1 SimGrid (3.6) unstable; urgency=low
4 * completely rewrote this module to allow parallel execution.
7 * Add a new function xbt_dynar_to_array that transforms a dynar into a
8 NULL-terminated array. This may solve backward compatibility issues
9 due to the change to return type of SD_Simulate. See also:
10 http://lists.gforge.inria.fr/pipermail/simgrid-user/2010-December/002206.html
12 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
14 SimGrid (3.5) stable; urgency=medium
17 * New feature to any SimGrid-based simulator: Model-Checking
18 Check SIN#1 for more details.
21 * New Model SMPI (three-interval linear regression for correction factors)
22 See RR-7426, available at http://hal.inria.fr/inria-00527150
23 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
24 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
25 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
26 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
27 * See also src/smpi/README
32 - Tracing API changes: TRACE_start and TRACE_end should not be called
33 by user-code. They are automatically called by simulators created
34 with SimDAG, MSG and SMPI if the toolkit is compiled with
35 tracing_enabled=ON. Categories declaration and utilization remain the
36 same for MSG and SimDag.
37 - A function was added to the tracing API to declare categories with
39 - TRACE_category_with_color (char *category, char *color)
40 where color must be in the following format
41 "%f %f %f", red, green, blue
42 and red, green, blue are float values in the interval [0, 1]
43 - User can specify NULL as color parameter, or continue calling
45 On that case, the tracing system will define random colors
46 - The following command-line options are supported:
47 --cfg=tracing/filename:msg.trace
48 --cfg=tracing:1 (activate tracing, needed to use others)
49 --cfg=tracing/platform:1 (categorized resource use)
50 --cfg=tracing/uncategorized:1 (uncategorized resource use)
51 --cfg=tracing/msg/task:1 (task creation)
52 --cfg=tracing/msg/process:1 (process creation, migration)
53 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
54 --cfg=tracing/smpi:1 (SMPI interface tracing)
55 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
56 - examples of examples/msg/tracing updated
58 - DAXLoader and DOTLoader functions can generate tasks with categories
59 - A new function to attribute a category to SD tasks:
60 TRACE_sd_set_task_category (SD_task_t task, char *category)
61 * Tracing the MPI interface implemented by SMPI
62 - Collective operations are traced with states
63 - Point-to-Point operations are traced with states/links
64 - Tracing activated by a parameter "-trace filename" passed
65 to smpirun during execution (considering that simgrid
66 is compiled with tracing enabled)
67 - To run the simulation with gdb, the simulator
68 accepts --cfg=tracing/smpi:1 to trace SMPI
69 - tesh files to check if smpi tracing is ok
70 - See examples/smpi/NAS/DT-trace
71 * GTNetS tracing re-worked
72 - adaptation to the tracing system of GTNets to cope
73 with modifications regarding the fullduplex mode
74 - new tesh files to check if gtnets tracing is ok
77 * Asynchronous communications through the functions:
78 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
79 * New function: MSG_load_platform_script()
80 to make possible using a lua script instead of XML files to set up platforms
81 * New function: MSG_set_function
82 to associate functions to processes, used when bypassing the parser
83 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
85 Platforms: Add some more examples in examples/platforms
86 * Grid'5000: see www.grid5000.fr
87 * *_30000_hosts.xml: various huge files [mainly scalability testing]
90 * Change the XML format. This is a very important modification. SimGrid
91 3.5 introduces a new hierarchical format based on the notion of
92 Autonomous Systems. Compatibility with old format is ensured
93 through the perl script provided in the install bin directory
94 bin/simgrid_update_xml.
95 It is now possible to build platforms with specific routing
96 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
97 connect several platforms together. We will try to provide soon
98 set of realistic platforms exploiting these properties (have a
99 look at examples/platforms/ for the moment).
100 * Take the opportunity of the XML format change to be a good XML citizen:
101 rename link:ctn to link_ctn and similar changes (also dealed with by
103 * Add a new routing scheme (rule-based) using regular expressions. It
104 enables to have an extremely low memory footprint when the
105 underlying routing is simple and can be compactly described. You
106 need to have libpcre4-dev (perl regular expressions) installed if
107 you want to use this routing scheme.
108 * Revive the cluster TAG and allow to easily and efficiently (both in
109 term of memory and speed) connect clusters together. Have a look
110 at teshsuite/simdag/platforms/ to see how this can be done. With
111 this tag, you can create clusters with thousands of tasks at no
112 cost (have a look at examples/platforms/).
113 Note: clusters are implemented as ASes, so there is no need for an
114 enclosing AS tag if you have only one cluster in your platform.
115 * Add new generic functions in the public interface that allows the user
116 to call SURF 'create_resource' methods from your code (same
117 functionality as the XML bypass mechanism but with a much lighter
119 * Add a new model (enabled through command line --cfg=network/model:SMPI)
120 that uses a piecewise linear approximation to produce better
121 results when exchanging small messages.
122 * Add a new parameter to handle correctly full duplex link and account
123 for interferences between uplink and downlink communications
124 (activate with --cfg=fullduplex:1).
127 * Rename the SD_READY (all dependencies are satisfied and task is
128 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
129 dependencies are satisfied) state.
130 This prevents a confusion between the notion of "ready to schedule"
131 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
132 simulated" (SD_RUNNABLE) used by the simulation kernel.
133 * Change the way a task is considered as ready. Instead of removing
134 dependencies when a task is done, a counter is decreased. This way,
135 it is always possible to reach ancestors thanks to the
136 SD_taks_get_parents function (even after the end of the simulation.)
137 * Change the return type of SD_Simulate from (SD_task_t*) into
138 xbt_dynar_t. This function was in handling a dynar internally and
139 converted it into a NULL terminated array for historical reasons.
140 * New function SD_dotload(char*) to load a DAG described in dot
141 format. This loader and the corresponding examples require the
142 installation of the graphviz library.
143 * Fix a bug in the management of tasks of size 0 in the surf network
144 models. This problem was only visible with SIMDAG and you should
145 thus disregard results produced with earlier versions if you
146 relied on this feature (some tasks were blocked because of this).
147 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
148 even though your applications were doing only point-to-point
149 communications and sequential computations. Now you can really use any
150 model you want (of course, if you create real parallel tasks, which are
151 not implemented in most models beside ptaskL07, this will abort).
152 * Add an example that schedules a DAX on an heterogeneous platform
153 using a Min-Min strategy.
154 * New function SD_workstation_get_current_task() that returns the kind
155 of task currently running on a workstation in the sequential access
157 * Raise some warnings when unexecuted tasks remains at the end of the
158 simulation. This is usually caused by cycles in the DAG.
161 * New function: SIMIX_process_set_function() called by MSG_set_function
162 * Change the underlying waiting queue in semaphores so that a process
163 can wait on several of them simultaneously (as in waitany).
164 * Fix the way to handle tokens in semaphores so that all access patterns
165 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
166 * kill the dirty pimple SIMIX_message_sizes_output()
167 Please use (proper) visualization instead
170 * New data container: setset (set of sets of elements)
171 * New module: mmalloc (mapped malloc, allowing to have several
172 independent segments of malloc)
173 * New function: xbt_dict_cursor_set_data()
174 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
175 * New function: xbt_dynar_is_empty()
176 * New function: xbt_fifo_get_last_item()
177 * Fix xbt_dynar_shrink(): use the right element size.
178 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
179 uninitialized areas during expand.
180 * Fix semaphores: previous implementation was severely broken.
181 * Use library init/fini functions for our initialization.
182 - you can use logs and other feature as soon as you want in your
183 code (even before the xbt_init / MSG_init)
184 - xbt_exit is now a no-op and produce a warning when used.
187 * Port GRAS to new SIMIX mechanisms. This allows gras users to
188 benefit from the latest improvement to the simulation kernel.
189 * Kill measurement sockets for now. If you rely on them, sorry. This
190 release is not for you. This feature will be reintroduced in the
191 future, but we cannot delay the release any further.
192 * New function: gras_msgtype_get_name().
193 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
194 * Fix (at last) the pmm example: it should not randomly fail anymore.
196 Build chain: bug fixes and overall polishing
197 * Cmake is now stable enough. Hence, we killed the autotools.
198 * Port to windows ( TM :)
199 * Fix the 'make install' target.
200 No need to use 'make install-simgrid' anymore
201 * Introduce a 'make dist' target compiling a *source* archive
202 'make package' compiles a binary archive
203 * Compile java files only on need
204 * Add --cd and --setenv command line options to tesh
205 * Out of source builds are not fully supported yet, but we are close
206 * Enable supernovae and optimization flags by default for our users
209 * Add layer to set up environment directly from lua, without XML.
210 * The effect of gras_stub_generator can be achieved through
211 lua too (check examples/gras/console/ping_generator.lua)
213 -- Wed, 01 Dec 2010 22:09:23 +0100 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
215 SimGrid (3.4.1) stable; urgency=low
217 The "Polishing easter eggs is probably a good idea" release.
218 This is a bug fixes release only.
222 * Fix a bug preventing the tasks from begin garbage collected.
225 * Fix a bug occuring when a host involved in a communication fails.
226 This was not detected properly by the other peer involved in the
227 communication. Now, it's reported as a network error.
230 * Warn the user about loop dependencies in data flow of DAX files
231 * Obey the control-flow dependencies of DAX files
234 * Add option "enable_smpi" allowing to not compile SMPI.
235 Probably useful for the (Mac) users experiencing a build error here
236 * Improve the detection of lua5.1 and ruby1.8
238 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
240 SimGrid (3.4) stable; urgency=low
242 The "Easter in Cargese" release. Also known as (major changes):
244 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
245 ~> bindings were greatly improved
246 ~> new tracing infrastructure for better visualization introduced
248 * the "Welcome to configury modernity" release.
249 ~> we switched from autotools to cmake, and improved our cdash
252 A more detailled list of changes follow (full detail in svn log).
254 Java Bindings: Various Cleanups
255 * (install java-gcj-compat-dev on debian-like to use them)
256 * Remove put/get: no need to export deprecated interface in Java
257 Use send/receive instead.
258 * Cleanup the examples and add a README per directory
259 * Remove example autoDestination (that's the only way to go now)
260 * Remove example explicitDestination (was a plain copy of basic)
261 * Make JniException a runtime exception, so that there is no need to
262 declare the fact that you may encounter such a beast. I guess that
263 nobody will ever want to survive such error.
264 * Create specific errors for each MSG case of failure:
265 host failure, transfer failure, timeout, task cancelled
266 * Cleanup the exceptions that may get thrown by each function
267 * Other internal cleanups in Java bindings. Performance still bad :/
268 Ruby and Lua Bindings: create them
269 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
270 * That's new and great, you should try them out.
271 Same functionalities than Java bindings, only even less polished
273 * Kill the useless "rate" argument of SD_task_get_execution_time()
274 Everyone used to provide -1 as a value, it was not used, and the
275 semantic of a possible use wasn't even clear.
276 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
277 either as comm costs or compute costs to mean that there is no
278 such thing for that specific task.
279 * Add a SD_task_set_name() function
280 * Fix SD_task_unschedule() on typed tasks
281 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
282 * In DAX loader, accept useless 'level' attributes to <job> since
283 LIGO DAGs have them (seem to be to ease graphical representation).
285 * Add an example masterslave_mailbox.c using send/receive and not
286 the deprecated put/get interface.
287 * Kill the MSG_paje_output() function. It's a noop since 2 years.
288 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
290 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
291 (declare MSG_USE_DEPRECATED to still have the old name)
292 * Add a MSG_task_set_data() function
293 * About trace replay (see examples/msg/actions):
295 - Allow to work with splitted trace files for each process
296 Give the specific trace file as argument of each process,
297 and call MSG_action_trace_run(NULL)
298 You can still have one merged file for all processes.
299 - Fix implementation of collective operations
300 * Allow task_execute() on 0-sized tasks (closes #10063)
302 * This is the first release of SimGrid where SMPI is not considered
303 beta anymore (even if some corners should still be improved)
304 * Port over the new SIMIX_network submodule (internal refactoring)
305 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
306 * Implement more missing elements of the standard:
308 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
310 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
311 MPI_Scan() MPI_Get_processor_name()
312 - Added implementation of missing case for Alltoall (warning: it's
313 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
314 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
315 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
316 SMPI_MPI_Allgatherv()
318 - MPI_Waitsome() was broken
319 - Allow relative includes in smpicc
320 - Command line cfg argument 'reference_speed' was ignored...
321 - Some functions did not properly lead to auto-benching of user code
322 - smpicc passes -O2 by default (just like openmpi one)
324 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
325 * Bug fixes about timeouts during communications
326 * add SIMIX_message_sizes_output() as a pimple to write to file the
327 amount of messages per size. Use gnuplot to get histogram.
328 Pimple because that's the only user-visible function of simix,
329 defined directly in xbt.h (irk, sorry)
331 - Add a SIMIX_sem_get_capacity() function
332 - Fix interactions with processe resume/suspende
333 - release_forever() was stupidly broken
334 - Fix SIMIX_display_process_status() for processes in a semaphore
335 - Make SIMIX_sem_block_onto() user-visible
336 * Refactoring context stuff:
337 - Use pseudo-OOP for better modularity
338 - reimplement SIMIX_process_kill() without process_schedule() so
339 that the latter can take as invariant that it is called from
341 - Merge context_start into context_new for sake of simplicity
343 * Add a Vivaldi network model, coded live during SUD'10 ;)
344 * Rename configuration variables to start a hierarchy:
345 o cpu_model -> cpu/model
346 o network_model -> network/model
347 o workstation_model -> workstation/model
348 * New configuration variables:
349 o network/bandwidth_factor: correction to bandwith
350 o network/latency_factor: correction to latency
351 o netwotk/weight_S: correction to the weight of competing streams
352 * Add a long description to the models, that users can see with such
353 argument on the command line: --cfg=cpu/model:help
354 * --help-models display the long description of all known models
356 * config: add the ability to set a default value after registration
357 Does not override any previously set value (e.g. from cmd line)
358 * dict: allow to have integer key and data.
359 When so, you need to use the following functions
360 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
361 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
362 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
363 In contrary to regular dicts, the key is not malloced before copy.
364 Mixing scalar and regular elements in the same dict is not tested
366 * Allow to use xbt_dynar_shrink() to expend the dynar instead
367 Tracing for Visualization:
368 * SimGrid is now instrumented in order to generate a trace file for
369 visualization analysis: to use it, need to compile SimGrid with the
370 "tracing" option enabled, and instrument the program using SimGrid with
371 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
372 (among other functions).
373 * The instrumentation only traces the platform utilization for now
374 * Documentation to use the tracing functions and how to analyze the
375 traces with the Triva tool is written.
376 * More information about: SimGrid FAQ (in the section Tracing Simulations
379 * We moved to cmake as default build system. Autotools support will
380 be dropped soon. Check the FAQ for more info about how to use it.
381 * Greatly improved our cdash/ctest interactions
382 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
383 * Added memory checking tests with valgrind; lot of memleak fixing.
384 This may be the first release of simgrid with so few memory issues
385 * Added code coverage tests.
386 Our coverage is still improvable, but at least we see it on cdash.
388 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
390 SimGrid (3.3.4) stable; urgency=low
392 The "Desktop Grid needs love too" release (also called Xmas release).
395 * Major speedup in the maxmin system solving by using lazy evaluation
396 Instead of solving completely the maxmin system at each iteration,
397 only invalidate (and recompute) the modified parts.
398 This new feature is enabled in default models but you can try to
399 turn it on with "--cfg:maxmin-selective-update=1" for other models.
400 * Cas01 IMproved as default CPU model
401 This CPU model is the same Cas01 model, but it uses the
402 maxmin-selective-update flag and a heap structure to manage
403 actions on SURF kernel.
404 It reduces the complexity to find the next action to finish and,
405 consequently, it's faster than the old Cas01.
406 This is the new default CPU model (Cas01).
407 * Rename the old Cas01 model to Cas01_fullupdate
408 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
409 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
410 * CpuTI (CPU Trace Integration)
411 A new CPU model whose objective is simulate faster when using
412 availability trace files.
413 Instead of using a full featured, over engineered maxmin system for
414 CPU modeling, this model does the pre-integration of traces files
415 to calculate the amount of CPU power available, and so, executes
416 faster than the old CPU models.
417 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
418 * Use LV08 as default network model since it gives better accuracy
419 for small messages and shouldn't change things for big ones.
420 Use --cfg=network_model:CM02 to get the previous behavior.
423 ******************************************
424 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
425 ******************************************
426 * The new CPU model may changes simulations!
427 The point is that events occurring at the exact same timestamp
428 are not scheduled in the same order with the old and new
429 version. This may be enough to completely change the execution
430 of simulations in some cases.
431 * The new network model will change simulations!
432 This new model is more realistic than the previous one, so you
433 should consider redoing your old experiments with this model.
434 Sorry for the inconvenience.
437 * Introduce the supernovae compilation mode
438 When compiled that way, the whole SimGrid (or almost) is put in a
439 single compilation unit and compiled in one shoot.
440 This is to help gcc which has difficulties to inline stuff from one
442 The speedup seem to be above 15%, althrough more tests are needed on
443 amd64 to confirm that gain.
446 * Port of MSG's mailbox on top of SIMIX network
447 The put/get mechanism was greatly simplified on the way.
450 * New SIMIX network module. Provides:
451 - Mailbox: rendez-vous mecanism to find with who you want to speak
452 - Synchronous send/recv: easier and hopefully faster since the
453 logic is handled in the maestro process directly now
454 - Asynchronous send/recv: you dreamt of it? It's here now
455 Too bad that nobody cared enough to propagate the change to MSG.
456 * Add semaphores as SIMIX synchronization mechanism.
459 * new function SD_daxload(char*) to load a DAX file
460 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
461 * Introduce typed tasks. Specify its kind and cost at creation.
462 At scheduling, just give where it should be placed, and the cost
463 for each involved resource is automatically computed.
464 Existing constructors so far (more to come of course):
465 - SD_task_create_comm_e2e() for end-to-end communication
466 - SD_task_create_comp_seq() for sequential computation
467 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
468 * new function SD_task_dump() for debuging display
469 * new function SD_task_dotty(task,FILE*) writing to file the info
470 about the task in dotty format
471 * SD_task_dependency_exists() can now cope with having one of its
472 arguments NULL. If so, it tests whether the other argument has any
474 * Add getters on list of preceding/following tasks:
475 SD_task_get_parents(task) and SD_task_get_children(task)
476 * Add getters on amount of workstations and list:
477 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
478 * Add getter on task kind: SD_task_get_kind(task)
479 * Update the start_time and finish_time of tasks on completion/failure
480 * Bugfix: Remove task from state swags when destroyed
483 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
486 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
487 dynar with one separate thread per value of the dynar.
488 * Change the prototype of xbt_thread_create(), sorry.
489 Added a boolean parameter indicating whether we want to join this
490 thread (used in SG only for now)
491 * Implement xbt_thread_join and xbt_thread_yield in SG also.
494 * GTNetS wrappers should now be usable again (and betterly tested too)
495 * Fix a major regression from 3.2 where the timeout provided to
496 MSG_task_put_with_timeout() was used as absolute time before which
497 the comm should be done.
498 * Start to fix the <cluster> tag.
499 - Internal links should be good now (beside of the loopback, which
500 use the private link instead)
501 - paths to the external world is still rather broken
502 - the <route:multi> tag is just broken. Actually that's brain-dead.
503 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
504 to make it less stupid
505 ** Check your platform with teshsuite/simdag/platforms/flatifier **
506 * Fix a source-level compatibility glitch from 3.2: after defining
507 MSG_USE_DEPRECATED, you can use the old name
508 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
509 * Allow to compile from the SVN with automake 1.11
510 * Fix some problems when using the "start_time" tag in deployment XMLs.
511 * Fix #8569: XBT/synchro.h has redundant declarations
512 * Fix #8563: MSG return values and exceptions
513 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
514 * Integrate patch #8636: Obey DESTDIR when installing documentation.
515 Thanks to Robson Peixoto.
516 * Fix a vicious bug in dictionaries inducing that some elements were
517 not freed on xbt_dict_free()
519 Portability report of this version:
520 * Main portability targets:
521 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
523 - mac leopard on i386
524 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
525 but nothing critical.
526 * Other platforms: windows, AIX and others were not tested for this release
528 Timing report of this version:
529 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
530 scenario parameters). From 8h to a few seconds in desktop grid settings.
531 * Supernovae brings about 25% speedup on i386.
533 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
535 SimGrid (3.3.3) stable; urgency=low
537 The "Need for Speed" release.
539 The timings done to validate the 3.3.2 were faulty.
540 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
542 The problem was a conversion from a manually handled vector to
543 xbt_dynar_t on the critical path.
544 xbt_dynar_foreach calls functions, inducing stack management crap.
546 We inlined these functions and xbt_dynar_foreach is now breath taking.
547 We also inlined xbt_swag_belong on the way.
549 Here are some approximate speedup measurements (on master/slaves
550 simulations lasting between 10s and 20s each):
551 3.3.1 -> 3.3.2: about same performance
552 3.3.2 -> 3.3.3: 40% speedup
553 3.3.1 -> 3.3.3: 40% speedup
554 3.3.1 with inline patch -> 3.3.3: 30% speedup
556 Our reading is that the refactoring which occurred in 3.3.2 made us
557 suffer much more from the xbt_dynar_foreach low performance, but
558 once we solved this, this refactoring proved to be very performance
559 effective. From the 40% speedup, somehow, 10% are due to the
560 inlining and 30% to the refactoring.
562 That's a pitty that gcc cannot inline functions placed in other files
563 alone. We have to choose between:
564 - break the encapsulation (by putting private data structures and
565 accessors in headers files to help gcc)
566 - live with low performance
567 - switch to a decent compiler such as icc (not quite possible).
569 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
571 SimGrid (3.3.2) stable; urgency=low
573 The "Simplicity does not preceed complexity, but follows it" release.
575 The main contributors of this release were (lexical order):
576 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
579 * Extract the routing logic into its own object.
580 (was dupplicated in network.c and workstation_LV07.c;
581 Allows to implement other ways of storing that info)
582 => kill now useless network_card concept
583 - Use dynar to represent routes (instead of void** + int*)
584 - kill link_set (use surf_network_model->resource_set instead)
585 - Add a command-line option to choose the routing schema to use
586 - Add three new models:
587 * Floyd (shortest path computed at initialization)
588 * Dijikstra (shortest path recomputed all the time)
589 * Cached Dijikstra (shortest path computed on need)
590 All these models where contributed by Silas De Munck, and are
591 described in his ICCS09 paper.
593 * Simplify model declaration
594 (less redirections, less function to write when defining a model)
595 - Factorize stuff between models:
598 surf_model_resource_set(model)
599 surf_model_resource_by_name(model, name)
600 - Unify the types of models in s_surf_model_t (using an union)
601 - Embeed fields of common_public directly into s_surf_model_t
602 - Rename model methods:
603 action_free ~> action_unref
604 action_change_state ~> action_state_set
605 action_get_state ~> action_state_get
606 - Change model methods into functions :
607 (model)->common_public->action_use ~> surf_action_ref
609 * Implement a generic resource; use it as ancestor to specific ones
610 (allows to kill duplicated code in models)
611 Drawback: timer command don't need no name nor properties;
612 workstation_CLM03 don't need no properties
613 (but I guess we can live with those few bytes wasted)
615 * Improve the action object model
616 - implement a constructor avoiding dupplicated code about field
617 initialization in generic_action part.
619 * Kill the SDP model: it has an external dependency, is deprecated
620 in flavor of modern lmm models, and didn't compile since a while
623 * Relocation of the context module from XBT to SIMIX.
624 (the context were decoupled from the simix processes, duplicating a lot of code)
625 => a lot of code was factorized
626 - less overhead is introduced during scheduling
627 - simpler API for the context factory
628 - the logic for process creation,destruction and manipulation was simplified
629 * Simplification of the s_smx_process_t data structure.
630 => accesing the simix level data associated to a process is faster now,
631 and the code is a lot more readable.
634 * Implement some more MPI primitives:
635 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
636 -implementation: Bcast: flat or 2-ary tree (default),
639 Allreduce: Reduce then Bcast
640 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
641 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
642 Alltoallv: flat tree, like ompi
644 * Add support for optimized collectives (Bcast is now binomial by default)
645 * Port smpirun and smpicc to OS X
648 * Kill SD_link_get_properties: hard to maintain and makes very little sense
649 Shout out if you used it.
652 * Display the list of still queued messages in SG mode when existing
656 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
657 * Add xbt_graph_node_get_outedges() [Silas De Munck]
658 * Add xbt_str_from_file(FILE*)
659 * Add xbt_dict_get_key achieving a linear reverse search
660 * Remove the context module
662 Portability report of this version:
663 * Main portability targets:
664 - Linux(debian)/x86/context
665 - Linux(debian)/x86/pthreads
666 - Linux(debian)/amd64/context
667 - Linux(debian)/amd64/pthreads
668 On these, we still have the eratic breakages of gras/pmm and
669 amok/saturate_sg reported in previous version. We still think
670 that the tests are the cause of the fault, not the tested code.
672 - Mac OSX Leopard/x86/context
673 Still false negative in tesh autotesting.
674 Smpi still fails, but this time because readlink does not accept -f
675 Everything seems to work properly beside of that.
678 - AIX version 5.3 (only tested contexts this time)
679 Smpi still fails there because mktemp is not installed.
680 Everything seems to work properly beside of that.
682 I managed to compile it for the first time, but several breakages.
683 Won't delay the release for this exotic platform.
685 * Windows: it's still lagging behind. If you want to help, please
688 Timing report of this version:
689 This version seem to be more than 5% faster than 3.3.1 (on linux
690 64bits with contextes). The gain is less than expected, we are
691 investigating this for next release.
693 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
695 SimGrid (3.3.1) stable; urgency=low
698 * Implement a --cfg-help to show existing configuration variables
699 * Build chain do not require doxygen in maintainer mode
702 * fix a bug on struct sizeof computation, which prevented the
703 exchange of arrays of structs in some conditions
704 - added a regression test about this in datadesc_usage
705 * Allow the exchange of 0-long dynamic vectors.
706 - for that, use -1 as indicator of dynamic size instead of 0
707 - This implied to change any size from unsigned long to long,
708 reducing a bit communication abilities, but I guess that with
709 64bits being quite common, this is more than enough.
710 - This also induce a protocol change, thus bumping network protocol
711 version from 0 to 1 (if we have external users, we have to get
712 clean on that point too ;)
713 - added two regression tests about this in datadesc_usage
714 * Be more verbose when propagating local exceptions
715 This helps debugging.
716 * Display the status of simulated processes when receiving SIGINT in
720 * Allow to control the simulation from a trace file.
721 New functions MSG_action_register() and MSG_action_trace_run()
722 The first one allows to associate a function execution to each
723 kind of action while the second one parses a trace file and
724 triggers the corresponding actions within the system.
725 For now, only a toy example is provided in examples/msg/actions
726 * Add an exemple of process migration in examples/msg/migration
727 * Fix a bug in task exchange which broke MSG_task_get_sender()
728 Add a teshsuite regression test for that.
729 [Bug: if MSG_task_get_sender() is called after sender exit,
731 * Fix a bug which prevented suspend/resume to work properly
732 * Display the status of simulated processes when receiving SIGINT
733 This fixes a regression of v3.3. due to the introduction of SIMIX
734 * Bug fixing in failure management:
735 - trace could not start by a failure at time 0
736 - failure during communications were not working
739 * Add SIMIX_process_set_name() to change the name of the current
740 process in the log messages.
741 * Store smx_hosts in a dict since we only retrieve them by name
742 * Move the configuration infrastructure to surf
745 * Move the configuration infrastructure to surf
748 * Massive internal cleanups:
749 - Store internal structures on processes instead of hosts (allows
750 to have more than one process per host, in addition of being more
752 - Cleanup the initialization/finalization process
753 - Kill a whole bunch of unneeded synchronization:
754 processes run in exclusive manner within the simulator
755 - Move queues from global tables to process data fields
757 - now accept -platform and -hostfile arguments
758 - Pass the right rank value to processes according to the hostfile
759 * Compile the examples by default, and use them as regression tests
760 * Implement MPI_Wtime()
761 * Change the reference speed to a command line option
764 * TCP_gamma can now be specified as command line option using
765 --cfg=TCP_gamma:10000000.0
766 * Change the --surf-path cmd line option into --cfg=path:
769 * Also include strbuff from xbt.h public header
770 * xbt_ex_display(): do not free the exception after displaying
771 This allows to do more with the given exception afterward.
772 Users should call xbt_ex_free() themselves.
776 Portability report of this version:
777 * Main portability targets:
778 - Linux(debian)/x86/context
779 - Linux(debian)/x86/pthreads
780 - Linux(debian)/amd64/context
781 - Linux(debian)/amd64/pthreads
782 These targets fail about 1/10 of times on gras/pmm, but we believe
783 that this is because of the test, not because of simgrid.
784 amok/saturate_sg fails even more rarely, and the test may not be
787 - Mac OSX Leopard/x86/context
788 The test suite still spits tons of errors because some obscure
789 force prevents us from removing the temporary directories
790 arguing that they still contain some metadata I've never heard of.
791 Smpi fails because seq is not installed.
792 Everything seems to work properly beside of that.
795 - AIX version 5.3 (both contexts and pthread)
796 Smpi still fails there because mktemp is not installed.
797 XML inclusions seems rosty on AIX.
799 * Windows: it's still lagging behind. If you want to help, please
802 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
804 SimGrid (3.3) stable; urgency=high
808 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
811 * Introduce the SIMIX module: factorize code between MSG and GRAS.
814 Until now, GRAS were using MSG as an interface to SURF. It was
815 quite difficult because both interface have several differences
816 (MSG channels vs GRAS sockets were the most notable point).
818 This also opens the gate to SMPI (which should occur soon) and speed
819 up simulations by to 40% (even if it were not the main goal).
821 **************************************
822 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
823 **************************************
824 The point is that events occuring at the exact same timestamp are
825 not scheduled in the same order with the old and new version. This
826 may be enough to completely change the execution of simulations in
827 some cases. Sorry for the inconvenience.
829 * Cleanup and upgrade the XML format to push further scalability
830 issues (check http://hal.inria.fr/inria-00256883/ for more info)
832 * Improve the testing infrastructure with tesh. Now a very large part of
833 the code is tested not only by being run but also by checking that the
834 output match an expected output [Mt].
836 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
837 is really less memory-demanding, which should allow you to use
838 larger files in SimGrid [AL].
840 * Inform valgrind about our contextes, so that it becomes usable
841 with the default (and more effecient) version of SimGrid
842 [contributed by Sékou Diakite, many thanks]
845 * Introduce a listener thread in charge of receiving incomming
846 messages from the network. It allows to overlap communication and
847 computation but most notably, it removes some stupid deadlocks due
848 to the fact that so far, a process could not send and receive at
849 the same time. This made most non trivial communication schema
851 * Convert the PIDs from long int to int to match the MSG ones (and
853 * New function: gras_agent_spawn() to launch a new process on
854 current host. Only working in simulation for now. [Mt]
855 * New function: gras_os_hostport() returning a constant form (ie,
856 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
859 * Make the backtrace of exceptions more human readable [Mt]
860 * New module: xbt/str [Mt]
861 a ton of string utility functions (split, join, printf to a newly
862 allocated buffer, trim, etc)
863 * New module: xbt/hash [Mt]
864 SHA1 hashing algorithm (more to come if needed)
865 * New module: xbt/synchro [Mt]
866 synchronization tools (mutex and conditions) working the same way
867 in simulation and in real life (mainly useful for GRAS, but not
869 * New module: xbt/queue [Mt]
870 classical producer/consumer synchronization scheme
871 * xbt_dynar_new_sync() creates a synchronized dynar. All access
872 (using the classical functions will get serialized) [Mt]
873 * Make dictionary internal table dynamic. No need to specify its size
874 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
876 * Make sure the log channels are organized as a tree under windows
877 (because of ANSI C compatibility issue, any channel were child of
881 * Cleaned many thing in surf and fixed a few bugs [AL].
882 * Add a nice command line configuration mechanism to compose models [AL].
883 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
884 the previous one (KCCFLN05). It relies on something that looks like
885 a max-min sharing mechanism but cannot be written as such. A new solver
886 was thus designed [AL].
887 * Add a new solver to lmm. Based on Lagrange optimization and
888 gradient-based descent, it enables to efficiently maximise systems s.a
890 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
892 This solver enables to propose two new network models for TCP Reno and
893 TCP Vegas based on Low's work. These models still need to be fully
894 tested though [Pedro Velho].
897 * Bug fix in SD_simulate. Now the time bound given as argument is
899 * Use the new parallel task model (ptask_L07) as default.
900 * Use the SURF command line configuration mechanism.
901 * 0-size tasks (for synchronization) should now work.
903 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
905 SimGrid (3.2) stable; urgency=high
909 We still experience issues on this platform, but we believe that at
912 GRAS API BREAKAGE (for simplification purpose, sorry):
913 * the gras_msgtype_by_name is not used anymore. Instead of
914 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
915 you can write (and must)
916 gras_msg_send(toserver, "request", &request);
917 - If you still want to pass a gras_msgtype_t to the function (to cache
918 the type and avoid the lookup time), use the gras_msg_send_() variant.
919 - Impacted functions:
920 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
921 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
922 * The callbacks are now expected to return 0 when everything went well
923 (just like the main() function)
925 GRAS new features and improvements:
926 * New module mecanism where user code can use per process globals [Mt]
927 This is similar to gras_userdata_*() functions, but for libraries. It
928 factorize some code developped over and over in the examples and AMOK.
929 It has still to be documented and used (only amok/peermanagement is
931 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
932 loss when they were small enough to fit into the buffer and sent quickly
933 enough so that they can all get received in one shoot.
934 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
935 if not found. Use the *_or_null() variant for the old semantic.
936 * In gras_msg_handle, do not discard messages without callback.
937 They are probably messages to be explicitly awaited later (ie, proofs of
938 mis-synchronization in userland since they are sent before being awaited)
940 * gras_socket_meas_send/recv: semantic changed!
941 The numerical arguments used to be (1) the total amount of data to send
942 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
943 messages. This was need for the fool willing to send more than MAXINT
944 bytes on quite fat pipes.
947 * Do really rename the hostmanagement module to peermanagement. [Mt]
948 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
949 but this is rather new and this was documented in the module
950 documentation (poor excuses, I admit)
951 * Bandwidth measurement semantic changed! This follows the changes to
952 gras_socket_meas_send/recv explained above.
955 * A sequential mode has been added to the workstations. When a workstation
956 is in sequential mode, it can execute only one task, and the other tasks
957 are waiting in a FIFO. [Christophe Thiery]
960 * The KCCFLN05 workstation model now handles parallel tasks. It is the
961 model for SIMDAG. [Christophe Thiery]
962 * Bug fix in the maxmin solver: Some values were close to 0 instead of
963 equal to 0, which caused some bad behaviors in
964 saturated_constraint_set_update. I now use a threshold mechanism like in
968 * When running manually src/testall, you select specific units [Mt]
969 testall is the result of our cunit mecanism, and should replace all
970 the scripty thingy around since bash don't run easily on billware.
972 * A mallocator system has been added. [Christophe Thiery]
973 Mallocators allow you to recycle your unused objects instead of freeing them
974 and allocating new ones.
976 Documentation update:
977 * FAQ reworking + New FAQs:
978 - "Valgrind spits tons of errors!" [Mt]
979 - "How to repport bugs" [Mt]
980 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
981 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
982 - Communication time measurement within MSG [AL]
983 - I experience weird communication times when I change the latency [AL]
986 - an introduction to the framework and to the used communication model
987 - an initiatic tour introducing the most proheminent features:
989 . Lesson 0: Installing GRAS
990 . Lesson 1: Setting up your own project
991 o Part 2: Message passing
992 . Lesson 2: Exchanging simple messages
993 . Lesson 3: Passing arguments to the processes (in SG)
994 . Lesson 4: Attaching callbacks to messages
995 . Lesson 5: Using globals in processes
996 . Lesson 6: Logging informations properly
997 . Lesson 7: Using internal timers
998 . Lesson 8: Handling errors through exceptions
999 . Lesson 9: Exchanging simple data
1000 . Lesson 10: Remote Procedure Calling (RPC)
1001 . Lesson 11: Explicitely waiting for messages
1002 . Recapping of message passing features in GRAS
1003 - A HOWTO section containing:
1004 o HOWTO design a GRAS application
1005 More are due, of course. They will come latter. In the meanwhile, you can
1006 check the examples which are still here.
1008 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
1010 SimGrid (3.1) stable; urgency=high
1014 There was a stack corruption somewhere, visible only when optimizing
1015 with these versions. [Vince]
1018 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1019 module that enabled to play with Directed Acyclic Graphs. It is built
1020 directly on top of SURF and provides an API rather close to the old
1021 SG. Some old codes using SG are currently under rewrite to check that
1022 all needful functions are provided. [Christophe Thiery]
1025 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1026 extension of the classical CLM03 model that gracefully handles
1027 failures. This is now the default model for MSG and GRAS. It doesn't
1028 handle parallel tasks yet though. [AL]
1029 * Bug fix: Weights were not correctly set in the network part.
1030 WARNING: This may have resulted in incorrect results with simulations
1031 where there are more than one flow on a given link. [AL]
1034 * After a (long ?) discussion on simgrid-devel, we have decided that the
1035 convention we had on units was stupid. That is why it has been decided
1036 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1037 WARNING : This means that all previous platform files will not work as
1038 such with this version! A warning is issued to ask users to update
1040 A conversion script can be found in the contrib module of the CVS, under
1041 the name contrib/platform_generation/surfxml_update.pl [MQ]
1044 * Bug fix: Processes were started in reverse order, wrt deployment file.
1045 WARNING: if your code relies on this bug, please fix it. [AL]
1046 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1047 being executed on two different locations. [AL]
1048 * Bug fix: Failures are now better supported thanks to Derrick's tests
1049 (there was many failure situations I hadn't thought of and that weren't
1050 correctly handled). [AL]
1051 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1055 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1056 Exception occurring on server-side are propagated back to client (!).
1058 API CHANGE: the callback changed their prototype. Change:
1059 int my_handler(gras_socket_t expeditor, void *payload_data) {
1061 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1062 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1064 * New! function: gras_msg_handleall to deal with all messages arriving
1065 within a given period.
1066 * New! function: gras_socket_server_range to get a server socket in a
1067 range of port numbers (ease to avoid port number conflicts) [MQ]
1068 * New! gras processes display their backtrace when they get a SIGUSR1
1069 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1070 Sweet to debug RL processes [MQ]
1074 - Do not force experiment sizes to be expressed in kb, or it becomes
1075 impossible to measure the latency this way (needs one byte-long tests)
1076 WARNING: this changes the amok_bw_* function semantic. [MQ]
1077 - Implements the link saturation stuff. [MQ]
1078 * Peer management module:
1079 New! module factorizing code that we wrote over and over [MQ].
1082 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1083 - Test units are placed directly into the library code, they get extracted
1084 automatically and placed into the src/testall binary.
1085 - Convert most of the XBT tests to this system.
1086 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1087 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1088 live in the same C code. This required to change a little bit the API
1089 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1090 * New module: sparse graph structure with basic algorithms (this is work
1091 in progress and the API is not considered to be frozen yet). [AL]
1092 * Display more information on backtraces: source line & function names are
1093 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1094 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1095 * Reworked a little bit some #include statements to load only required
1096 headers. Some user code that relied on SimGrid to include stdlib or
1097 stdio may need to include it by themselves. [AL]
1098 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1100 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1101 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1102 but I guess that almost nobody used those parts. [MQ]
1104 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1106 SimGrid (3.0.1) stable; urgency=low
1109 * Unfortunately, I had missed 5 misnamed functions:
1110 xbt_fifo_item_t xbt_fifo_newitem(void);
1111 void xbt_fifo_freeitem(xbt_fifo_item_t);
1112 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1113 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1114 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1115 They're now deprecated. Please use their new versions:
1116 xbt_fifo_item_t xbt_fifo_new_item(void);
1117 void xbt_fifo_free_item(xbt_fifo_item_t);
1118 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1119 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1120 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1122 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1123 * Documentation: xbt_log module unmercifully reworked [MQ]
1124 * Bugfix: there was a problem with the ending of contexts with
1125 the pthread backend. It caused some weird deadlock or behavior
1126 depending on the pthread implementation. [AL]
1127 * Bugfix: get the exceptions raised in the simulator repport where
1128 and why they come from when they are not catched in time [AL, MQ]
1131 * Bugfix: Do repport the error when two non-connected hosts try to
1132 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1135 * Add additionnal checkings on communications. Assert that two
1136 communicating hosts are connected by a set of links... [AL]
1139 * Add additionnal checkings on channel values in communication [AL]
1140 * New: MSG_task_get_source to see on which host a task was generated [HC]
1141 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1142 the number of tasks waiting to be received on channel and sent
1144 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1145 waits for the first task coming from a given host.. [AL]
1147 GRAS new functionnalities: [MQ]
1148 * Enhance the parsing macro to allow the size of multidimentional objects
1149 to be given thru annotations.
1150 * New example (and documentation): Matrix Multiplication a la RPC
1151 (as when I was young!) and fix a bunch of bugs found on the way.
1153 GRAS performance improvements: [MQ]
1155 * Reduce the amount of cbps creation/destruction by making it static to
1156 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1159 * Change libdata to a set so that we can search for stuff by ID (and thus
1160 reduce the insane amount of dict lookups)
1163 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1164 RL communication on top of SG since it's so uneffective.
1165 It may also be used for inter-thread communication in RL, one day.
1166 * Use gras_datadesc_copy() to exchange messages on top of SG
1168 - improve message exchange performance on top of SG
1169 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1170 * Don't exchange on the network the size of the used part of buffer,
1171 instead, specify the possible buffer size to read().
1173 - reduces the amount of read/write calls (one pair per exchange)
1174 - reduces the amount of exchanged data (the size)
1175 - allows to retrieve all arrived data on receiver side, if we don't need
1176 it right now (subsequent read will peek the buffer)
1177 - allows the receiver to proceed with the begining of the stream before
1178 everything is arrived
1179 - make it possible to build an iov transport (using readv/writev)
1181 - take care of the data with non-stable storage (like stacked data),
1183 * If possible, TCP send uses vector I/O (when writev() is here)
1184 - Don't use it for receive since we send data sizes and data on the
1185 same stream, so we wouldn't be able to chain large amount of chunks
1186 before having to flush the stuff to read the size.
1187 * Rework the transport plugin mecanism to simplify it and reduce the
1188 amount of pointer dereferencement when searching for the right function
1191 * I guess that now, we do almost as few system calls as possible while
1192 doing as few data copy as possible.
1194 To improve it further, we could try to send all the sizes first and then
1195 all the data (to use iov on receiving size), but it's only a partial
1196 solution: when you have 2 dimensional data, the sizes of the second
1197 dimension is data of the first dimension, so you need 3 streams.
1199 I'm not sure the potential performance gains justify the coding burden.
1201 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1203 SimGrid (3.00) stable; urgency=high
1206 * New! Give the possibility to hijack the surf parser and thus bypass
1207 MSG_create_environment and MSG_launch_application. Have a look at
1208 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1210 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1212 SimGrid (2.96) unstable; urgency=low
1217 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1218 This deprecates the xbt_error_t mecanisms.
1219 It modifies (simplifies) all XBT and GRAS API.
1220 MSG API keeps unchanged (exceptions raised by XBT are catched from
1221 within MSG and masked with existing error handling facilities)
1224 * New! Add a FATPIPE model. [AL]
1225 * New! Add a parallel task model. [AL]
1226 * New! Add automatically a loopback interface (in the default
1227 network model) if none was precised.
1230 * Bugfix: MSG_process_resume now works with the current running process.
1232 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1233 * Modification of MSG_task_get_compute_duration. Once a task has been
1234 processed, the value returned by this function is now equal to 0. [AL]
1235 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1236 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1237 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1238 of a m_task. For now, MSG_task_cancel only works with computation
1240 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1241 of the processor (in Mflop/s) regardless of the current load on the
1243 * API Change: use proper naming convention for MSG_getClock and
1244 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1246 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1247 This function changes the priority of a computation task. This priority
1248 doesn't affect the transfer rate. A priority of 2 will make a task
1249 receive two times more cpu power than the other ones. This function
1250 has been added to suit the needs of Nguyen The Loc and hasn't been that
1251 much tested yet. So if it fails, please report it and send me your code.
1253 * API Change: removed all functions and types that were marked "deprecated"
1254 since many months. Renamed MSG_global_init_args to MSG_global_init.
1256 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1258 SimGrid (2.95) unstable; urgency=low
1261 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1262 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1263 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1266 * Bugfix: Allow absolute paths for platform description files [MQ]
1267 * Bugfix: do free the variables after use. Leads to drastic performance
1269 * Implement max_duration (ie, timeouts) on resources [AL]
1272 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1274 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1275 the needed support (timeouts) [AL]
1278 * Implement measurement sockets. You can now get the bandwidth between two
1279 hosts thanks to AMOK (see below). [MQ]
1280 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1281 dynar over the network (yeah) [MQ]
1282 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1283 * simple/static token-ring example. [Alexandre Colucci and MQ]
1284 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1285 in place before [MQ]
1288 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1289 type parsing. "short" and "long" modifiers were also ignored; other
1290 modifier (such as reference level) are still ignored. [MQ]
1291 * Embeed the buffer size within the buffer itself on SG. [MQ]
1292 That way, send() are atomic and cannot get intermixed anymore (at least
1293 the ones which are less than 100k; bigger messages still have the issue)
1294 * Array size pushed by the field, not by the field type (or each
1295 and every long int will push stuff to the cbps) [MQ]
1296 * use select() to sleep since it allows to portably sleep less than one
1299 GRAS (minor cleanups)
1300 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1303 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1304 * type category 'ignored' killed as it was never used and were difficult
1306 * whether a type can cycle or not is now a flag, leaving room for more
1307 flags (as "ignored", if we feel the need one day ;)
1308 * Rename raw sockets to measurement sockets since "raw" has another
1309 meaning in networking community.
1312 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1313 and offers features not belonging to GRAS but that most applications
1314 need. One day, it may be a set of plugins loadable at runtime.
1315 * New module: bandwidth
1316 bandwidth measurement between arbitrary nodes running this module. [MQ]
1318 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1320 SimGrid (2.94) unstable; urgency=low
1322 The first beta release of SimGrid 3 !
1326 * Update the main page and the FAQ. Adding references to gforge.
1329 * Add a gras_os_getpid function.
1332 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1333 * Extend the logs so that they also print PID, hostname, date, ... if
1335 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1336 and kill the old version which were in testsuite/
1337 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1341 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1342 for users. Instead, each programming environment (GRAS, MSG,...) use it
1343 when needed to provide such a feature to users.
1344 Don't shortcut the mecanism or you will also shortcut the virtualization
1345 you need on the simulator.
1349 * Cleanups in configury with regard to compile optimization/warning flags.
1350 Also add -fno-loop-optimize to any powerpc since it's the optimization
1351 killing gcc (< 3.4.0).
1352 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1353 needed by MSG examples complications
1354 * Borrow configury beautifications from PHP
1357 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1358 warning (thanks loris for stumbling into this one).
1359 * Bugfix: stop loading private headers (gras_config.h) from the public
1363 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1364 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1365 RL would imply a DNS resolver, which is *hard* to do in a portable way
1366 (and therefore delayed).
1367 * Implement a real timer mecanism and use it in timing macros. This allows
1368 to avoid rounding errors and get a 0.000005 sec precision in timing
1369 macros. While I was at it, various cleanups:
1370 - allow to declare more than one timed section per file (fix a stupid bug)
1371 - move some private declaration to the right place
1372 - merge conditional execution and timing macros into emulation module
1373 - document the module
1374 - make sure the module cleanups its mess on gras_exit
1375 * Documentation improvements:
1376 - (new) how to compile applications using GRAS
1377 - (new) emulation support (timing macros)
1379 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1381 SimGrid (2.93) unstable; urgency=low
1383 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1386 - Use Paje properly where used. Still to be sanitized properly.
1387 - Portability fix: Add an implementation of the contexts using pthread
1391 - Add xbt_procname(): returns the name of the current process.
1392 Use it to show the current process's name in all logging.
1394 - fix detection of older flex version and the reaction, since we do
1395 depend on modern ones (we use lex_destroy)
1396 - Better separation of SG and RL in the libs: remove all simulation code
1397 from libgras. As a result, this lib is now only 200k when stripped.
1398 Some of the xbt modules may also be duplicated (two sets and such) and
1399 should be cleaned/killed before SG3.
1400 - Insist on using xlC on AIX because of weird problems involving gcc there.
1401 - Cleanup the make remote stuff. This is now done by scripts
1402 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1403 mainly for our private use, but we're working on changing them to user
1406 - Bugfix: flush the socket on close only if there is some *output*.
1407 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1409 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1412 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1414 SimGrid (2.92) unstable; urgency=low
1416 Alpha 3 on the path to SimGrid 3
1420 - New! First try of benchmarking macros.
1421 - New! First try so that gras_stub_generator generate deployment and
1422 remote compilation helpers.
1424 - Bugfix: Initialization fix in msg_test.
1428 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1430 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1432 - API Change: gras_os_sleep to take the amount of seconds as a double.
1433 Accepting an int was error prone since it was the only location where
1434 seconds were coded as such. It leaded to damn rounding errors.
1435 - Bugfix: Hard to belive that timers ever worked before this.
1437 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1439 SimGrid (2.91) unstable; urgency=low
1441 Alpha 2 on the path to SimGrid 3
1445 - Bug fix in the lmm_solver.
1447 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1448 through the function MSG_paje_output.
1449 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1450 - It is possible to bound the rate of a communication in MSG with
1451 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1452 to write it in the changelog).
1453 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1458 - Various cleanups to the autotools stuff
1459 - Begin to move Gras examples to examples/gras/
1460 - Let make distcheck work again (yeah!)
1462 - documentation overhauled using doxygen.
1463 gtk-doc-tools is dead in SimGrid now.
1464 - Automatically extract all existing logging categories, and add the list
1465 to the documentation (long standing one, to say the less)
1467 - Cleanup the known architecture table. Reorder the entries to group what
1468 should be, and use a more consistent naming scheme.
1469 (some of the test dataset are still to be regenerated)
1470 - New! Allow library to register globals on each process just as userdata
1472 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1473 lookup time (for now).
1474 Use it in msg and trp.
1475 This cleans a lot the internals and helps enforcing privacy of the
1476 headers between the gras components.
1477 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1478 - Bugfix: gras_os_time was delirious in RL.
1479 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1481 - Reenable GRAS now that it works.
1483 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1485 SimGrid (2.90) unstable; urgency=low
1487 Alpha 1 on the path to SimGrid 3
1489 * It is a long time since the last release of SimGrid. I'm sorry about
1490 that but as I had told you, I was rewriting a lot of things. I apologize
1491 to those who had been reporting bugs to me and that I had not answered.
1492 If your bug is still in the new version, please tell me. Here is a
1493 summary of the main changes.
1495 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1496 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1497 portability, speed, and a lot more but you'll figure it out later.
1498 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1499 I wanted to release the new MSG as soon as possible and I have
1500 broken GRAS, which is the reason why, for now, only MSG is fully
1501 functional. A laconic description of these projects is available
1502 in the documentation.
1504 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1505 kernel optimized for our needs (called SURF but only the developers
1506 should use it). Hence, MSG is now roughly 30 times faster and I think
1507 that by rewriting a little bit MSG, I could event speed it up a little
1508 bit more. Beside the gain in speed, it is also much easier to encode a
1509 new platform model with SURF than it was with SG. More to come...
1511 * REVOLUTION 3: I have tried to change a little as possible the API of
1512 MSG but a few things really had to disappear. The main differences
1513 with the previous version are :
1514 1) no more m_links_t and the corresponding functions. Platforms are
1515 directly read from a XML description and cannot be hard-coded
1516 anymore. The same format is used for application deployment
1517 description. The new format is described in the documentation.
1518 Have a look in tools/platform_generation. There is a tiny script
1519 that converts from the old platform format to the new one. Concerning
1520 the application deployment format, parsing the old one is tricky.
1521 I think most of you should however be able to convert your files. If
1522 it is really an issue, I can write a C code that does the conversion.
1524 2) the toolbox tbx does not exist anymore. We now have a library
1525 with much more data-structures but without the hash-tables (we have
1526 dictionaries that are much faster).
1528 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1530 *****************************************************************************
1531 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1532 * don't feel like converting the oldies. (Mt) *
1533 *****************************************************************************
1536 Version 2.90: "the long awaited one"
1537 - Finished rewriting and debugging MSG. Rewrote the documentation.
1538 - disable GRAS for now since it needs to be ported to the newest SG
1541 - Finish the port to windows (using mingw32 for cross-compile)
1544 - Main loop and datastructures of SURF. A cpu resource object is
1545 functional. Surf can thus be used to create cpu's with variable
1546 performance on which you can execute some actions.
1548 2004-11-15 Martin Quinson
1549 - Port to ARM. Simply added the alignment and size descriptions. Should
1550 work, but the ARM machines are so slow that I didn't had the opportunity
1551 to 'make check' over there yet.
1553 2004-11-15 Arnaud Legrand
1554 - Trace manager now written. It uses a heap structure and is therefore
1555 expected to be efficient. It may however be speeded up (particularly
1556 when many events occur at the same date) by using red and black
1557 trees. One day maybe...
1558 - Max-min linear system solver written. It uses a sparse matrix
1559 structure taking advantage of its expected use. Most operations are
1560 O(1) and free/calloc are called as few as possible. The computation of
1561 the minimum could however be improved by using a red and black tree
1564 2004-11-03 Arnaud Legrand
1565 - Rename every gras_* function that was in xbt/ to its xbt_
1567 - Add a heap and a doubly-linked list to xbt
1568 - Added a dichotomy to the dictionaries. make check works as well before
1569 so I assume that the patch is correct. I do not know however if things
1570 run effectively faster than before now. :)
1572 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1573 SimGrid, and the version number is bumped to 2.x
1575 2004-10-29 Martin Quinson
1576 - Introduction of the remote errors.
1577 They are the result of a RMI/RPC on the remote machine.
1578 ErrCodes being scalar values, you can't get the host on which those
1579 errors did happen. Extending the error mechanism as in Gnome is possible.
1580 No idea yet whether it is a good idea.
1582 2004-10-28 Martin Quinson
1583 - Interface revolution: the Starred Structure Eradication.
1584 I used to do typedef struct {} toto_t; and then handle *toto_t.
1585 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1586 - ???_t is a valid type (builded with typedef)
1587 - s_toto_t is a structure (access to fields with .)
1588 - s_toto is a structure needing 'struct' keyword to be used
1589 - e_toto_t is an enum
1590 - toto_t is an 'object' (struct*)
1592 typedef struct s_toto {} s_toto_t, *toto_t;
1593 typedef enum {} e_toto_t;
1594 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1595 s_toto_t) is private.
1597 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1598 so that it can set the variable to NULL. It was so for dicts and sets,
1599 it changed for dynars.
1601 - Fix a bunch of memleaks in dict_remove
1602 - Fix a bug in sg/server_socket opening: it failed all the time.
1604 2004-10-07 Martin Quinson
1605 - Speed up dynar lookup operation a bit.
1607 gras_dynar_get is dead.
1609 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1610 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1611 (which gives you the address of the stored data).
1613 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1614 copy of the data using an affectation to do the job and not a memcpy.
1616 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1617 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1619 It does not really speedup the dynar test because they are
1620 setting elements all the time (and look them seldom). But the dict does
1621 far more lookup than setting.
1623 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1625 2004-10-05 Martin Quinson
1626 - Allow to (en/dis)able the cycle detection at run time.
1628 Whether we should check for cycle or not is now a property of each
1629 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1630 datadesc_cycle_unset allow to remove this property when previously set.
1632 Note that the cycle detection is off by default since it impacts the
1633 performance. Watch the data you feed GRAS with ;)
1635 This property is hereditary. Any element embedded in a structure having it
1636 set have it set for the time of this data exchange.
1638 You should set it both on sender and receiver side. If you don't set it on
1639 sender side, it will enter an endless loop. If you forget on receiver
1640 side, the cycles won't be recreated after communication.
1642 - Header reorganization.
1643 Kill gras_private.h, each submodule must load the headers it needs.
1645 2004-10-04 Martin Quinson
1646 - Interface revolution: do not try to survive to malloc failure.
1648 Now, gras_malloc and friends call gras_abort() on failure.
1649 As a conclusion, malloc_error is not a valid error anymore, and all
1650 functions for which it was the only gras_error_t return value are
1651 changed. They now return void, or there result directly.
1652 This simplify the API a lot.
1654 2004-09-29 Martin Quinson
1655 - Re-enable raw sockets.
1656 Created by gras_socket_{client,server}_ext;
1657 Used with gras_raw_{send,recv}
1660 It should allow to kill the last bits of gras first version soon.
1662 This is not completely satisfactory yet (duplicate code with
1663 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1666 - Simplify transport plugin (internal) interface by not passing any
1667 argument to _server and _client, but embedding them in the socket
1670 2004-09-28 Martin Quinson
1671 - Finish the port to AIX.
1672 autoconf was my problem (segfault within the malloc replacement
1673 function. No idea why)
1675 2004-09-16 Martin Quinson
1676 - Fix some size_t madness on 64bit architectures.
1678 2004-09-08 Martin Quinson
1679 - Reduce the number of system headers loaded, overload some more system
1680 calls (such as malloc to cast the result of the system one, and work
1682 - Fix and reintroduce the config support
1684 2004-09-07 Martin Quinson
1685 - Source code reorganization to allow Arnaud to surf all over there.
1686 - Allow to document the logging categories.
1687 - Remove all uppercase from logging categories and useless cleanup in names.
1689 2004-08-18 Martin Quinson
1690 Version 0.6.2 (protocol not changed; API changed)
1691 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1692 gras_error_t), and NULL when not found. Functions expecting a msgtype
1693 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1694 hopefully usefull message.
1695 - Portability to prehistoric sparcs again
1697 2004-08-17 Martin Quinson
1698 Version 0.6.1 (protocol not changed; ABI not changed)
1699 - prealloc some buffers to speed things up
1701 2004-08-11 Martin Quinson
1702 Version 0.6 (protocol not changed; ABI expended)
1703 - The parsing macro can deal with the references, provided that you add
1704 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1706 2004-08-09 Martin Quinson
1707 Version 0.5 (protocol not changed; ABI changed)
1708 - Allow to off turn the cycle detection code in data exchange at
1709 compilation time. It should be at run time, but I'm short of time (and
1710 the config stuff is still broken). That way, we keep dict out of the
1711 critical path, which is good because the performance is poor:
1712 - search not dichotomial yet
1713 - dynar give no way to access their content and memcpy everytime
1714 - In composed data description (struct, ref and so on), stop foolness of
1715 keeping the subtype's ID, but store the type itself. This keeps sets out
1716 of the critical path, which is good since they rely on dynar and
1717 dictionnaries. The only loose of that is that we cannot detect the
1718 redeclaration of a structure/union with another content (but I'm not sure
1719 the code detected well this error before anyway). We still can detect
1720 the redefinition discrepancy for the other types.
1721 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1722 it breaks the code because of type-punning used all over the place).
1723 This breaks on all non-gcc architectures (for now).
1725 All those changes (plus the buffer of last time) allow me to gain 2 order
1726 of magnitude on cruel tests consisting of 800000 array of integers on two
1727 level of a hierarchical structure (200 secondes -> 4 secondes)
1730 - the selector of reference must now return the type it points to, not
1731 the ID of this type.
1733 2004-08-06 Martin Quinson
1734 Version 0.4 (protocol changed; ABI not changed)
1735 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1736 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1737 - Implement a buffer transport, and use it by default (it relies on tcp in
1738 real life and on sg in simulation).
1739 That's a bit hackish since I had a new field to the structure to store
1740 its data without interfering with the subtype ones. Inheritance
1741 is tricky in C. And that's a kind of reverse inheritance with one class
1742 derivating two classes. Or maybe a game with java interfaces. Anyway,
1743 that's damn hard in C (at least).
1744 Moreover, I got tired while trying to ensure plugin separation and
1745 genericity in SG mode. MSG wants me to do weird things, so let's go for
1746 cruel hacks (temporarily of course ;).
1747 See comment in transport_private.h:71
1748 - do not include all the _interface headers in private but in the files
1749 which really need them (to cut the compilation time when they are
1752 2004-07-26 Martin Quinson
1753 Version 0.3 (protocol not changed; ABI changed)
1754 - Major overhault of the datadesc interface to simplify it:
1755 - shorted the function names:
1756 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1757 - add a trivial way to push/pop integers into the cbps without malloc.
1758 This allows to make really generic sub_type description, which simply
1759 pop their size of the stack.
1760 - add a function gras_datadesc_ref_pop_arr() which does what users want
1761 most of the time: Declare a dynamic array (which pops its size of the
1762 stack) and declare a reference to it. Poor name, but anyway.
1763 - kill the post-send callback, add a post-receive one
1765 2004-07-23 Martin Quinson
1766 Version 0.2 (protocol changed; ABI changed)
1767 - add some testing for cpbs in the test cases, and fix some more bugs.
1768 This invalidate again the little64 data file, since I cannot regenerate
1770 - remove an awfull optimization in the logging stuff, allowing me to:
1771 - understand it again
1772 - learn gcc how to check that the argument match the provided format
1773 - fix all errors revealed by gcc after that
1774 - internal keys of dict are not \0 terminated. Deal with it properly in
1775 loggings instead of segfaulting when the user want to see the logs :-/
1777 2004-07-22 Martin Quinson
1778 - Fix some stupid bug preventing cbps (callback postit) from working
1780 2004-07-21 Martin Quinson
1781 - Some documentation cleanups
1782 - remove the useless last argument of msgtype_declare
1783 - rename the Virtu functions to fit into the 'os' namespace
1784 - move headers src/include -> src/include/gras/ and stop fooling with
1785 gras -> . symbolic link
1786 - make distcheck is now successful
1788 2004-07-19 Martin Quinson
1790 - Build shared library also
1791 - Install html doc to the right location
1792 - stop removing maintainer files in make clean
1793 - build tests only on make check
1795 2004-07-13 Martin Quinson
1797 - No major issue in previous version => change versionning schema
1798 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1799 corresponding dataset.
1801 2004-07-11 Martin Quinson
1803 - Get it working with any kind of structure (we can compute the padding
1804 bytes remotely for all the architectures I have access to)
1805 - Implement the structure parsing macro (still not quite robust/complete)
1806 - Improvement to the remote testing toysuite
1808 2004-07-10 Martin Quinson
1809 [autoconf mechanism]
1810 - get ride of a bunch of deprecated macros
1811 - actually run the test for two-compliment, not only compile it :-/
1812 - test whether the structures get packed (and bail out if yes. Damn.
1813 Alignment is a serious matter)
1814 - test whether the structures get compacted (but respecting the alignment
1815 constraints of each types)
1816 - test whether the array fields of structures can straddle alignment boundaries
1818 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1819 (took me 2 hours to find that bug, looking at the wrong place)
1820 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1821 before sending/receiving and are used to compute the offsets of fields
1822 - Given that padding size depend even on compiler options, keep track of
1823 alignment and aligned_size only for the current architecture. Not a big
1824 deal since we send structure fields one after the other (seems
1826 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1827 program, let it work on linux/gcc/little32. portability todo.
1829 2004-07-08 Martin Quinson
1830 - import and improve remote compilation support from FAST
1831 - make sure make check works on half a dozen of machines out there
1833 2004-07-07 Martin Quinson
1834 Let's say it's version 0.0.3 ;)
1835 - Implement conversions (yuhu!)
1836 - Let it work on solaris (beside conversion, of course)
1837 - Stupid me, using rand() to generate the conversion datatests in not wise.
1839 2004-07-06 Martin Quinson
1840 - Let make dist work, since I'm gonna need it to compile on remote hosts
1841 - Let Tests/datadesc_usage write the architecture on which the file was
1842 generated as first byte.
1843 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1844 architecture descriptions.
1845 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1846 on those architectures.
1847 - Optimization: send/recv array of scalar in one shoot
1849 2004-07-05 Martin Quinson
1850 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1852 - Plug a whole bunch of memleaks
1853 - each process now have to call gras_{init,exit}. One day, their log
1854 settings will be separated
1855 - Continue the code factorisation between SG, RL and common in Transport.
1857 2004-07-04 Martin Quinson
1859 - Redistribution between SG and RL.
1860 We wanna have to accept in SG, so move accepted related parts of RL in
1861 the common part. (more precisely, the dynar of all known sockets is no
1862 more a static in transport.c, but part of the process_data)
1864 [gras_stub_generator]
1865 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1867 2004-07-03 Martin Quinson
1868 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1869 them all up in one shot)
1871 - Fix the ugly case of reference to dynamic array.
1872 - New (semi-public) function gras_datadesc_size to allow the messaging
1873 layer to malloc the needed space for the buffer.
1875 - gras_socket_close now expect the socket to close (and not its address to
1876 put NULL in it after it). This is because the socket passed to handlers
1877 is one of their argument (=> not writable).
1879 - propagate the interface cleanup from last week in datadesc, ie remove a
1880 superfluous level of indirection. User pass adress of variable
1881 containing data (both when sending and receiving), and not of a variable
1882 being a pointer to the data. Let's say that I like it better ;)
1883 The price for that is constructs like "int msg=*(int*)payload" in
1884 handlers, but it's a fine price, IMHO.
1886 - Let it work in RL (yuhu)
1888 2004-06-21 Martin Quinson
1890 - porting SG plugin and SG select to new standards (works almost).
1891 - plug memleaks and fix bugs around.
1894 - cleanup the prototype of data recv and force users to specify when they
1895 want to handle references to objects. Test case working even for cycles.
1896 - plug memleaks. Valgrind is perfectly ok with this.
1898 2004-06-12 Martin Quinson
1900 - cleanup the separation between plugin and main code in plugin creation
1902 2004-06-11 Martin Quinson
1904 - Reput hook for raw sockets, needed for BW experiments
1905 - kill a few lines of dead code
1906 [Data description] Interface cleanup
1907 - gras_datadesc_by_name returns the searched type or NULL.
1908 That way, no variable is needed to use a type desc once, which makes
1910 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1911 two parameters were strings (field name, type name), leading to
1913 [Dicos] Interface cleanup
1914 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1915 This is consistant with the dynar API.
1917 2004-04-21 Martin Quinson
1919 - Porting to new standards.
1921 - interface cleanup.
1922 There is no bag anymore, no need to take extra provision to mask the
1923 pointers behind "ID".
1924 Better splitup of functions between files create/exchange/convert.
1925 This is still a bit artificial since convert and receive are so
1926 interleaved, but anyway.
1928 - add a queued message list to procdata (the ones not matching criteria
1930 - factorize some more code between SG and RL wrt procdata
1932 - use gras_exit in example to track memleaks
1933 - get rid of gs_example now that GS is properly integrated into gras
1934 - update run_test to integrate the lastest tests (datadesc)
1936 - rename WARNINGn macros to WARNn since it prooved error-prone
1938 2004-04-19 Martin Quinson
1940 - register init/exit functions within gras module mechanism
1941 - send/receive function.
1942 Convertion is not implemented, but short-cutted if not needed.
1943 struct/array elements are sent one by one (instead of block-wise), but
1944 nobody really cares (yet). Get a prototype before optimizing.
1945 - tests (using a file socket) for DD send/receive on:
1946 - base types: int, float
1947 - array: fixed size, string (ie ref to dynamic string)
1948 - structure: homogeneous, heterogeneous
1949 - chained list, graph with cycle
1950 Believe it or not, valgrind is not too unhappy with the results. The
1951 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1955 - Bugfix when using a filename explicitely (instead of '-')
1957 2004-04-09 Martin Quinson
1959 - factorize more code between RL and SG in socket creation
1960 - Complete the implementation and tests of:
1962 o file (only in RL, and mainly for debugging)
1964 I lost 3 days to design a portable address resolver, and then decided
1965 that the prototype mainly have to run on my box.
1966 Addressing portability too early may be like optimizing too early :-/
1968 - use gras_init in the Tests instead of the crappy parse_log_opt
1969 (the latter function is removed)
1970 [Conditional execution]
1971 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1972 [Code reorganisation]
1973 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1974 Build examples against the RL library, since there is no way to disable
1975 its creation for now.
1977 For information, the beginning of coding on GRAS was back in june
1978 2003. I guess that every line has been rewritten at least twice since