availability trace files.
435 Instead of using a full featured, over engineered maxmin system for
436 CPU modeling, this model does the pre-integration of traces files
437 to calculate the amount of CPU power available, and so, executes
438 faster than the old CPU models.
439 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
440 * Use LV08 as default network model since it gives better accuracy
441 for small messages and shouldn't change things for big ones.
442 Use --cfg=network_model:CM02 to get the previous behavior.
445 ******************************************
446 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
447 ******************************************
448 * The new CPU model may changes simulations!
449 The point is that events occurring at the exact same timestamp
450 are not scheduled in the same order with the old and new
451 version. This may be enough to completely change the execution
452 of simulations in some cases.
453 * The new network model will change simulations!
454 This new model is more realistic than the previous one, so you
455 should consider redoing your old experiments with this model.
456 Sorry for the inconvenience.
459 * Introduce the supernovae compilation mode
460 When compiled that way, the whole SimGrid (or almost) is put in a
461 single compilation unit and compiled in one shoot.
462 This is to help gcc which has difficulties to inline stuff from one
464 The speedup seem to be above 15%, althrough more tests are needed on
465 amd64 to confirm that gain.
468 * Port of MSG's mailbox on top of SIMIX network
469 The put/get mechanism was greatly simplified on the way.
472 * New SIMIX network module. Provides:
473 - Mailbox: rendez-vous mecanism to find with who you want to speak
474 - Synchronous send/recv: easier and hopefully faster since the
475 logic is handled in the maestro process directly now
476 - Asynchronous send/recv: you dreamt of it? It's here now
477 Too bad that nobody cared enough to propagate the change to MSG.
478 * Add semaphores as SIMIX synchronization mechanism.
481 * new function SD_daxload(char*) to load a DAX file
482 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
483 * Introduce typed tasks. Specify its kind and cost at creation.
484 At scheduling, just give where it should be placed, and the cost
485 for each involved resource is automatically computed.
486 Existing constructors so far (more to come of course):
487 - SD_task_create_comm_e2e() for end-to-end communication
488 - SD_task_create_comp_seq() for sequential computation
489 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
490 * new function SD_task_dump() for debuging display
491 * new function SD_task_dotty(task,FILE*) writing to file the info
492 about the task in dotty format
493 * SD_task_dependency_exists() can now cope with having one of its
494 arguments NULL. If so, it tests whether the other argument has any
496 * Add getters on list of preceding/following tasks:
497 SD_task_get_parents(task) and SD_task_get_children(task)
498 * Add getters on amount of workstations and list:
499 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
500 * Add getter on task kind: SD_task_get_kind(task)
501 * Update the start_time and finish_time of tasks on completion/failure
502 * Bugfix: Remove task from state swags when destroyed
505 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
508 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
509 dynar with one separate thread per value of the dynar.
510 * Change the prototype of xbt_thread_create(), sorry.
511 Added a boolean parameter indicating whether we want to join this
512 thread (used in SG only for now)
513 * Implement xbt_thread_join and xbt_thread_yield in SG also.
516 * GTNetS wrappers should now be usable again (and betterly tested too)
517 * Fix a major regression from 3.2 where the timeout provided to
518 MSG_task_put_with_timeout() was used as absolute time before which
519 the comm should be done.
520 * Start to fix the <cluster> tag.
521 - Internal links should be good now (beside of the loopback, which
522 use the private link instead)
523 - paths to the external world is still rather broken
524 - the <route:multi> tag is just broken. Actually that's brain-dead.
525 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
526 to make it less stupid
527 ** Check your platform with teshsuite/simdag/platforms/flatifier **
528 * Fix a source-level compatibility glitch from 3.2: after defining
529 MSG_USE_DEPRECATED, you can use the old name
530 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
531 * Allow to compile from the SVN with automake 1.11
532 * Fix some problems when using the "start_time" tag in deployment XMLs.
533 * Fix #8569: XBT/synchro.h has redundant declarations
534 * Fix #8563: MSG return values and exceptions
535 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
536 * Integrate patch #8636: Obey DESTDIR when installing documentation.
537 Thanks to Robson Peixoto.
538 * Fix a vicious bug in dictionaries inducing that some elements were
539 not freed on xbt_dict_free()
541 Portability report of this version:
542 * Main portability targets:
543 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
545 - mac leopard on i386
546 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
547 but nothing critical.
548 * Other platforms: windows, AIX and others were not tested for this release
550 Timing report of this version:
551 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
552 scenario parameters). From 8h to a few seconds in desktop grid settings.
553 * Supernovae brings about 25% speedup on i386.
555 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
557 SimGrid (3.3.3) stable; urgency=low
559 The "Need for Speed" release.
561 The timings done to validate the 3.3.2 were faulty.
562 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
564 The problem was a conversion from a manually handled vector to
565 xbt_dynar_t on the critical path.
566 xbt_dynar_foreach calls functions, inducing stack management crap.
568 We inlined these functions and xbt_dynar_foreach is now breath taking.
569 We also inlined xbt_swag_belong on the way.
571 Here are some approximate speedup measurements (on master/slaves
572 simulations lasting between 10s and 20s each):
573 3.3.1 -> 3.3.2: about same performance
574 3.3.2 -> 3.3.3: 40% speedup
575 3.3.1 -> 3.3.3: 40% speedup
576 3.3.1 with inline patch -> 3.3.3: 30% speedup
578 Our reading is that the refactoring which occurred in 3.3.2 made us
579 suffer much more from the xbt_dynar_foreach low performance, but
580 once we solved this, this refactoring proved to be very performance
581 effective. From the 40% speedup, somehow, 10% are due to the
582 inlining and 30% to the refactoring.
584 That's a pitty that gcc cannot inline functions placed in other files
585 alone. We have to choose between:
586 - break the encapsulation (by putting private data structures and
587 accessors in headers files to help gcc)
588 - live with low performance
589 - switch to a decent compiler such as icc (not quite possible).
591 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
593 SimGrid (3.3.2) stable; urgency=low
595 The "Simplicity does not preceed complexity, but follows it" release.
597 The main contributors of this release were (lexical order):
598 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
601 * Extract the routing logic into its own object.
602 (was dupplicated in network.c and workstation_LV07.c;
603 Allows to implement other ways of storing that info)
604 => kill now useless network_card concept
605 - Use dynar to represent routes (instead of void** + int*)
606 - kill link_set (use surf_network_model->resource_set instead)
607 - Add a command-line option to choose the routing schema to use
608 - Add three new models:
609 * Floyd (shortest path computed at initialization)
610 * Dijikstra (shortest path recomputed all the time)
611 * Cached Dijikstra (shortest path computed on need)
612 All these models where contributed by Silas De Munck, and are
613 described in his ICCS09 paper.
615 * Simplify model declaration
616 (less redirections, less function to write when defining a model)
617 - Factorize stuff between models:
620 surf_model_resource_set(model)
621 surf_model_resource_by_name(model, name)
622 - Unify the types of models in s_surf_model_t (using an union)
623 - Embeed fields of common_public directly into s_surf_model_t
624 - Rename model methods:
625 action_free ~> action_unref
626 action_change_state ~> action_state_set
627 action_get_state ~> action_state_get
628 - Change model methods into functions :
629 (model)->common_public->action_use ~> surf_action_ref
631 * Implement a generic resource; use it as ancestor to specific ones
632 (allows to kill duplicated code in models)
633 Drawback: timer command don't need no name nor properties;
634 workstation_CLM03 don't need no properties
635 (but I guess we can live with those few bytes wasted)
637 * Improve the action object model
638 - implement a constructor avoiding dupplicated code about field
639 initialization in generic_action part.
641 * Kill the SDP model: it has an external dependency, is deprecated
642 in flavor of modern lmm models, and didn't compile since a while
645 * Relocation of the context module from XBT to SIMIX.
646 (the context were decoupled from the simix processes, duplicating a lot of code)
647 => a lot of code was factorized
648 - less overhead is introduced during scheduling
649 - simpler API for the context factory
650 - the logic for process creation,destruction and manipulation was simplified
651 * Simplification of the s_smx_process_t data structure.
652 => accesing the simix level data associated to a process is faster now,
653 and the code is a lot more readable.
656 * Implement some more MPI primitives:
657 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
658 -implementation: Bcast: flat or 2-ary tree (default),
661 Allreduce: Reduce then Bcast
662 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
663 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
664 Alltoallv: flat tree, like ompi
666 * Add support for optimized collectives (Bcast is now binomial by default)
667 * Port smpirun and smpicc to OS X
670 * Kill SD_link_get_properties: hard to maintain and makes very little sense
671 Shout out if you used it.
674 * Display the list of still queued messages in SG mode when existing
678 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
679 * Add xbt_graph_node_get_outedges() [Silas De Munck]
680 * Add xbt_str_from_file(FILE*)
681 * Add xbt_dict_get_key achieving a linear reverse search
682 * Remove the context module
684 Portability report of this version:
685 * Main portability targets:
686 - Linux(debian)/x86/context
687 - Linux(debian)/x86/pthreads
688 - Linux(debian)/amd64/context
689 - Linux(debian)/amd64/pthreads
690 On these, we still have the eratic breakages of gras/pmm and
691 amok/saturate_sg reported in previous version. We still think
692 that the tests are the cause of the fault, not the tested code.
694 - Mac OSX Leopard/x86/context
695 Still false negative in tesh autotesting.
696 Smpi still fails, but this time because readlink does not accept -f
697 Everything seems to work properly beside of that.
700 - AIX version 5.3 (only tested contexts this time)
701 Smpi still fails there because mktemp is not installed.
702 Everything seems to work properly beside of that.
704 I managed to compile it for the first time, but several breakages.
705 Won't delay the release for this exotic platform.
707 * Windows: it's still lagging behind. If you want to help, please
710 Timing report of this version:
711 This version seem to be more than 5% faster than 3.3.1 (on linux
712 64bits with contextes). The gain is less than expected, we are
713 investigating this for next release.
715 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
717 SimGrid (3.3.1) stable; urgency=low
720 * Implement a --cfg-help to show existing configuration variables
721 * Build chain do not require doxygen in maintainer mode
724 * fix a bug on struct sizeof computation, which prevented the
725 exchange of arrays of structs in some conditions
726 - added a regression test about this in datadesc_usage
727 * Allow the exchange of 0-long dynamic vectors.
728 - for that, use -1 as indicator of dynamic size instead of 0
729 - This implied to change any size from unsigned long to long,
730 reducing a bit communication abilities, but I guess that with
731 64bits being quite common, this is more than enough.
732 - This also induce a protocol change, thus bumping network protocol
733 version from 0 to 1 (if we have external users, we have to get
734 clean on that point too ;)
735 - added two regression tests about this in datadesc_usage
736 * Be more verbose when propagating local exceptions
737 This helps debugging.
738 * Display the status of simulated processes when receiving SIGINT in
742 * Allow to control the simulation from a trace file.
743 New functions MSG_action_register() and MSG_action_trace_run()
744 The first one allows to associate a function execution to each
745 kind of action while the second one parses a trace file and
746 triggers the corresponding actions within the system.
747 For now, only a toy example is provided in examples/msg/actions
748 * Add an exemple of process migration in examples/msg/migration
749 * Fix a bug in task exchange which broke MSG_task_get_sender()
750 Add a teshsuite regression test for that.
751 [Bug: if MSG_task_get_sender() is called after sender exit,
753 * Fix a bug which prevented suspend/resume to work properly
754 * Display the status of simulated processes when receiving SIGINT
755 This fixes a regression of v3.3. due to the introduction of SIMIX
756 * Bug fixing in failure management:
757 - trace could not start by a failure at time 0
758 - failure during communications were not working
761 * Add SIMIX_process_set_name() to change the name of the current
762 process in the log messages.
763 * Store smx_hosts in a dict since we only retrieve them by name
764 * Move the configuration infrastructure to surf
767 * Move the configuration infrastructure to surf
770 * Massive internal cleanups:
771 - Store internal structures on processes instead of hosts (allows
772 to have more than one process per host, in addition of being more
774 - Cleanup the initialization/finalization process
775 - Kill a whole bunch of unneeded synchronization:
776 processes run in exclusive manner within the simulator
777 - Move queues from global tables to process data fields
779 - now accept -platform and -hostfile arguments
780 - Pass the right rank value to processes according to the hostfile
781 * Compile the examples by default, and use them as regression tests
782 * Implement MPI_Wtime()
783 * Change the reference speed to a command line option
786 * TCP_gamma can now be specified as command line option using
787 --cfg=TCP_gamma:10000000.0
788 * Change the --surf-path cmd line option into --cfg=path:
791 * Also include strbuff from xbt.h public header
792 * xbt_ex_display(): do not free the exception after displaying
793 This allows to do more with the given exception afterward.
794 Users should call xbt_ex_free() themselves.
798 Portability report of this version:
799 * Main portability targets:
800 - Linux(debian)/x86/context
801 - Linux(debian)/x86/pthreads
802 - Linux(debian)/amd64/context
803 - Linux(debian)/amd64/pthreads
804 These targets fail about 1/10 of times on gras/pmm, but we believe
805 that this is because of the test, not because of simgrid.
806 amok/saturate_sg fails even more rarely, and the test may not be
809 - Mac OSX Leopard/x86/context
810 The test suite still spits tons of errors because some obscure
811 force prevents us from removing the temporary directories
812 arguing that they still contain some metadata I've never heard of.
813 Smpi fails because seq is not installed.
814 Everything seems to work properly beside of that.
817 - AIX version 5.3 (both contexts and pthread)
818 Smpi still fails there because mktemp is not installed.
819 XML inclusions seems rosty on AIX.
821 * Windows: it's still lagging behind. If you want to help, please
824 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
826 SimGrid (3.3) stable; urgency=high
830 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
833 * Introduce the SIMIX module: factorize code between MSG and GRAS.
836 Until now, GRAS were using MSG as an interface to SURF. It was
837 quite difficult because both interface have several differences
838 (MSG channels vs GRAS sockets were the most notable point).
840 This also opens the gate to SMPI (which should occur soon) and speed
841 up simulations by to 40% (even if it were not the main goal).
843 **************************************
844 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
845 **************************************
846 The point is that events occuring at the exact same timestamp are
847 not scheduled in the same order with the old and new version. This
848 may be enough to completely change the execution of simulations in
849 some cases. Sorry for the inconvenience.
851 * Cleanup and upgrade the XML format to push further scalability
852 issues (check http://hal.inria.fr/inria-00256883/ for more info)
854 * Improve the testing infrastructure with tesh. Now a very large part of
855 the code is tested not only by being run but also by checking that the
856 output match an expected output [Mt].
858 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
859 is really less memory-demanding, which should allow you to use
860 larger files in SimGrid [AL].
862 * Inform valgrind about our contextes, so that it becomes usable
863 with the default (and more effecient) version of SimGrid
864 [contributed by Sékou Diakite, many thanks]
867 * Introduce a listener thread in charge of receiving incomming
868 messages from the network. It allows to overlap communication and
869 computation but most notably, it removes some stupid deadlocks due
870 to the fact that so far, a process could not send and receive at
871 the same time. This made most non trivial communication schema
873 * Convert the PIDs from long int to int to match the MSG ones (and
875 * New function: gras_agent_spawn() to launch a new process on
876 current host. Only working in simulation for now. [Mt]
877 * New function: gras_os_hostport() returning a constant form (ie,
878 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
881 * Make the backtrace of exceptions more human readable [Mt]
882 * New module: xbt/str [Mt]
883 a ton of string utility functions (split, join, printf to a newly
884 allocated buffer, trim, etc)
885 * New module: xbt/hash [Mt]
886 SHA1 hashing algorithm (more to come if needed)
887 * New module: xbt/synchro [Mt]
888 synchronization tools (mutex and conditions) working the same way
889 in simulation and in real life (mainly useful for GRAS, but not
891 * New module: xbt/queue [Mt]
892 classical producer/consumer synchronization scheme
893 * xbt_dynar_new_sync() creates a synchronized dynar. All access
894 (using the classical functions will get serialized) [Mt]
895 * Make dictionary internal table dynamic. No need to specify its size
896 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
898 * Make sure the log channels are organized as a tree under windows
899 (because of ANSI C compatibility issue, any channel were child of
903 * Cleaned many thing in surf and fixed a few bugs [AL].
904 * Add a nice command line configuration mechanism to compose models [AL].
905 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
906 the previous one (KCCFLN05). It relies on something that looks like
907 a max-min sharing mechanism but cannot be written as such. A new solver
908 was thus designed [AL].
909 * Add a new solver to lmm. Based on Lagrange optimization and
910 gradient-based descent, it enables to efficiently maximise systems s.a
912 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
914 This solver enables to propose two new network models for TCP Reno and
915 TCP Vegas based on Low's work. These models still need to be fully
916 tested though [Pedro Velho].
919 * Bug fix in SD_simulate. Now the time bound given as argument is
921 * Use the new parallel task model (ptask_L07) as default.
922 * Use the SURF command line configuration mechanism.
923 * 0-size tasks (for synchronization) should now work.
925 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
927 SimGrid (3.2) stable; urgency=high
931 We still experience issues on this platform, but we believe that at
934 GRAS API BREAKAGE (for simplification purpose, sorry):
935 * the gras_msgtype_by_name is not used anymore. Instead of
936 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
937 you can write (and must)
938 gras_msg_send(toserver, "request", &request);
939 - If you still want to pass a gras_msgtype_t to the function (to cache
940 the type and avoid the lookup time), use the gras_msg_send_() variant.
941 - Impacted functions:
942 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
943 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
944 * The callbacks are now expected to return 0 when everything went well
945 (just like the main() function)
947 GRAS new features and improvements:
948 * New module mecanism where user code can use per process globals [Mt]
949 This is similar to gras_userdata_*() functions, but for libraries. It
950 factorize some code developped over and over in the examples and AMOK.
951 It has still to be documented and used (only amok/peermanagement is
953 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
954 loss when they were small enough to fit into the buffer and sent quickly
955 enough so that they can all get received in one shoot.
956 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
957 if not found. Use the *_or_null() variant for the old semantic.
958 * In gras_msg_handle, do not discard messages without callback.
959 They are probably messages to be explicitly awaited later (ie, proofs of
960 mis-synchronization in userland since they are sent before being awaited)
962 * gras_socket_meas_send/recv: semantic changed!
963 The numerical arguments used to be (1) the total amount of data to send
964 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
965 messages. This was need for the fool willing to send more than MAXINT
966 bytes on quite fat pipes.
969 * Do really rename the hostmanagement module to peermanagement. [Mt]
970 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
971 but this is rather new and this was documented in the module
972 documentation (poor excuses, I admit)
973 * Bandwidth measurement semantic changed! This follows the changes to
974 gras_socket_meas_send/recv explained above.
977 * A sequential mode has been added to the workstations. When a workstation
978 is in sequential mode, it can execute only one task, and the other tasks
979 are waiting in a FIFO. [Christophe Thiery]
982 * The KCCFLN05 workstation model now handles parallel tasks. It is the
983 model for SIMDAG. [Christophe Thiery]
984 * Bug fix in the maxmin solver: Some values were close to 0 instead of
985 equal to 0, which caused some bad behaviors in
986 saturated_constraint_set_update. I now use a threshold mechanism like in
990 * When running manually src/testall, you select specific units [Mt]
991 testall is the result of our cunit mecanism, and should replace all
992 the scripty thingy around since bash don't run easily on billware.
994 * A mallocator system has been added. [Christophe Thiery]
995 Mallocators allow you to recycle your unused objects instead of freeing them
996 and allocating new ones.
998 Documentation update:
999 * FAQ reworking + New FAQs:
1000 - "Valgrind spits tons of errors!" [Mt]
1001 - "How to repport bugs" [Mt]
1002 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
1003 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
1004 - Communication time measurement within MSG [AL]
1005 - I experience weird communication times when I change the latency [AL]
1006 * GRAS tutorial [Mt]
1008 - an introduction to the framework and to the used communication model
1009 - an initiatic tour introducing the most proheminent features:
1011 . Lesson 0: Installing GRAS
1012 . Lesson 1: Setting up your own project
1013 o Part 2: Message passing
1014 . Lesson 2: Exchanging simple messages
1015 . Lesson 3: Passing arguments to the processes (in SG)
1016 . Lesson 4: Attaching callbacks to messages
1017 . Lesson 5: Using globals in processes
1018 . Lesson 6: Logging informations properly
1019 . Lesson 7: Using internal timers
1020 . Lesson 8: Handling errors through exceptions
1021 . Lesson 9: Exchanging simple data
1022 . Lesson 10: Remote Procedure Calling (RPC)
1023 . Lesson 11: Explicitely waiting for messages
1024 . Recapping of message passing features in GRAS
1025 - A HOWTO section containing:
1026 o HOWTO design a GRAS application
1027 More are due, of course. They will come latter. In the meanwhile, you can
1028 check the examples which are still here.
1030 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
1032 SimGrid (3.1) stable; urgency=high
1036 There was a stack corruption somewhere, visible only when optimizing
1037 with these versions. [Vince]
1040 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1041 module that enabled to play with Directed Acyclic Graphs. It is built
1042 directly on top of SURF and provides an API rather close to the old
1043 SG. Some old codes using SG are currently under rewrite to check that
1044 all needful functions are provided. [Christophe Thiery]
1047 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1048 extension of the classical CLM03 model that gracefully handles
1049 failures. This is now the default model for MSG and GRAS. It doesn't
1050 handle parallel tasks yet though. [AL]
1051 * Bug fix: Weights were not correctly set in the network part.
1052 WARNING: This may have resulted in incorrect results with simulations
1053 where there are more than one flow on a given link. [AL]
1056 * After a (long ?) discussion on simgrid-devel, we have decided that the
1057 convention we had on units was stupid. That is why it has been decided
1058 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1059 WARNING : This means that all previous platform files will not work as
1060 such with this version! A warning is issued to ask users to update
1062 A conversion script can be found in the contrib module of the CVS, under
1063 the name contrib/platform_generation/surfxml_update.pl [MQ]
1066 * Bug fix: Processes were started in reverse order, wrt deployment file.
1067 WARNING: if your code relies on this bug, please fix it. [AL]
1068 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1069 being executed on two different locations. [AL]
1070 * Bug fix: Failures are now better supported thanks to Derrick's tests
1071 (there was many failure situations I hadn't thought of and that weren't
1072 correctly handled). [AL]
1073 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1077 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1078 Exception occurring on server-side are propagated back to client (!).
1080 API CHANGE: the callback changed their prototype. Change:
1081 int my_handler(gras_socket_t expeditor, void *payload_data) {
1083 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1084 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1086 * New! function: gras_msg_handleall to deal with all messages arriving
1087 within a given period.
1088 * New! function: gras_socket_server_range to get a server socket in a
1089 range of port numbers (ease to avoid port number conflicts) [MQ]
1090 * New! gras processes display their backtrace when they get a SIGUSR1
1091 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1092 Sweet to debug RL processes [MQ]
1096 - Do not force experiment sizes to be expressed in kb, or it becomes
1097 impossible to measure the latency this way (needs one byte-long tests)
1098 WARNING: this changes the amok_bw_* function semantic. [MQ]
1099 - Implements the link saturation stuff. [MQ]
1100 * Peer management module:
1101 New! module factorizing code that we wrote over and over [MQ].
1104 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1105 - Test units are placed directly into the library code, they get extracted
1106 automatically and placed into the src/testall binary.
1107 - Convert most of the XBT tests to this system.
1108 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1109 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1110 live in the same C code. This required to change a little bit the API
1111 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1112 * New module: sparse graph structure with basic algorithms (this is work
1113 in progress and the API is not considered to be frozen yet). [AL]
1114 * Display more information on backtraces: source line & function names are
1115 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1116 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1117 * Reworked a little bit some #include statements to load only required
1118 headers. Some user code that relied on SimGrid to include stdlib or
1119 stdio may need to include it by themselves. [AL]
1120 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1122 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1123 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1124 but I guess that almost nobody used those parts. [MQ]
1126 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1128 SimGrid (3.0.1) stable; urgency=low
1131 * Unfortunately, I had missed 5 misnamed functions:
1132 xbt_fifo_item_t xbt_fifo_newitem(void);
1133 void xbt_fifo_freeitem(xbt_fifo_item_t);
1134 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1135 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1136 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1137 They're now deprecated. Please use their new versions:
1138 xbt_fifo_item_t xbt_fifo_new_item(void);
1139 void xbt_fifo_free_item(xbt_fifo_item_t);
1140 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1141 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1142 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1144 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1145 * Documentation: xbt_log module unmercifully reworked [MQ]
1146 * Bugfix: there was a problem with the ending of contexts with
1147 the pthread backend. It caused some weird deadlock or behavior
1148 depending on the pthread implementation. [AL]
1149 * Bugfix: get the exceptions raised in the simulator repport where
1150 and why they come from when they are not catched in time [AL, MQ]
1153 * Bugfix: Do repport the error when two non-connected hosts try to
1154 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1157 * Add additionnal checkings on communications. Assert that two
1158 communicating hosts are connected by a set of links... [AL]
1161 * Add additionnal checkings on channel values in communication [AL]
1162 * New: MSG_task_get_source to see on which host a task was generated [HC]
1163 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1164 the number of tasks waiting to be received on channel and sent
1166 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1167 waits for the first task coming from a given host.. [AL]
1169 GRAS new functionnalities: [MQ]
1170 * Enhance the parsing macro to allow the size of multidimentional objects
1171 to be given thru annotations.
1172 * New example (and documentation): Matrix Multiplication a la RPC
1173 (as when I was young!) and fix a bunch of bugs found on the way.
1175 GRAS performance improvements: [MQ]
1177 * Reduce the amount of cbps creation/destruction by making it static to
1178 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1181 * Change libdata to a set so that we can search for stuff by ID (and thus
1182 reduce the insane amount of dict lookups)
1185 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1186 RL communication on top of SG since it's so uneffective.
1187 It may also be used for inter-thread communication in RL, one day.
1188 * Use gras_datadesc_copy() to exchange messages on top of SG
1190 - improve message exchange performance on top of SG
1191 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1192 * Don't exchange on the network the size of the used part of buffer,
1193 instead, specify the possible buffer size to read().
1195 - reduces the amount of read/write calls (one pair per exchange)
1196 - reduces the amount of exchanged data (the size)
1197 - allows to retrieve all arrived data on receiver side, if we don't need
1198 it right now (subsequent read will peek the buffer)
1199 - allows the receiver to proceed with the begining of the stream before
1200 everything is arrived
1201 - make it possible to build an iov transport (using readv/writev)
1203 - take care of the data with non-stable storage (like stacked data),
1205 * If possible, TCP send uses vector I/O (when writev() is here)
1206 - Don't use it for receive since we send data sizes and data on the
1207 same stream, so we wouldn't be able to chain large amount of chunks
1208 before having to flush the stuff to read the size.
1209 * Rework the transport plugin mecanism to simplify it and reduce the
1210 amount of pointer dereferencement when searching for the right function
1213 * I guess that now, we do almost as few system calls as possible while
1214 doing as few data copy as possible.
1216 To improve it further, we could try to send all the sizes first and then
1217 all the data (to use iov on receiving size), but it's only a partial
1218 solution: when you have 2 dimensional data, the sizes of the second
1219 dimension is data of the first dimension, so you need 3 streams.
1221 I'm not sure the potential performance gains justify the coding burden.
1223 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1225 SimGrid (3.00) stable; urgency=high
1228 * New! Give the possibility to hijack the surf parser and thus bypass
1229 MSG_create_environment and MSG_launch_application. Have a look at
1230 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1232 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1234 SimGrid (2.96) unstable; urgency=low
1239 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1240 This deprecates the xbt_error_t mecanisms.
1241 It modifies (simplifies) all XBT and GRAS API.
1242 MSG API keeps unchanged (exceptions raised by XBT are catched from
1243 within MSG and masked with existing error handling facilities)
1246 * New! Add a FATPIPE model. [AL]
1247 * New! Add a parallel task model. [AL]
1248 * New! Add automatically a loopback interface (in the default
1249 network model) if none was precised.
1252 * Bugfix: MSG_process_resume now works with the current running process.
1254 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1255 * Modification of MSG_task_get_compute_duration. Once a task has been
1256 processed, the value returned by this function is now equal to 0. [AL]
1257 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1258 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1259 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1260 of a m_task. For now, MSG_task_cancel only works with computation
1262 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1263 of the processor (in Mflop/s) regardless of the current load on the
1265 * API Change: use proper naming convention for MSG_getClock and
1266 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1268 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1269 This function changes the priority of a computation task. This priority
1270 doesn't affect the transfer rate. A priority of 2 will make a task
1271 receive two times more cpu power than the other ones. This function
1272 has been added to suit the needs of Nguyen The Loc and hasn't been that
1273 much tested yet. So if it fails, please report it and send me your code.
1275 * API Change: removed all functions and types that were marked "deprecated"
1276 since many months. Renamed MSG_global_init_args to MSG_global_init.
1278 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1280 SimGrid (2.95) unstable; urgency=low
1283 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1284 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1285 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1288 * Bugfix: Allow absolute paths for platform description files [MQ]
1289 * Bugfix: do free the variables after use. Leads to drastic performance
1291 * Implement max_duration (ie, timeouts) on resources [AL]
1294 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1296 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1297 the needed support (timeouts) [AL]
1300 * Implement measurement sockets. You can now get the bandwidth between two
1301 hosts thanks to AMOK (see below). [MQ]
1302 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1303 dynar over the network (yeah) [MQ]
1304 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1305 * simple/static token-ring example. [Alexandre Colucci and MQ]
1306 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1307 in place before [MQ]
1310 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1311 type parsing. "short" and "long" modifiers were also ignored; other
1312 modifier (such as reference level) are still ignored. [MQ]
1313 * Embeed the buffer size within the buffer itself on SG. [MQ]
1314 That way, send() are atomic and cannot get intermixed anymore (at least
1315 the ones which are less than 100k; bigger messages still have the issue)
1316 * Array size pushed by the field, not by the field type (or each
1317 and every long int will push stuff to the cbps) [MQ]
1318 * use select() to sleep since it allows to portably sleep less than one
1321 GRAS (minor cleanups)
1322 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1325 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1326 * type category 'ignored' killed as it was never used and were difficult
1328 * whether a type can cycle or not is now a flag, leaving room for more
1329 flags (as "ignored", if we feel the need one day ;)
1330 * Rename raw sockets to measurement sockets since "raw" has another
1331 meaning in networking community.
1334 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1335 and offers features not belonging to GRAS but that most applications
1336 need. One day, it may be a set of plugins loadable at runtime.
1337 * New module: bandwidth
1338 bandwidth measurement between arbitrary nodes running this module. [MQ]
1340 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1342 SimGrid (2.94) unstable; urgency=low
1344 The first beta release of SimGrid 3 !
1348 * Update the main page and the FAQ. Adding references to gforge.
1351 * Add a gras_os_getpid function.
1354 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1355 * Extend the logs so that they also print PID, hostname, date, ... if
1357 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1358 and kill the old version which were in testsuite/
1359 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1363 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1364 for users. Instead, each programming environment (GRAS, MSG,...) use it
1365 when needed to provide such a feature to users.
1366 Don't shortcut the mecanism or you will also shortcut the virtualization
1367 you need on the simulator.
1371 * Cleanups in configury with regard to compile optimization/warning flags.
1372 Also add -fno-loop-optimize to any powerpc since it's the optimization
1373 killing gcc (< 3.4.0).
1374 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1375 needed by MSG examples complications
1376 * Borrow configury beautifications from PHP
1379 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1380 warning (thanks loris for stumbling into this one).
1381 * Bugfix: stop loading private headers (gras_config.h) from the public
1385 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1386 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1387 RL would imply a DNS resolver, which is *hard* to do in a portable way
1388 (and therefore delayed).
1389 * Implement a real timer mecanism and use it in timing macros. This allows
1390 to avoid rounding errors and get a 0.000005 sec precision in timing
1391 macros. While I was at it, various cleanups:
1392 - allow to declare more than one timed section per file (fix a stupid bug)
1393 - move some private declaration to the right place
1394 - merge conditional execution and timing macros into emulation module
1395 - document the module
1396 - make sure the module cleanups its mess on gras_exit
1397 * Documentation improvements:
1398 - (new) how to compile applications using GRAS
1399 - (new) emulation support (timing macros)
1401 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1403 SimGrid (2.93) unstable; urgency=low
1405 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1408 - Use Paje properly where used. Still to be sanitized properly.
1409 - Portability fix: Add an implementation of the contexts using pthread
1413 - Add xbt_procname(): returns the name of the current process.
1414 Use it to show the current process's name in all logging.
1416 - fix detection of older flex version and the reaction, since we do
1417 depend on modern ones (we use lex_destroy)
1418 - Better separation of SG and RL in the libs: remove all simulation code
1419 from libgras. As a result, this lib is now only 200k when stripped.
1420 Some of the xbt modules may also be duplicated (two sets and such) and
1421 should be cleaned/killed before SG3.
1422 - Insist on using xlC on AIX because of weird problems involving gcc there.
1423 - Cleanup the make remote stuff. This is now done by scripts
1424 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1425 mainly for our private use, but we're working on changing them to user
1428 - Bugfix: flush the socket on close only if there is some *output*.
1429 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1431 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1434 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1436 SimGrid (2.92) unstable; urgency=low
1438 Alpha 3 on the path to SimGrid 3
1442 - New! First try of benchmarking macros.
1443 - New! First try so that gras_stub_generator generate deployment and
1444 remote compilation helpers.
1446 - Bugfix: Initialization fix in msg_test.
1450 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1452 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1454 - API Change: gras_os_sleep to take the amount of seconds as a double.
1455 Accepting an int was error prone since it was the only location where
1456 seconds were coded as such. It leaded to damn rounding errors.
1457 - Bugfix: Hard to belive that timers ever worked before this.
1459 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1461 SimGrid (2.91) unstable; urgency=low
1463 Alpha 2 on the path to SimGrid 3
1467 - Bug fix in the lmm_solver.
1469 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1470 through the function MSG_paje_output.
1471 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1472 - It is possible to bound the rate of a communication in MSG with
1473 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1474 to write it in the changelog).
1475 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1480 - Various cleanups to the autotools stuff
1481 - Begin to move Gras examples to examples/gras/
1482 - Let make distcheck work again (yeah!)
1484 - documentation overhauled using doxygen.
1485 gtk-doc-tools is dead in SimGrid now.
1486 - Automatically extract all existing logging categories, and add the list
1487 to the documentation (long standing one, to say the less)
1489 - Cleanup the known architecture table. Reorder the entries to group what
1490 should be, and use a more consistent naming scheme.
1491 (some of the test dataset are still to be regenerated)
1492 - New! Allow library to register globals on each process just as userdata
1494 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1495 lookup time (for now).
1496 Use it in msg and trp.
1497 This cleans a lot the internals and helps enforcing privacy of the
1498 headers between the gras components.
1499 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1500 - Bugfix: gras_os_time was delirious in RL.
1501 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1503 - Reenable GRAS now that it works.
1505 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1507 SimGrid (2.90) unstable; urgency=low
1509 Alpha 1 on the path to SimGrid 3
1511 * It is a long time since the last release of SimGrid. I'm sorry about
1512 that but as I had told you, I was rewriting a lot of things. I apologize
1513 to those who had been reporting bugs to me and that I had not answered.
1514 If your bug is still in the new version, please tell me. Here is a
1515 summary of the main changes.
1517 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1518 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1519 portability, speed, and a lot more but you'll figure it out later.
1520 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1521 I wanted to release the new MSG as soon as possible and I have
1522 broken GRAS, which is the reason why, for now, only MSG is fully
1523 functional. A laconic description of these projects is available
1524 in the documentation.
1526 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1527 kernel optimized for our needs (called SURF but only the developers
1528 should use it). Hence, MSG is now roughly 30 times faster and I think
1529 that by rewriting a little bit MSG, I could event speed it up a little
1530 bit more. Beside the gain in speed, it is also much easier to encode a
1531 new platform model with SURF than it was with SG. More to come...
1533 * REVOLUTION 3: I have tried to change a little as possible the API of
1534 MSG but a few things really had to disappear. The main differences
1535 with the previous version are :
1536 1) no more m_links_t and the corresponding functions. Platforms are
1537 directly read from a XML description and cannot be hard-coded
1538 anymore. The same format is used for application deployment
1539 description. The new format is described in the documentation.
1540 Have a look in tools/platform_generation. There is a tiny script
1541 that converts from the old platform format to the new one. Concerning
1542 the application deployment format, parsing the old one is tricky.
1543 I think most of you should however be able to convert your files. If
1544 it is really an issue, I can write a C code that does the conversion.
1546 2) the toolbox tbx does not exist anymore. We now have a library
1547 with much more data-structures but without the hash-tables (we have
1548 dictionaries that are much faster).
1550 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1552 *****************************************************************************
1553 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1554 * don't feel like converting the oldies. (Mt) *
1555 *****************************************************************************
1558 Version 2.90: "the long awaited one"
1559 - Finished rewriting and debugging MSG. Rewrote the documentation.
1560 - disable GRAS for now since it needs to be ported to the newest SG
1563 - Finish the port to windows (using mingw32 for cross-compile)
1566 - Main loop and datastructures of SURF. A cpu resource object is
1567 functional. Surf can thus be used to create cpu's with variable
1568 performance on which you can execute some actions.
1570 2004-11-15 Martin Quinson
1571 - Port to ARM. Simply added the alignment and size descriptions. Should
1572 work, but the ARM machines are so slow that I didn't had the opportunity
1573 to 'make check' over there yet.
1575 2004-11-15 Arnaud Legrand
1576 - Trace manager now written. It uses a heap structure and is therefore
1577 expected to be efficient. It may however be speeded up (particularly
1578 when many events occur at the same date) by using red and black
1579 trees. One day maybe...
1580 - Max-min linear system solver written. It uses a sparse matrix
1581 structure taking advantage of its expected use. Most operations are
1582 O(1) and free/calloc are called as few as possible. The computation of
1583 the minimum could however be improved by using a red and black tree
1586 2004-11-03 Arnaud Legrand
1587 - Rename every gras_* function that was in xbt/ to its xbt_
1589 - Add a heap and a doubly-linked list to xbt
1590 - Added a dichotomy to the dictionaries. make check works as well before
1591 so I assume that the patch is correct. I do not know however if things
1592 run effectively faster than before now. :)
1594 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1595 SimGrid, and the version number is bumped to 2.x
1597 2004-10-29 Martin Quinson
1598 - Introduction of the remote errors.
1599 They are the result of a RMI/RPC on the remote machine.
1600 ErrCodes being scalar values, you can't get the host on which those
1601 errors did happen. Extending the error mechanism as in Gnome is possible.
1602 No idea yet whether it is a good idea.
1604 2004-10-28 Martin Quinson
1605 - Interface revolution: the Starred Structure Eradication.
1606 I used to do typedef struct {} toto_t; and then handle *toto_t.
1607 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1608 - ???_t is a valid type (builded with typedef)
1609 - s_toto_t is a structure (access to fields with .)
1610 - s_toto is a structure needing 'struct' keyword to be used
1611 - e_toto_t is an enum
1612 - toto_t is an 'object' (struct*)
1614 typedef struct s_toto {} s_toto_t, *toto_t;
1615 typedef enum {} e_toto_t;
1616 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1617 s_toto_t) is private.
1619 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1620 so that it can set the variable to NULL. It was so for dicts and sets,
1621 it changed for dynars.
1623 - Fix a bunch of memleaks in dict_remove
1624 - Fix a bug in sg/server_socket opening: it failed all the time.
1626 2004-10-07 Martin Quinson
1627 - Speed up dynar lookup operation a bit.
1629 gras_dynar_get is dead.
1631 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1632 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1633 (which gives you the address of the stored data).
1635 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1636 copy of the data using an affectation to do the job and not a memcpy.
1638 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1639 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1641 It does not really speedup the dynar test because they are
1642 setting elements all the time (and look them seldom). But the dict does
1643 far more lookup than setting.
1645 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1647 2004-10-05 Martin Quinson
1648 - Allow to (en/dis)able the cycle detection at run time.
1650 Whether we should check for cycle or not is now a property of each
1651 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1652 datadesc_cycle_unset allow to remove this property when previously set.
1654 Note that the cycle detection is off by default since it impacts the
1655 performance. Watch the data you feed GRAS with ;)
1657 This property is hereditary. Any element embedded in a structure having it
1658 set have it set for the time of this data exchange.
1660 You should set it both on sender and receiver side. If you don't set it on
1661 sender side, it will enter an endless loop. If you forget on receiver
1662 side, the cycles won't be recreated after communication.
1664 - Header reorganization.
1665 Kill gras_private.h, each submodule must load the headers it needs.
1667 2004-10-04 Martin Quinson
1668 - Interface revolution: do not try to survive to malloc failure.
1670 Now, gras_malloc and friends call gras_abort() on failure.
1671 As a conclusion, malloc_error is not a valid error anymore, and all
1672 functions for which it was the only gras_error_t return value are
1673 changed. They now return void, or there result directly.
1674 This simplify the API a lot.
1676 2004-09-29 Martin Quinson
1677 - Re-enable raw sockets.
1678 Created by gras_socket_{client,server}_ext;
1679 Used with gras_raw_{send,recv}
1682 It should allow to kill the last bits of gras first version soon.
1684 This is not completely satisfactory yet (duplicate code with
1685 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1688 - Simplify transport plugin (internal) interface by not passing any
1689 argument to _server and _client, but embedding them in the socket
1692 2004-09-28 Martin Quinson
1693 - Finish the port to AIX.
1694 autoconf was my problem (segfault within the malloc replacement
1695 function. No idea why)
1697 2004-09-16 Martin Quinson
1698 - Fix some size_t madness on 64bit architectures.
1700 2004-09-08 Martin Quinson
1701 - Reduce the number of system headers loaded, overload some more system
1702 calls (such as malloc to cast the result of the system one, and work
1704 - Fix and reintroduce the config support
1706 2004-09-07 Martin Quinson
1707 - Source code reorganization to allow Arnaud to surf all over there.
1708 - Allow to document the logging categories.
1709 - Remove all uppercase from logging categories and useless cleanup in names.
1711 2004-08-18 Martin Quinson
1712 Version 0.6.2 (protocol not changed; API changed)
1713 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1714 gras_error_t), and NULL when not found. Functions expecting a msgtype
1715 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1716 hopefully usefull message.
1717 - Portability to prehistoric sparcs again
1719 2004-08-17 Martin Quinson
1720 Version 0.6.1 (protocol not changed; ABI not changed)
1721 - prealloc some buffers to speed things up
1723 2004-08-11 Martin Quinson
1724 Version 0.6 (protocol not changed; ABI expended)
1725 - The parsing macro can deal with the references, provided that you add
1726 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1728 2004-08-09 Martin Quinson
1729 Version 0.5 (protocol not changed; ABI changed)
1730 - Allow to off turn the cycle detection code in data exchange at
1731 compilation time. It should be at run time, but I'm short of time (and
1732 the config stuff is still broken). That way, we keep dict out of the
1733 critical path, which is good because the performance is poor:
1734 - search not dichotomial yet
1735 - dynar give no way to access their content and memcpy everytime
1736 - In composed data description (struct, ref and so on), stop foolness of
1737 keeping the subtype's ID, but store the type itself. This keeps sets out
1738 of the critical path, which is good since they rely on dynar and
1739 dictionnaries. The only loose of that is that we cannot detect the
1740 redeclaration of a structure/union with another content (but I'm not sure
1741 the code detected well this error before anyway). We still can detect
1742 the redefinition discrepancy for the other types.
1743 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1744 it breaks the code because of type-punning used all over the place).
1745 This breaks on all non-gcc architectures (for now).
1747 All those changes (plus the buffer of last time) allow me to gain 2 order
1748 of magnitude on cruel tests consisting of 800000 array of integers on two
1749 level of a hierarchical structure (200 secondes -> 4 secondes)
1752 - the selector of reference must now return the type it points to, not
1753 the ID of this type.
1755 2004-08-06 Martin Quinson
1756 Version 0.4 (protocol changed; ABI not changed)
1757 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1758 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1759 - Implement a buffer transport, and use it by default (it relies on tcp in
1760 real life and on sg in simulation).
1761 That's a bit hackish since I had a new field to the structure to store
1762 its data without interfering with the subtype ones. Inheritance
1763 is tricky in C. And that's a kind of reverse inheritance with one class
1764 derivating two classes. Or maybe a game with java interfaces. Anyway,
1765 that's damn hard in C (at least).
1766 Moreover, I got tired while trying to ensure plugin separation and
1767 genericity in SG mode. MSG wants me to do weird things, so let's go for
1768 cruel hacks (temporarily of course ;).
1769 See comment in transport_private.h:71
1770 - do not include all the _interface headers in private but in the files
1771 which really need them (to cut the compilation time when they are
1774 2004-07-26 Martin Quinson
1775 Version 0.3 (protocol not changed; ABI changed)
1776 - Major overhault of the datadesc interface to simplify it:
1777 - shorted the function names:
1778 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1779 - add a trivial way to push/pop integers into the cbps without malloc.
1780 This allows to make really generic sub_type description, which simply
1781 pop their size of the stack.
1782 - add a function gras_datadesc_ref_pop_arr() which does what users want
1783 most of the time: Declare a dynamic array (which pops its size of the
1784 stack) and declare a reference to it. Poor name, but anyway.
1785 - kill the post-send callback, add a post-receive one
1787 2004-07-23 Martin Quinson
1788 Version 0.2 (protocol changed; ABI changed)
1789 - add some testing for cpbs in the test cases, and fix some more bugs.
1790 This invalidate again the little64 data file, since I cannot regenerate
1792 - remove an awfull optimization in the logging stuff, allowing me to:
1793 - understand it again
1794 - learn gcc how to check that the argument match the provided format
1795 - fix all errors revealed by gcc after that
1796 - internal keys of dict are not \0 terminated. Deal with it properly in
1797 loggings instead of segfaulting when the user want to see the logs :-/
1799 2004-07-22 Martin Quinson
1800 - Fix some stupid bug preventing cbps (callback postit) from working
1802 2004-07-21 Martin Quinson
1803 - Some documentation cleanups
1804 - remove the useless last argument of msgtype_declare
1805 - rename the Virtu functions to fit into the 'os' namespace
1806 - move headers src/include -> src/include/gras/ and stop fooling with
1807 gras -> . symbolic link
1808 - make distcheck is now successful
1810 2004-07-19 Martin Quinson
1812 - Build shared library also
1813 - Install html doc to the right location
1814 - stop removing maintainer files in make clean
1815 - build tests only on make check
1817 2004-07-13 Martin Quinson
1819 - No major issue in previous version => change versionning schema
1820 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1821 corresponding dataset.
1823 2004-07-11 Martin Quinson
1825 - Get it working with any kind of structure (we can compute the padding
1826 bytes remotely for all the architectures I have access to)
1827 - Implement the structure parsing macro (still not quite robust/complete)
1828 - Improvement to the remote testing toysuite
1830 2004-07-10 Martin Quinson
1831 [autoconf mechanism]
1832 - get ride of a bunch of deprecated macros
1833 - actually run the test for two-compliment, not only compile it :-/
1834 - test whether the structures get packed (and bail out if yes. Damn.
1835 Alignment is a serious matter)
1836 - test whether the structures get compacted (but respecting the alignment
1837 constraints of each types)
1838 - test whether the array fields of structures can straddle alignment boundaries
1840 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1841 (took me 2 hours to find that bug, looking at the wrong place)
1842 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1843 before sending/receiving and are used to compute the offsets of fields
1844 - Given that padding size depend even on compiler options, keep track of
1845 alignment and aligned_size only for the current architecture. Not a big
1846 deal since we send structure fields one after the other (seems
1848 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1849 program, let it work on linux/gcc/little32. portability todo.
1851 2004-07-08 Martin Quinson
1852 - import and improve remote compilation support from FAST
1853 - make sure make check works on half a dozen of machines out there
1855 2004-07-07 Martin Quinson
1856 Let's say it's version 0.0.3 ;)
1857 - Implement conversions (yuhu!)
1858 - Let it work on solaris (beside conversion, of course)
1859 - Stupid me, using rand() to generate the conversion datatests in not wise.
1861 2004-07-06 Martin Quinson
1862 - Let make dist work, since I'm gonna need it to compile on remote hosts
1863 - Let Tests/datadesc_usage write the architecture on which the file was
1864 generated as first byte.
1865 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1866 architecture descriptions.
1867 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1868 on those architectures.
1869 - Optimization: send/recv array of scalar in one shoot
1871 2004-07-05 Martin Quinson
1872 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1874 - Plug a whole bunch of memleaks
1875 - each process now have to call gras_{init,exit}. One day, their log
1876 settings will be separated
1877 - Continue the code factorisation between SG, RL and common in Transport.
1879 2004-07-04 Martin Quinson
1881 - Redistribution between SG and RL.
1882 We wanna have to accept in SG, so move accepted related parts of RL in
1883 the common part. (more precisely, the dynar of all known sockets is no
1884 more a static in transport.c, but part of the process_data)
1886 [gras_stub_generator]
1887 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1889 2004-07-03 Martin Quinson
1890 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1891 them all up in one shot)
1893 - Fix the ugly case of reference to dynamic array.
1894 - New (semi-public) function gras_datadesc_size to allow the messaging
1895 layer to malloc the needed space for the buffer.
1897 - gras_socket_close now expect the socket to close (and not its address to
1898 put NULL in it after it). This is because the socket passed to handlers
1899 is one of their argument (=> not writable).
1901 - propagate the interface cleanup from last week in datadesc, ie remove a
1902 superfluous level of indirection. User pass adress of variable
1903 containing data (both when sending and receiving), and not of a variable
1904 being a pointer to the data. Let's say that I like it better ;)
1905 The price for that is constructs like "int msg=*(int*)payload" in
1906 handlers, but it's a fine price, IMHO.
1908 - Let it work in RL (yuhu)
1910 2004-06-21 Martin Quinson
1912 - porting SG plugin and SG select to new standards (works almost).
1913 - plug memleaks and fix bugs around.
1916 - cleanup the prototype of data recv and force users to specify when they
1917 want to handle references to objects. Test case working even for cycles.
1918 - plug memleaks. Valgrind is perfectly ok with this.
1920 2004-06-12 Martin Quinson
1922 - cleanup the separation between plugin and main code in plugin creation
1924 2004-06-11 Martin Quinson
1926 - Reput hook for raw sockets, needed for BW experiments
1927 - kill a few lines of dead code
1928 [Data description] Interface cleanup
1929 - gras_datadesc_by_name returns the searched type or NULL.
1930 That way, no variable is needed to use a type desc once, which makes
1932 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1933 two parameters were strings (field name, type name), leading to
1935 [Dicos] Interface cleanup
1936 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1937 This is consistant with the dynar API.
1939 2004-04-21 Martin Quinson
1941 - Porting to new standards.
1943 - interface cleanup.
1944 There is no bag anymore, no need to take extra provision to mask the
1945 pointers behind "ID".
1946 Better splitup of functions between files create/exchange/convert.
1947 This is still a bit artificial since convert and receive are so
1948 interleaved, but anyway.
1950 - add a queued message list to procdata (the ones not matching criteria
1952 - factorize some more code between SG and RL wrt procdata
1954 - use gras_exit in example to track memleaks
1955 - get rid of gs_example now that GS is properly integrated into gras
1956 - update run_test to integrate the lastest tests (datadesc)
1958 - rename WARNINGn macros to WARNn since it prooved error-prone
1960 2004-04-19 Martin Quinson
1962 - register init/exit functions within gras module mechanism
1963 - send/receive function.
1964 Convertion is not implemented, but short-cutted if not needed.
1965 struct/array elements are sent one by one (instead of block-wise), but
1966 nobody really cares (yet). Get a prototype before optimizing.
1967 - tests (using a file socket) for DD send/receive on:
1968 - base types: int, float
1969 - array: fixed size, string (ie ref to dynamic string)
1970 - structure: homogeneous, heterogeneous
1971 - chained list, graph with cycle
1972 Believe it or not, valgrind is not too unhappy with the results. The
1973 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1977 - Bugfix when using a filename explicitely (instead of '-')
1979 2004-04-09 Martin Quinson
1981 - factorize more code between RL and SG in socket creation
1982 - Complete the implementation and tests of:
1984 o file (only in RL, and mainly for debugging)
1986 I lost 3 days to design a portable address resolver, and then decided
1987 that the prototype mainly have to run on my box.
1988 Addressing portability too early may be like optimizing too early :-/
1990 - use gras_init in the Tests instead of the crappy parse_log_opt
1991 (the latter function is removed)
1992 [Conditional execution]
1993 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1994 [Code reorganisation]
1995 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1996 Build examples against the RL library, since there is no way to disable
1997 its creation for now.
1999 For information, the beginning of coding on GRAS was back in june
2000 2003. I guess that every line has been rewritten at least twice since