1 SimGrid (3.7) NOT RELEASED; urgency=low
3 The "Simplicity does not preceed complexity, but follows it" release.
6 * Change the correction factors used in LMM model, according to
7 the lastest experiments described in INRIA RR-7821.
8 Accuracy should be improved this way.
9 * Use the partial invalidation optimization by default for the
10 network too. Should produce the exact same results, only faster.
11 * Major cleanup in surf to merge models and split some optimization
12 mechanisms from the core of the models. As a result you can now
13 specify which model to use (e.g., --cfg=network/model:LV08
14 --cfg=cpu/model:Cas01) and which optimization mode to use
15 (e.g., --cfg=network/optim:lazy --cfg=cpu/optim:TI).
16 Incompatible combinations should err at initialization. See
17 --help-models for the list of all models and optimization modes.
18 * The CLM03 workstation model were dropped for simplicity because it
19 used the deprecated CM02 network model. Use default instead.
20 * Rename the TCP_gamma configuration option to network/TCP_gamma
21 * Rename the coordinates configuration option to
22 network/coordinates, and document it
23 * Use now crosstraffic keyword instead of the terribly missleading
24 fullduplex keyword. It is activated by default now in the current
25 default model, use --cfg=network/crosstraffic:0 to turn it off.
28 * Stabilize the parallel execution mode of user contexts
29 * Introduce configuration variables to control parallel execution:
30 - contexts/synchro: Synchronization mode to use when running
31 contexts in parallel (either futex, posix or busy_wait)
32 - contexts/parallel_threshold: Minimal number of user contexts
33 to be run in parallel (raw contexts only)
36 * Performance boost by using a swag internally to compute the set of
37 tasks that are finished and should constitute the return value of
41 * Mallocators: allow value NULL for the reset function.
42 * Dicts: new function xbt_dict_new_homogeneous(void(*)(void*)) to
43 create homogeneous dictionaries, where all the elements share the
46 -- $date Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
48 SimGrid (3.6.2) stable; urgency=low
50 The "Not coding new stuff allows to polish old things" release.
53 * New bindings to the NS3 packet level simulator (experimental)
54 * Use the raw (efficient) execution contextes instead of the sysv
55 (portable) ones when possible.
56 * libpcre is now mandatory in any cases since not using it led to
57 severe performance loss and possibly other issues
58 * Update the XML platforms:
59 - G5K: include the lastest machine in Nancy
60 - GridPP and LCG: new platforms
61 * Documentation was partially updated, at least (more to come)
63 Bug fixes, cosmetics and small improvements
64 * Free terminated processes before the end of the simulation to avoid
65 exhausting the memory of users having very dynamic amount of
67 * Bug fix and cosmetics about canceling non-running tasks
68 * Bug fix about the dot loader's issues when using libcgraph
71 * Create an installer for windows with nsis (amd64 and win32)
72 - Add an hello world project to illustrate simgrid project creation.
73 - Embed libpcre into the Simgrid installer to avoid
74 its compilation burden
75 * The raw execution contextes should work on Apple now
76 * Port to Windows 64 bits
77 - Sysv contextes now have an implementation for this arch
78 - GRAS communication features now support this arch
79 * Drop support for borland compiler on windows
80 - this code was not maintained, and we kinda depend on gcc nowadays
81 * Fix portability issues on kfreebsd/gnu: build error about semaphores
82 * Fix portability issue on unstable ubuntu: linker became picky on
85 -- Wed Oct 5 15:51:01 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
88 SimGrid (3.6.1) stable; urgency=low
90 The "Oops, we broke Macs too" release
93 * Fixed contextes detection so that raw ones are used when possible
94 * On Mac, do not use Posix Ucontextes with gcc v4.[1-5] since this
95 leads to a strange error, with user code segfaulting sometimes when
96 the generated code is not perfectly aligned (which is not
97 controlable from the user side, depends on the amount of code)
100 * New macro: CATCH_ANONYMOUS, which is like CATCH(e) but without argument.
102 -- Mon Jun 27 13:59:03 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
104 SimGrid (3.6) stable; urgency=medium
106 The Summer Release, also known as the "OMG! They Killed Kenny!" version
109 * Bindings now constitute their own package, separated from the main one.
110 Rationale: reduce our maintainance nightmare by reducing the module coupling
111 They will soon be released on their own on gforge.
113 svn co svn://scm.gforge.inria.fr/svn/simgrid/contrib/trunk/simgrid-java
114 svn co svn://scm.gforge.inria.fr/svn/simgrid/contrib/trunk/simgrid-ruby
116 GRAS: It is not considered as stable anymore, but experimental. Sorry.
117 * It's not quite deprecated for now because we have no replacement,
118 but it may soon become the case.
121 * New MPI functions supported: MPI_Comm_disconnect, MPI_Comm_get_name
122 * Fortran: New user-level cache variable to store the rank of the running
123 process. This improves performance by an order of magnitude.
124 * C: New coccinelle script to automatically locate and modifiy global and
125 local static variables.
126 * Improved SMPI network model with a sender-side gap to account for multiple
130 * New function MSG_comm_get_status(). MSG_comm_test() and MSG_comm_testany()
131 only say if a communication is finished, no matter whether it succeeded or
132 failed. You can call MSG_comm_get_status() to know the status of a finished
134 * New function MSG_task_dsend() to send a task and detach it. When a
135 communication is detached, you are never notified of its success or failure
136 and the memory is released automatically once it is finished. This function
137 is useful when you don't care about the end nor the success of a
139 * Change the prototypes of action replay. Sorry for inconvenience,
140 but this is really more efficient this way (and to adapt your code,
141 you just have to fix the initialization, that shouldn't be too long)
142 * Kill the braindead MSG_task_refcount_dec() function. I guess nobody
143 ever managed to do anything useful with it.
144 * New function MSG_comm_testany(). Similarly to MSG_comm_waitany(), it
145 takes a dynar of communications. It returns immediately and gives the
146 index of a finished communication (if any).
147 * New example: a basic implementation of the Chord P2P algorithm.
150 * New model for multi-core CPUs. You can now use the core attribute to
151 precise the number of cores of a host. This is a basic model. Every
152 process running on the host receives at most the power provided in
153 the DTD (throughput<=power). Total throughput of process cannot exceed
155 * New peer tag. This peer tag creates a tiny AS comprising a host and a
156 router linked by an up-link and a down-link (possibly asymmetrical).
157 This kind of pattern allows to easily build last-mile model style platforms.
158 Aggregating such patterns in a rule-based AS is thus the technique of
159 choice for modeling large peer-to-peer/volunteer computing/cloud platforms.
160 * New model for Vivaldi routing. We transformed the Vivaldi network model
161 into a Vivaldi routing model (based on the rule-based model). This allows to
162 combine Vivaldi based latencies with last-mile platforms.
165 * Added a check for NaN of IEEE754 infinite in the double entries of
167 * Introduce a new context factory "raw", highly inspirated from the
168 ucontext factory, but using manually crafted functions in assembly to
169 do the work in an efficient manner.
170 * Allow to change the used context factory at run time, not only at
171 compilation time. Use --cfg=contexts/factory:raw for maximal speed.
172 * Add an option --cfg=contexts/stacksize:N to set the stack size of the user
173 contextes at runtime (only with raw contexts or ucontexts).
174 * Completely rewrote this module to allow parallel execution of user
175 processes. Use --cfg=contexts/nthreads:N to execute user processes
176 with N parallel threads (the default is 1, meaning no parallelism).
177 * Allow to decide dynamically between sequential and parallel modes.
178 When nthreads > 1, you can use --cfg=contexts/threshold:P to run the user
179 processes in parallel only when their number is greater than or equal to P
181 * Added a check for NaN of IEEE754 infinite in the double entries of
185 * New command line option: if you pass --cfg=verbose-exit:0, SimGrid
186 won't output the state of processes when interrupted with Ctrl-C
187 * Add a new function xbt_dynar_to_array that transforms a dynar into a
188 NULL-terminated array. This may solve backward compatibility issues
189 due to the change to return type of SD_simulate. See also:
190 http://lists.gforge.inria.fr/pipermail/simgrid-user/2010-December/002206.html
191 * Add new macros with variable number of arguments.
192 - in xbt/log.h: XBT_DEBUG, XBT_VERB, XBT_INFO, etc.
193 - in xbt/asserts.h: xbt_assert
194 - in xbt/cunit.h: xbt_test_{add,fail,assert,log}
195 - in xbt/ex.h: THROWF and RETHROWF.
196 Define XBT_USE_DEPRECATED if you want to use the old numbered macros like
198 * Change xbt_die() to accept a format string with arguments, just like printf.
199 * New data structure: xbt_lib_t, like a dict but more general and with better
203 * New configuration options
204 Options triva/categorized and triva/uncategorized can be used to generate
205 graph configuration files for Triva visualization tool.
206 * Configuration option tracing/platform is renamed to tracing/categorized
207 * XBT logging makes tracing error checks easier, new root log hierarchy: instr
208 * New TRACE_user_link_variable interface:
209 User provides the name of the link and the tracing variable to attach to it
210 * the declaration of tracing categories must be done after the environment
212 * simpler tracing interface, just one way to declare categories
213 TRACE_category or TRACE_category_with_color, it is up to you
214 * links in the trace file are again identified by their names
215 * trace contains the full platform hierarchy exactly as declared using the ASes
216 * Options tracing/msg/[task|process]:1 groups the process by hosts
217 for both cases, tasks and processes must have names that are unique during the simulation
218 these options generate traces that are suited to gantt-charts, such as the space-time view of Paje
219 * The experimental option tracing/msg/volume is deprecated
220 its functionality may be reincorporated if needed
222 The tracing generates a trace file with unordered timestamped events,
223 because of the way the core simulator (surf) works. A script available
224 at the tools directory (fix-paje-trace.sh) can be used to put the events
225 in order. We have changed the tracing so it can generate ordered timestamped
226 events in the final trace, but depending on the simulator (and how much time
227 is simulated) that can lead to a huge memory utilization. It is desactivated
228 by default, but it can be activated using the --cfg=tracing/buffer:1 switch.
231 * Define a SIMGRID_VERSION macro in simgrid_config.h.
232 - We are trying hard to keep the API stable, but it may happen that
233 some things change (we're a research project after all, not a
234 nuclear plant operating system). If such things should happen, you
235 could rely on that macro to adapt.
236 - current value: 30600 for 3.06.00, aka 3.6
237 * Define macro MAKE_SIMGRID_VERSION(major, minor, patch) to help building
238 a number that can be compared with SIMGRID_VERSION.
239 * Add a build option -Denable_debug (set to ON by default): when set to OFF,
240 assertions and verbose/debug logging events are disabled at compile time.
242 -- Tue Jun 21 08:57:43 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
244 SimGrid (3.5) stable; urgency=medium
247 * New feature to any SimGrid-based simulator: Model-Checking
248 Check SIN#1 for more details.
251 * New Model SMPI (three-interval linear regression for correction factors)
252 See RR-7426, available at http://hal.inria.fr/inria-00527150
253 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
254 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
255 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
256 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
257 * See also src/smpi/README
262 - Tracing API changes: TRACE_start and TRACE_end should not be called
263 by user-code. They are automatically called by simulators created
264 with SimDAG, MSG and SMPI if the toolkit is compiled with
265 tracing_enabled=ON. Categories declaration and utilization remain the
266 same for MSG and SimDag.
267 - A function was added to the tracing API to declare categories with
269 - TRACE_category_with_color (char *category, char *color)
270 where color must be in the following format
271 "%f %f %f", red, green, blue
272 and red, green, blue are float values in the interval [0, 1]
273 - User can specify NULL as color parameter, or continue calling
275 On that case, the tracing system will define random colors
276 - The following command-line options are supported:
277 --cfg=tracing/filename:msg.trace
278 --cfg=tracing:1 (activate tracing, needed to use others)
279 --cfg=tracing/platform:1 (categorized resource use)
280 --cfg=tracing/uncategorized:1 (uncategorized resource use)
281 --cfg=tracing/msg/task:1 (task creation)
282 --cfg=tracing/msg/process:1 (process creation, migration)
283 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
284 --cfg=tracing/smpi:1 (SMPI interface tracing)
285 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
286 - examples of examples/msg/tracing updated
288 - DAXLoader and DOTLoader functions can generate tasks with categories
289 - A new function to attribute a category to SD tasks:
290 TRACE_sd_set_task_category (SD_task_t task, char *category)
291 * Tracing the MPI interface implemented by SMPI
292 - Collective operations are traced with states
293 - Point-to-Point operations are traced with states/links
294 - Tracing activated by a parameter "-trace filename" passed
295 to smpirun during execution (considering that simgrid
296 is compiled with tracing enabled)
297 - To run the simulation with gdb, the simulator
298 accepts --cfg=tracing/smpi:1 to trace SMPI
299 - tesh files to check if smpi tracing is ok
300 - See examples/smpi/NAS/DT-trace
301 * GTNetS tracing re-worked
302 - adaptation to the tracing system of GTNets to cope
303 with modifications regarding the fullduplex mode
304 - new tesh files to check if gtnets tracing is ok
307 * Asynchronous communications through the functions:
308 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
309 * New function: MSG_load_platform_script()
310 to make possible using a lua script instead of XML files to set up platforms
311 * New function: MSG_set_function
312 to associate functions to processes, used when bypassing the parser
313 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
315 Platforms: Add some more examples in examples/platforms
316 * Grid'5000: see www.grid5000.fr
317 * *_30000_hosts.xml: various huge files [mainly scalability testing]
320 * Change the XML format. This is a very important modification. SimGrid
321 3.5 introduces a new hierarchical format based on the notion of
322 Autonomous Systems. Compatibility with old format is ensured
323 through the perl script provided in the install bin directory
324 bin/simgrid_update_xml.
325 It is now possible to build platforms with specific routing
326 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
327 connect several platforms together. We will try to provide soon
328 set of realistic platforms exploiting these properties (have a
329 look at examples/platforms/ for the moment).
330 * Take the opportunity of the XML format change to be a good XML citizen:
331 rename link:ctn to link_ctn and similar changes (also dealed with by
333 * Add a new routing scheme (rule-based) using regular expressions. It
334 enables to have an extremely low memory footprint when the
335 underlying routing is simple and can be compactly described. You
336 need to have libpcre4-dev (perl regular expressions) installed if
337 you want to use this routing scheme.
338 * Revive the cluster TAG and allow to easily and efficiently (both in
339 term of memory and speed) connect clusters together. Have a look
340 at teshsuite/simdag/platforms/ to see how this can be done. With
341 this tag, you can create clusters with thousands of tasks at no
342 cost (have a look at examples/platforms/).
343 Note: clusters are implemented as ASes, so there is no need for an
344 enclosing AS tag if you have only one cluster in your platform.
345 * Add new generic functions in the public interface that allows the user
346 to call SURF 'create_resource' methods from your code (same
347 functionality as the XML bypass mechanism but with a much lighter
349 * Add a new model (enabled through command line --cfg=network/model:SMPI)
350 that uses a piecewise linear approximation to produce better
351 results when exchanging small messages.
352 * Add a new parameter to handle correctly full duplex link and account
353 for interferences between uplink and downlink communications
354 (activate with --cfg=fullduplex:1).
357 * Rename the SD_READY (all dependencies are satisfied and task is
358 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
359 dependencies are satisfied) state.
360 This prevents a confusion between the notion of "ready to schedule"
361 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
362 simulated" (SD_RUNNABLE) used by the simulation kernel.
363 * Change the way a task is considered as ready. Instead of removing
364 dependencies when a task is done, a counter is decreased. This way,
365 it is always possible to reach ancestors thanks to the
366 SD_taks_get_parents function (even after the end of the simulation.)
367 * Change the return type of SD_Simulate from (SD_task_t*) into
368 xbt_dynar_t. This function was in handling a dynar internally and
369 converted it into a NULL terminated array for historical reasons.
370 * New function SD_dotload(char*) to load a DAG described in dot
371 format. This loader and the corresponding examples require the
372 installation of the graphviz library.
373 * Fix a bug in the management of tasks of size 0 in the surf network
374 models. This problem was only visible with SIMDAG and you should
375 thus disregard results produced with earlier versions if you
376 relied on this feature (some tasks were blocked because of this).
377 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
378 even though your applications were doing only point-to-point
379 communications and sequential computations. Now you can really use any
380 model you want (of course, if you create real parallel tasks, which are
381 not implemented in most models beside ptaskL07, this will abort).
382 * Add an example that schedules a DAX on an heterogeneous platform
383 using a Min-Min strategy.
384 * New function SD_workstation_get_current_task() that returns the kind
385 of task currently running on a workstation in the sequential access
387 * Raise some warnings when unexecuted tasks remains at the end of the
388 simulation. This is usually caused by cycles in the DAG.
391 * New function: SIMIX_process_set_function() called by MSG_set_function
392 * Change the underlying waiting queue in semaphores so that a process
393 can wait on several of them simultaneously (as in waitany).
394 * Fix the way to handle tokens in semaphores so that all access patterns
395 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
396 * kill the dirty pimple SIMIX_message_sizes_output()
397 Please use (proper) visualization instead
400 * New data container: setset (set of sets of elements)
401 * New module: mmalloc (mapped malloc, allowing to have several
402 independent segments of malloc)
403 * New function: xbt_dict_cursor_set_data()
404 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
405 * New function: xbt_dynar_is_empty()
406 * New function: xbt_fifo_get_last_item()
407 * Fix xbt_dynar_shrink(): use the right element size.
408 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
409 uninitialized areas during expand.
410 * Fix semaphores: previous implementation was severely broken.
411 * Use library init/fini functions for our initialization.
412 - you can use logs and other feature as soon as you want in your
413 code (even before the xbt_init / MSG_init)
414 - xbt_exit is now a no-op and produce a warning when used.
417 * Port GRAS to new SIMIX mechanisms. This allows gras users to
418 benefit from the latest improvement to the simulation kernel.
419 * Kill measurement sockets for now. If you rely on them, sorry. This
420 release is not for you. This feature will be reintroduced in the
421 future, but we cannot delay the release any further.
422 * New function: gras_msgtype_get_name().
423 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
424 * Fix (at last) the pmm example: it should not randomly fail anymore.
426 Build chain: bug fixes and overall polishing
427 * Cmake is now stable enough. Hence, we killed the autotools.
428 * Port to windows ( TM :)
429 * Fix the 'make install' target.
430 No need to use 'make install-simgrid' anymore
431 * Introduce a 'make dist' target compiling a *source* archive
432 'make package' compiles a binary archive
433 * Compile java files only on need
434 * Add --cd and --setenv command line options to tesh
435 * Out of source builds are not fully supported yet, but we are close
436 * Enable supernovae and optimization flags by default for our users
439 * Add layer to set up environment directly from lua, without XML.
440 * The effect of gras_stub_generator can be achieved through
441 lua too (check examples/gras/console/ping_generator.lua)
443 -- Wed, 01 Dec 2010 22:09:23 +0100 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
445 SimGrid (3.4.1) stable; urgency=low
447 The "Polishing easter eggs is probably a good idea" release.
448 This is a bug fixes release only.
452 * Fix a bug preventing the tasks from begin garbage collected.
455 * Fix a bug occuring when a host involved in a communication fails.
456 This was not detected properly by the other peer involved in the
457 communication. Now, it's reported as a network error.
460 * Warn the user about loop dependencies in data flow of DAX files
461 * Obey the control-flow dependencies of DAX files
464 * Add option "enable_smpi" allowing to not compile SMPI.
465 Probably useful for the (Mac) users experiencing a build error here
466 * Improve the detection of lua5.1 and ruby1.8
468 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
470 SimGrid (3.4) stable; urgency=low
472 The "Easter in Cargese" release. Also known as (major changes):
474 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
475 ~> bindings were greatly improved
476 ~> new tracing infrastructure for better visualization introduced
478 * the "Welcome to configury modernity" release.
479 ~> we switched from autotools to cmake, and improved our cdash
482 A more detailled list of changes follow (full detail in svn log).
484 Java Bindings: Various Cleanups
485 * (install java-gcj-compat-dev on debian-like to use them)
486 * Remove put/get: no need to export deprecated interface in Java
487 Use send/receive instead.
488 * Cleanup the examples and add a README per directory
489 * Remove example autoDestination (that's the only way to go now)
490 * Remove example explicitDestination (was a plain copy of basic)
491 * Make JniException a runtime exception, so that there is no need to
492 declare the fact that you may encounter such a beast. I guess that
493 nobody will ever want to survive such error.
494 * Create specific errors for each MSG case of failure:
495 host failure, transfer failure, timeout, task cancelled
496 * Cleanup the exceptions that may get thrown by each function
497 * Other internal cleanups in Java bindings. Performance still bad :/
498 Ruby and Lua Bindings: create them
499 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
500 * That's new and great, you should try them out.
501 Same functionalities than Java bindings, only even less polished
503 * Kill the useless "rate" argument of SD_task_get_execution_time()
504 Everyone used to provide -1 as a value, it was not used, and the
505 semantic of a possible use wasn't even clear.
506 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
507 either as comm costs or compute costs to mean that there is no
508 such thing for that specific task.
509 * Add a SD_task_set_name() function
510 * Fix SD_task_unschedule() on typed tasks
511 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
512 * In DAX loader, accept useless 'level' attributes to <job> since
513 LIGO DAGs have them (seem to be to ease graphical representation).
515 * Add an example masterslave_mailbox.c using send/receive and not
516 the deprecated put/get interface.
517 * Kill the MSG_paje_output() function. It's a noop since 2 years.
518 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
520 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
521 (declare MSG_USE_DEPRECATED to still have the old name)
522 * Add a MSG_task_set_data() function
523 * About trace replay (see examples/msg/actions):
525 - Allow to work with splitted trace files for each process
526 Give the specific trace file as argument of each process,
527 and call MSG_action_trace_run(NULL)
528 You can still have one merged file for all processes.
529 - Fix implementation of collective operations
530 * Allow task_execute() on 0-sized tasks (closes #10063)
532 * This is the first release of SimGrid where SMPI is not considered
533 beta anymore (even if some corners should still be improved)
534 * Port over the new SIMIX_network submodule (internal refactoring)
535 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
536 * Implement more missing elements of the standard:
538 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
540 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
541 MPI_Scan() MPI_Get_processor_name()
542 - Added implementation of missing case for Alltoall (warning: it's
543 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
544 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
545 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
546 SMPI_MPI_Allgatherv()
548 - MPI_Waitsome() was broken
549 - Allow relative includes in smpicc
550 - Command line cfg argument 'reference_speed' was ignored...
551 - Some functions did not properly lead to auto-benching of user code
552 - smpicc passes -O2 by default (just like openmpi one)
554 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
555 * Bug fixes about timeouts during communications
556 * add SIMIX_message_sizes_output() as a pimple to write to file the
557 amount of messages per size. Use gnuplot to get histogram.
558 Pimple because that's the only user-visible function of simix,
559 defined directly in xbt.h (irk, sorry)
561 - Add a SIMIX_sem_get_capacity() function
562 - Fix interactions with processe resume/suspende
563 - release_forever() was stupidly broken
564 - Fix SIMIX_display_process_status() for processes in a semaphore
565 - Make SIMIX_sem_block_onto() user-visible
566 * Refactoring context stuff:
567 - Use pseudo-OOP for better modularity
568 - reimplement SIMIX_process_kill() without process_schedule() so
569 that the latter can take as invariant that it is called from
571 - Merge context_start into context_new for sake of simplicity
573 * Add a Vivaldi network model, coded live during SUD'10 ;)
574 * Rename configuration variables to start a hierarchy:
575 o cpu_model -> cpu/model
576 o network_model -> network/model
577 o workstation_model -> workstation/model
578 * New configuration variables:
579 o network/bandwidth_factor: correction to bandwith
580 o network/latency_factor: correction to latency
581 o netwotk/weight_S: correction to the weight of competing streams
582 * Add a long description to the models, that users can see with such
583 argument on the command line: --cfg=cpu/model:help
584 * --help-models display the long description of all known models
586 * config: add the ability to set a default value after registration
587 Does not override any previously set value (e.g. from cmd line)
588 * dict: allow to have integer key and data.
589 When so, you need to use the following functions
590 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
591 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
592 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
593 In contrary to regular dicts, the key is not malloced before copy.
594 Mixing scalar and regular elements in the same dict is not tested
596 * Allow to use xbt_dynar_shrink() to expend the dynar instead
597 Tracing for Visualization:
598 * SimGrid is now instrumented in order to generate a trace file for
599 visualization analysis: to use it, need to compile SimGrid with the
600 "tracing" option enabled, and instrument the program using SimGrid with
601 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
602 (among other functions).
603 * The instrumentation only traces the platform utilization for now
604 * Documentation to use the tracing functions and how to analyze the
605 traces with the Triva tool is written.
606 * More information about: SimGrid FAQ (in the section Tracing Simulations
609 * We moved to cmake as default build system. Autotools support will
610 be dropped soon. Check the FAQ for more info about how to use it.
611 * Greatly improved our cdash/ctest interactions
612 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
613 * Added memory checking tests with valgrind; lot of memleak fixing.
614 This may be the first release of simgrid with so few memory issues
615 * Added code coverage tests.
616 Our coverage is still improvable, but at least we see it on cdash.
618 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
620 SimGrid (3.3.4) stable; urgency=low
622 The "Desktop Grid needs love too" release (also called Xmas release).
625 * Major speedup in the maxmin system solving by using lazy evaluation
626 Instead of solving completely the maxmin system at each iteration,
627 only invalidate (and recompute) the modified parts.
628 This new feature is enabled in default models but you can try to
629 turn it on with "--cfg:maxmin-selective-update=1" for other models.
630 * Cas01 IMproved as default CPU model
631 This CPU model is the same Cas01 model, but it uses the
632 maxmin-selective-update flag and a heap structure to manage
633 actions on SURF kernel.
634 It reduces the complexity to find the next action to finish and,
635 consequently, it's faster than the old Cas01.
636 This is the new default CPU model (Cas01).
637 * Rename the old Cas01 model to Cas01_fullupdate
638 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
639 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
640 * CpuTI (CPU Trace Integration)
641 A new CPU model whose objective is simulate faster when using
642 availability trace files.
643 Instead of using a full featured, over engineered maxmin system for
644 CPU modeling, this model does the pre-integration of traces files
645 to calculate the amount of CPU power available, and so, executes
646 faster than the old CPU models.
647 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
648 * Use LV08 as default network model since it gives better accuracy
649 for small messages and shouldn't change things for big ones.
650 Use --cfg=network_model:CM02 to get the previous behavior.
653 ******************************************
654 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
655 ******************************************
656 * The new CPU model may changes simulations!
657 The point is that events occurring at the exact same timestamp
658 are not scheduled in the same order with the old and new
659 version. This may be enough to completely change the execution
660 of simulations in some cases.
661 * The new network model will change simulations!
662 This new model is more realistic than the previous one, so you
663 should consider redoing your old experiments with this model.
664 Sorry for the inconvenience.
667 * Introduce the supernovae compilation mode
668 When compiled that way, the whole SimGrid (or almost) is put in a
669 single compilation unit and compiled in one shoot.
670 This is to help gcc which has difficulties to inline stuff from one
672 The speedup seem to be above 15%, althrough more tests are needed on
673 amd64 to confirm that gain.
676 * Port of MSG's mailbox on top of SIMIX network
677 The put/get mechanism was greatly simplified on the way.
680 * New SIMIX network module. Provides:
681 - Mailbox: rendez-vous mecanism to find with who you want to speak
682 - Synchronous send/recv: easier and hopefully faster since the
683 logic is handled in the maestro process directly now
684 - Asynchronous send/recv: you dreamt of it? It's here now
685 Too bad that nobody cared enough to propagate the change to MSG.
686 * Add semaphores as SIMIX synchronization mechanism.
689 * new function SD_daxload(char*) to load a DAX file
690 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
691 * Introduce typed tasks. Specify its kind and cost at creation.
692 At scheduling, just give where it should be placed, and the cost
693 for each involved resource is automatically computed.
694 Existing constructors so far (more to come of course):
695 - SD_task_create_comm_e2e() for end-to-end communication
696 - SD_task_create_comp_seq() for sequential computation
697 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
698 * new function SD_task_dump() for debuging display
699 * new function SD_task_dotty(task,FILE*) writing to file the info
700 about the task in dotty format
701 * SD_task_dependency_exists() can now cope with having one of its
702 arguments NULL. If so, it tests whether the other argument has any
704 * Add getters on list of preceding/following tasks:
705 SD_task_get_parents(task) and SD_task_get_children(task)
706 * Add getters on amount of workstations and list:
707 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
708 * Add getter on task kind: SD_task_get_kind(task)
709 * Update the start_time and finish_time of tasks on completion/failure
710 * Bugfix: Remove task from state swags when destroyed
713 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
716 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
717 dynar with one separate thread per value of the dynar.
718 * Change the prototype of xbt_thread_create(), sorry.
719 Added a boolean parameter indicating whether we want to join this
720 thread (used in SG only for now)
721 * Implement xbt_thread_join and xbt_thread_yield in SG also.
724 * GTNetS wrappers should now be usable again (and betterly tested too)
725 * Fix a major regression from 3.2 where the timeout provided to
726 MSG_task_put_with_timeout() was used as absolute time before which
727 the comm should be done.
728 * Start to fix the <cluster> tag.
729 - Internal links should be good now (beside of the loopback, which
730 use the private link instead)
731 - paths to the external world is still rather broken
732 - the <route:multi> tag is just broken. Actually that's brain-dead.
733 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
734 to make it less stupid
735 ** Check your platform with teshsuite/simdag/platforms/flatifier **
736 * Fix a source-level compatibility glitch from 3.2: after defining
737 MSG_USE_DEPRECATED, you can use the old name
738 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
739 * Allow to compile from the SVN with automake 1.11
740 * Fix some problems when using the "start_time" tag in deployment XMLs.
741 * Fix #8569: XBT/synchro.h has redundant declarations
742 * Fix #8563: MSG return values and exceptions
743 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
744 * Integrate patch #8636: Obey DESTDIR when installing documentation.
745 Thanks to Robson Peixoto.
746 * Fix a vicious bug in dictionaries inducing that some elements were
747 not freed on xbt_dict_free()
749 Portability report of this version:
750 * Main portability targets:
751 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
753 - mac leopard on i386
754 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
755 but nothing critical.
756 * Other platforms: windows, AIX and others were not tested for this release
758 Timing report of this version:
759 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
760 scenario parameters). From 8h to a few seconds in desktop grid settings.
761 * Supernovae brings about 25% speedup on i386.
763 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
765 SimGrid (3.3.3) stable; urgency=low
767 The "Need for Speed" release.
769 The timings done to validate the 3.3.2 were faulty.
770 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
772 The problem was a conversion from a manually handled vector to
773 xbt_dynar_t on the critical path.
774 xbt_dynar_foreach calls functions, inducing stack management crap.
776 We inlined these functions and xbt_dynar_foreach is now breath taking.
777 We also inlined xbt_swag_belong on the way.
779 Here are some approximate speedup measurements (on master/slaves
780 simulations lasting between 10s and 20s each):
781 3.3.1 -> 3.3.2: about same performance
782 3.3.2 -> 3.3.3: 40% speedup
783 3.3.1 -> 3.3.3: 40% speedup
784 3.3.1 with inline patch -> 3.3.3: 30% speedup
786 Our reading is that the refactoring which occurred in 3.3.2 made us
787 suffer much more from the xbt_dynar_foreach low performance, but
788 once we solved this, this refactoring proved to be very performance
789 effective. From the 40% speedup, somehow, 10% are due to the
790 inlining and 30% to the refactoring.
792 That's a pitty that gcc cannot inline functions placed in other files
793 alone. We have to choose between:
794 - break the encapsulation (by putting private data structures and
795 accessors in headers files to help gcc)
796 - live with low performance
797 - switch to a decent compiler such as icc (not quite possible).
799 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
801 SimGrid (3.3.2) stable; urgency=low
803 The "Simplicity does not preceed complexity, but follows it" release.
805 The main contributors of this release were (lexical order):
806 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
809 * Extract the routing logic into its own object.
810 (was dupplicated in network.c and workstation_LV07.c;
811 Allows to implement other ways of storing that info)
812 => kill now useless network_card concept
813 - Use dynar to represent routes (instead of void** + int*)
814 - kill link_set (use surf_network_model->resource_set instead)
815 - Add a command-line option to choose the routing schema to use
816 - Add three new models:
817 * Floyd (shortest path computed at initialization)
818 * Dijikstra (shortest path recomputed all the time)
819 * Cached Dijikstra (shortest path computed on need)
820 All these models where contributed by Silas De Munck, and are
821 described in his ICCS09 paper.
823 * Simplify model declaration
824 (less redirections, less function to write when defining a model)
825 - Factorize stuff between models:
828 surf_model_resource_set(model)
829 surf_model_resource_by_name(model, name)
830 - Unify the types of models in s_surf_model_t (using an union)
831 - Embeed fields of common_public directly into s_surf_model_t
832 - Rename model methods:
833 action_free ~> action_unref
834 action_change_state ~> action_state_set
835 action_get_state ~> action_state_get
836 - Change model methods into functions :
837 (model)->common_public->action_use ~> surf_action_ref
839 * Implement a generic resource; use it as ancestor to specific ones
840 (allows to kill duplicated code in models)
841 Drawback: timer command don't need no name nor properties;
842 workstation_CLM03 don't need no properties
843 (but I guess we can live with those few bytes wasted)
845 * Improve the action object model
846 - implement a constructor avoiding dupplicated code about field
847 initialization in generic_action part.
849 * Kill the SDP model: it has an external dependency, is deprecated
850 in flavor of modern lmm models, and didn't compile since a while
853 * Relocation of the context module from XBT to SIMIX.
854 (the context were decoupled from the simix processes, duplicating a lot of code)
855 => a lot of code was factorized
856 - less overhead is introduced during scheduling
857 - simpler API for the context factory
858 - the logic for process creation,destruction and manipulation was simplified
859 * Simplification of the s_smx_process_t data structure.
860 => accesing the simix level data associated to a process is faster now,
861 and the code is a lot more readable.
864 * Implement some more MPI primitives:
865 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
866 -implementation: Bcast: flat or 2-ary tree (default),
869 Allreduce: Reduce then Bcast
870 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
871 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
872 Alltoallv: flat tree, like ompi
874 * Add support for optimized collectives (Bcast is now binomial by default)
875 * Port smpirun and smpicc to OS X
878 * Kill SD_link_get_properties: hard to maintain and makes very little sense
879 Shout out if you used it.
882 * Display the list of still queued messages in SG mode when existing
886 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
887 * Add xbt_graph_node_get_outedges() [Silas De Munck]
888 * Add xbt_str_from_file(FILE*)
889 * Add xbt_dict_get_key achieving a linear reverse search
890 * Remove the context module
892 Portability report of this version:
893 * Main portability targets:
894 - Linux(debian)/x86/context
895 - Linux(debian)/x86/pthreads
896 - Linux(debian)/amd64/context
897 - Linux(debian)/amd64/pthreads
898 On these, we still have the eratic breakages of gras/pmm and
899 amok/saturate_sg reported in previous version. We still think
900 that the tests are the cause of the fault, not the tested code.
902 - Mac OSX Leopard/x86/context
903 Still false negative in tesh autotesting.
904 Smpi still fails, but this time because readlink does not accept -f
905 Everything seems to work properly beside of that.
908 - AIX version 5.3 (only tested contexts this time)
909 Smpi still fails there because mktemp is not installed.
910 Everything seems to work properly beside of that.
912 I managed to compile it for the first time, but several breakages.
913 Won't delay the release for this exotic platform.
915 * Windows: it's still lagging behind. If you want to help, please
918 Timing report of this version:
919 This version seem to be more than 5% faster than 3.3.1 (on linux
920 64bits with contextes). The gain is less than expected, we are
921 investigating this for next release.
923 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
925 SimGrid (3.3.1) stable; urgency=low
928 * Implement a --cfg-help to show existing configuration variables
929 * Build chain do not require doxygen in maintainer mode
932 * fix a bug on struct sizeof computation, which prevented the
933 exchange of arrays of structs in some conditions
934 - added a regression test about this in datadesc_usage
935 * Allow the exchange of 0-long dynamic vectors.
936 - for that, use -1 as indicator of dynamic size instead of 0
937 - This implied to change any size from unsigned long to long,
938 reducing a bit communication abilities, but I guess that with
939 64bits being quite common, this is more than enough.
940 - This also induce a protocol change, thus bumping network protocol
941 version from 0 to 1 (if we have external users, we have to get
942 clean on that point too ;)
943 - added two regression tests about this in datadesc_usage
944 * Be more verbose when propagating local exceptions
945 This helps debugging.
946 * Display the status of simulated processes when receiving SIGINT in
950 * Allow to control the simulation from a trace file.
951 New functions MSG_action_register() and MSG_action_trace_run()
952 The first one allows to associate a function execution to each
953 kind of action while the second one parses a trace file and
954 triggers the corresponding actions within the system.
955 For now, only a toy example is provided in examples/msg/actions
956 * Add an exemple of process migration in examples/msg/migration
957 * Fix a bug in task exchange which broke MSG_task_get_sender()
958 Add a teshsuite regression test for that.
959 [Bug: if MSG_task_get_sender() is called after sender exit,
961 * Fix a bug which prevented suspend/resume to work properly
962 * Display the status of simulated processes when receiving SIGINT
963 This fixes a regression of v3.3. due to the introduction of SIMIX
964 * Bug fixing in failure management:
965 - trace could not start by a failure at time 0
966 - failure during communications were not working
969 * Add SIMIX_process_set_name() to change the name of the current
970 process in the log messages.
971 * Store smx_hosts in a dict since we only retrieve them by name
972 * Move the configuration infrastructure to surf
975 * Move the configuration infrastructure to surf
978 * Massive internal cleanups:
979 - Store internal structures on processes instead of hosts (allows
980 to have more than one process per host, in addition of being more
982 - Cleanup the initialization/finalization process
983 - Kill a whole bunch of unneeded synchronization:
984 processes run in exclusive manner within the simulator
985 - Move queues from global tables to process data fields
987 - now accept -platform and -hostfile arguments
988 - Pass the right rank value to processes according to the hostfile
989 * Compile the examples by default, and use them as regression tests
990 * Implement MPI_Wtime()
991 * Change the reference speed to a command line option
994 * TCP_gamma can now be specified as command line option using
995 --cfg=TCP_gamma:10000000.0
996 * Change the --surf-path cmd line option into --cfg=path:
999 * Also include strbuff from xbt.h public header
1000 * xbt_ex_display(): do not free the exception after displaying
1001 This allows to do more with the given exception afterward.
1002 Users should call xbt_ex_free() themselves.
1006 Portability report of this version:
1007 * Main portability targets:
1008 - Linux(debian)/x86/context
1009 - Linux(debian)/x86/pthreads
1010 - Linux(debian)/amd64/context
1011 - Linux(debian)/amd64/pthreads
1012 These targets fail about 1/10 of times on gras/pmm, but we believe
1013 that this is because of the test, not because of simgrid.
1014 amok/saturate_sg fails even more rarely, and the test may not be
1017 - Mac OSX Leopard/x86/context
1018 The test suite still spits tons of errors because some obscure
1019 force prevents us from removing the temporary directories
1020 arguing that they still contain some metadata I've never heard of.
1021 Smpi fails because seq is not installed.
1022 Everything seems to work properly beside of that.
1025 - AIX version 5.3 (both contexts and pthread)
1026 Smpi still fails there because mktemp is not installed.
1027 XML inclusions seems rosty on AIX.
1029 * Windows: it's still lagging behind. If you want to help, please
1032 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
1034 SimGrid (3.3) stable; urgency=high
1038 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
1039 [Malek Cherier & Mt]
1041 * Introduce the SIMIX module: factorize code between MSG and GRAS.
1044 Until now, GRAS were using MSG as an interface to SURF. It was
1045 quite difficult because both interface have several differences
1046 (MSG channels vs GRAS sockets were the most notable point).
1048 This also opens the gate to SMPI (which should occur soon) and speed
1049 up simulations by to 40% (even if it were not the main goal).
1051 **************************************
1052 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
1053 **************************************
1054 The point is that events occuring at the exact same timestamp are
1055 not scheduled in the same order with the old and new version. This
1056 may be enough to completely change the execution of simulations in
1057 some cases. Sorry for the inconvenience.
1059 * Cleanup and upgrade the XML format to push further scalability
1060 issues (check http://hal.inria.fr/inria-00256883/ for more info)
1062 * Improve the testing infrastructure with tesh. Now a very large part of
1063 the code is tested not only by being run but also by checking that the
1064 output match an expected output [Mt].
1066 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
1067 is really less memory-demanding, which should allow you to use
1068 larger files in SimGrid [AL].
1070 * Inform valgrind about our contextes, so that it becomes usable
1071 with the default (and more effecient) version of SimGrid
1072 [contributed by Sékou Diakite, many thanks]
1075 * Introduce a listener thread in charge of receiving incomming
1076 messages from the network. It allows to overlap communication and
1077 computation but most notably, it removes some stupid deadlocks due
1078 to the fact that so far, a process could not send and receive at
1079 the same time. This made most non trivial communication schema
1081 * Convert the PIDs from long int to int to match the MSG ones (and
1082 linux ones too) [Mt]
1083 * New function: gras_agent_spawn() to launch a new process on
1084 current host. Only working in simulation for now. [Mt]
1085 * New function: gras_os_hostport() returning a constant form (ie,
1086 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
1089 * Make the backtrace of exceptions more human readable [Mt]
1090 * New module: xbt/str [Mt]
1091 a ton of string utility functions (split, join, printf to a newly
1092 allocated buffer, trim, etc)
1093 * New module: xbt/hash [Mt]
1094 SHA1 hashing algorithm (more to come if needed)
1095 * New module: xbt/synchro [Mt]
1096 synchronization tools (mutex and conditions) working the same way
1097 in simulation and in real life (mainly useful for GRAS, but not
1099 * New module: xbt/queue [Mt]
1100 classical producer/consumer synchronization scheme
1101 * xbt_dynar_new_sync() creates a synchronized dynar. All access
1102 (using the classical functions will get serialized) [Mt]
1103 * Make dictionary internal table dynamic. No need to specify its size
1104 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
1106 * Make sure the log channels are organized as a tree under windows
1107 (because of ANSI C compatibility issue, any channel were child of
1108 root directly) [Mt].
1111 * Cleaned many thing in surf and fixed a few bugs [AL].
1112 * Add a nice command line configuration mechanism to compose models [AL].
1113 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
1114 the previous one (KCCFLN05). It relies on something that looks like
1115 a max-min sharing mechanism but cannot be written as such. A new solver
1116 was thus designed [AL].
1117 * Add a new solver to lmm. Based on Lagrange optimization and
1118 gradient-based descent, it enables to efficiently maximise systems s.a
1120 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
1122 This solver enables to propose two new network models for TCP Reno and
1123 TCP Vegas based on Low's work. These models still need to be fully
1124 tested though [Pedro Velho].
1127 * Bug fix in SD_simulate. Now the time bound given as argument is
1129 * Use the new parallel task model (ptask_L07) as default.
1130 * Use the SURF command line configuration mechanism.
1131 * 0-size tasks (for synchronization) should now work.
1133 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
1135 SimGrid (3.2) stable; urgency=high
1139 We still experience issues on this platform, but we believe that at
1140 least MSG is usable.
1142 GRAS API BREAKAGE (for simplification purpose, sorry):
1143 * the gras_msgtype_by_name is not used anymore. Instead of
1144 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
1145 you can write (and must)
1146 gras_msg_send(toserver, "request", &request);
1147 - If you still want to pass a gras_msgtype_t to the function (to cache
1148 the type and avoid the lookup time), use the gras_msg_send_() variant.
1149 - Impacted functions:
1150 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
1151 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
1152 * The callbacks are now expected to return 0 when everything went well
1153 (just like the main() function)
1155 GRAS new features and improvements:
1156 * New module mecanism where user code can use per process globals [Mt]
1157 This is similar to gras_userdata_*() functions, but for libraries. It
1158 factorize some code developped over and over in the examples and AMOK.
1159 It has still to be documented and used (only amok/peermanagement is
1161 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
1162 loss when they were small enough to fit into the buffer and sent quickly
1163 enough so that they can all get received in one shoot.
1164 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
1165 if not found. Use the *_or_null() variant for the old semantic.
1166 * In gras_msg_handle, do not discard messages without callback.
1167 They are probably messages to be explicitly awaited later (ie, proofs of
1168 mis-synchronization in userland since they are sent before being awaited)
1169 No big deal usually.
1170 * gras_socket_meas_send/recv: semantic changed!
1171 The numerical arguments used to be (1) the total amount of data to send
1172 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
1173 messages. This was need for the fool willing to send more than MAXINT
1174 bytes on quite fat pipes.
1177 * Do really rename the hostmanagement module to peermanagement. [Mt]
1178 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
1179 but this is rather new and this was documented in the module
1180 documentation (poor excuses, I admit)
1181 * Bandwidth measurement semantic changed! This follows the changes to
1182 gras_socket_meas_send/recv explained above.
1185 * A sequential mode has been added to the workstations. When a workstation
1186 is in sequential mode, it can execute only one task, and the other tasks
1187 are waiting in a FIFO. [Christophe Thiery]
1190 * The KCCFLN05 workstation model now handles parallel tasks. It is the
1191 model for SIMDAG. [Christophe Thiery]
1192 * Bug fix in the maxmin solver: Some values were close to 0 instead of
1193 equal to 0, which caused some bad behaviors in
1194 saturated_constraint_set_update. I now use a threshold mechanism like in
1198 * When running manually src/testall, you select specific units [Mt]
1199 testall is the result of our cunit mecanism, and should replace all
1200 the scripty thingy around since bash don't run easily on billware.
1202 * A mallocator system has been added. [Christophe Thiery]
1203 Mallocators allow you to recycle your unused objects instead of freeing them
1204 and allocating new ones.
1206 Documentation update:
1207 * FAQ reworking + New FAQs:
1208 - "Valgrind spits tons of errors!" [Mt]
1209 - "How to repport bugs" [Mt]
1210 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
1211 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
1212 - Communication time measurement within MSG [AL]
1213 - I experience weird communication times when I change the latency [AL]
1214 * GRAS tutorial [Mt]
1216 - an introduction to the framework and to the used communication model
1217 - an initiatic tour introducing the most proheminent features:
1219 . Lesson 0: Installing GRAS
1220 . Lesson 1: Setting up your own project
1221 o Part 2: Message passing
1222 . Lesson 2: Exchanging simple messages
1223 . Lesson 3: Passing arguments to the processes (in SG)
1224 . Lesson 4: Attaching callbacks to messages
1225 . Lesson 5: Using globals in processes
1226 . Lesson 6: Logging informations properly
1227 . Lesson 7: Using internal timers
1228 . Lesson 8: Handling errors through exceptions
1229 . Lesson 9: Exchanging simple data
1230 . Lesson 10: Remote Procedure Calling (RPC)
1231 . Lesson 11: Explicitely waiting for messages
1232 . Recapping of message passing features in GRAS
1233 - A HOWTO section containing:
1234 o HOWTO design a GRAS application
1235 More are due, of course. They will come latter. In the meanwhile, you can
1236 check the examples which are still here.
1238 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
1240 SimGrid (3.1) stable; urgency=high
1244 There was a stack corruption somewhere, visible only when optimizing
1245 with these versions. [Vince]
1248 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1249 module that enabled to play with Directed Acyclic Graphs. It is built
1250 directly on top of SURF and provides an API rather close to the old
1251 SG. Some old codes using SG are currently under rewrite to check that
1252 all needful functions are provided. [Christophe Thiery]
1255 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1256 extension of the classical CLM03 model that gracefully handles
1257 failures. This is now the default model for MSG and GRAS. It doesn't
1258 handle parallel tasks yet though. [AL]
1259 * Bug fix: Weights were not correctly set in the network part.
1260 WARNING: This may have resulted in incorrect results with simulations
1261 where there are more than one flow on a given link. [AL]
1264 * After a (long ?) discussion on simgrid-devel, we have decided that the
1265 convention we had on units was stupid. That is why it has been decided
1266 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1267 WARNING : This means that all previous platform files will not work as
1268 such with this version! A warning is issued to ask users to update
1270 A conversion script can be found in the contrib module of the CVS, under
1271 the name contrib/platform_generation/surfxml_update.pl [MQ]
1274 * Bug fix: Processes were started in reverse order, wrt deployment file.
1275 WARNING: if your code relies on this bug, please fix it. [AL]
1276 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1277 being executed on two different locations. [AL]
1278 * Bug fix: Failures are now better supported thanks to Derrick's tests
1279 (there was many failure situations I hadn't thought of and that weren't
1280 correctly handled). [AL]
1281 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1285 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1286 Exception occurring on server-side are propagated back to client (!).
1288 API CHANGE: the callback changed their prototype. Change:
1289 int my_handler(gras_socket_t expeditor, void *payload_data) {
1291 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1292 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1294 * New! function: gras_msg_handleall to deal with all messages arriving
1295 within a given period.
1296 * New! function: gras_socket_server_range to get a server socket in a
1297 range of port numbers (ease to avoid port number conflicts) [MQ]
1298 * New! gras processes display their backtrace when they get a SIGUSR1
1299 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1300 Sweet to debug RL processes [MQ]
1304 - Do not force experiment sizes to be expressed in kb, or it becomes
1305 impossible to measure the latency this way (needs one byte-long tests)
1306 WARNING: this changes the amok_bw_* function semantic. [MQ]
1307 - Implements the link saturation stuff. [MQ]
1308 * Peer management module:
1309 New! module factorizing code that we wrote over and over [MQ].
1312 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1313 - Test units are placed directly into the library code, they get extracted
1314 automatically and placed into the src/testall binary.
1315 - Convert most of the XBT tests to this system.
1316 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1317 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1318 live in the same C code. This required to change a little bit the API
1319 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1320 * New module: sparse graph structure with basic algorithms (this is work
1321 in progress and the API is not considered to be frozen yet). [AL]
1322 * Display more information on backtraces: source line & function names are
1323 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1324 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1325 * Reworked a little bit some #include statements to load only required
1326 headers. Some user code that relied on SimGrid to include stdlib or
1327 stdio may need to include it by themselves. [AL]
1328 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1330 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1331 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1332 but I guess that almost nobody used those parts. [MQ]
1334 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1336 SimGrid (3.0.1) stable; urgency=low
1339 * Unfortunately, I had missed 5 misnamed functions:
1340 xbt_fifo_item_t xbt_fifo_newitem(void);
1341 void xbt_fifo_freeitem(xbt_fifo_item_t);
1342 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1343 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1344 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1345 They're now deprecated. Please use their new versions:
1346 xbt_fifo_item_t xbt_fifo_new_item(void);
1347 void xbt_fifo_free_item(xbt_fifo_item_t);
1348 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1349 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1350 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1352 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1353 * Documentation: xbt_log module unmercifully reworked [MQ]
1354 * Bugfix: there was a problem with the ending of contexts with
1355 the pthread backend. It caused some weird deadlock or behavior
1356 depending on the pthread implementation. [AL]
1357 * Bugfix: get the exceptions raised in the simulator repport where
1358 and why they come from when they are not catched in time [AL, MQ]
1361 * Bugfix: Do repport the error when two non-connected hosts try to
1362 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1365 * Add additionnal checkings on communications. Assert that two
1366 communicating hosts are connected by a set of links... [AL]
1369 * Add additionnal checkings on channel values in communication [AL]
1370 * New: MSG_task_get_source to see on which host a task was generated [HC]
1371 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1372 the number of tasks waiting to be received on channel and sent
1374 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1375 waits for the first task coming from a given host.. [AL]
1377 GRAS new functionnalities: [MQ]
1378 * Enhance the parsing macro to allow the size of multidimentional objects
1379 to be given thru annotations.
1380 * New example (and documentation): Matrix Multiplication a la RPC
1381 (as when I was young!) and fix a bunch of bugs found on the way.
1383 GRAS performance improvements: [MQ]
1385 * Reduce the amount of cbps creation/destruction by making it static to
1386 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1389 * Change libdata to a set so that we can search for stuff by ID (and thus
1390 reduce the insane amount of dict lookups)
1393 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1394 RL communication on top of SG since it's so uneffective.
1395 It may also be used for inter-thread communication in RL, one day.
1396 * Use gras_datadesc_copy() to exchange messages on top of SG
1398 - improve message exchange performance on top of SG
1399 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1400 * Don't exchange on the network the size of the used part of buffer,
1401 instead, specify the possible buffer size to read().
1403 - reduces the amount of read/write calls (one pair per exchange)
1404 - reduces the amount of exchanged data (the size)
1405 - allows to retrieve all arrived data on receiver side, if we don't need
1406 it right now (subsequent read will peek the buffer)
1407 - allows the receiver to proceed with the begining of the stream before
1408 everything is arrived
1409 - make it possible to build an iov transport (using readv/writev)
1411 - take care of the data with non-stable storage (like stacked data),
1413 * If possible, TCP send uses vector I/O (when writev() is here)
1414 - Don't use it for receive since we send data sizes and data on the
1415 same stream, so we wouldn't be able to chain large amount of chunks
1416 before having to flush the stuff to read the size.
1417 * Rework the transport plugin mecanism to simplify it and reduce the
1418 amount of pointer dereferencement when searching for the right function
1421 * I guess that now, we do almost as few system calls as possible while
1422 doing as few data copy as possible.
1424 To improve it further, we could try to send all the sizes first and then
1425 all the data (to use iov on receiving size), but it's only a partial
1426 solution: when you have 2 dimensional data, the sizes of the second
1427 dimension is data of the first dimension, so you need 3 streams.
1429 I'm not sure the potential performance gains justify the coding burden.
1431 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1433 SimGrid (3.00) stable; urgency=high
1436 * New! Give the possibility to hijack the surf parser and thus bypass
1437 MSG_create_environment and MSG_launch_application. Have a look at
1438 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1440 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1442 SimGrid (2.96) unstable; urgency=low
1447 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1448 This deprecates the xbt_error_t mecanisms.
1449 It modifies (simplifies) all XBT and GRAS API.
1450 MSG API keeps unchanged (exceptions raised by XBT are catched from
1451 within MSG and masked with existing error handling facilities)
1454 * New! Add a FATPIPE model. [AL]
1455 * New! Add a parallel task model. [AL]
1456 * New! Add automatically a loopback interface (in the default
1457 network model) if none was precised.
1460 * Bugfix: MSG_process_resume now works with the current running process.
1462 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1463 * Modification of MSG_task_get_compute_duration. Once a task has been
1464 processed, the value returned by this function is now equal to 0. [AL]
1465 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1466 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1467 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1468 of a m_task. For now, MSG_task_cancel only works with computation
1470 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1471 of the processor (in Mflop/s) regardless of the current load on the
1473 * API Change: use proper naming convention for MSG_getClock and
1474 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1476 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1477 This function changes the priority of a computation task. This priority
1478 doesn't affect the transfer rate. A priority of 2 will make a task
1479 receive two times more cpu power than the other ones. This function
1480 has been added to suit the needs of Nguyen The Loc and hasn't been that
1481 much tested yet. So if it fails, please report it and send me your code.
1483 * API Change: removed all functions and types that were marked "deprecated"
1484 since many months. Renamed MSG_global_init_args to MSG_global_init.
1486 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1488 SimGrid (2.95) unstable; urgency=low
1491 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1492 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1493 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1496 * Bugfix: Allow absolute paths for platform description files [MQ]
1497 * Bugfix: do free the variables after use. Leads to drastic performance
1499 * Implement max_duration (ie, timeouts) on resources [AL]
1502 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1504 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1505 the needed support (timeouts) [AL]
1508 * Implement measurement sockets. You can now get the bandwidth between two
1509 hosts thanks to AMOK (see below). [MQ]
1510 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1511 dynar over the network (yeah) [MQ]
1512 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1513 * simple/static token-ring example. [Alexandre Colucci and MQ]
1514 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1515 in place before [MQ]
1518 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1519 type parsing. "short" and "long" modifiers were also ignored; other
1520 modifier (such as reference level) are still ignored. [MQ]
1521 * Embeed the buffer size within the buffer itself on SG. [MQ]
1522 That way, send() are atomic and cannot get intermixed anymore (at least
1523 the ones which are less than 100k; bigger messages still have the issue)
1524 * Array size pushed by the field, not by the field type (or each
1525 and every long int will push stuff to the cbps) [MQ]
1526 * use select() to sleep since it allows to portably sleep less than one
1529 GRAS (minor cleanups)
1530 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1533 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1534 * type category 'ignored' killed as it was never used and were difficult
1536 * whether a type can cycle or not is now a flag, leaving room for more
1537 flags (as "ignored", if we feel the need one day ;)
1538 * Rename raw sockets to measurement sockets since "raw" has another
1539 meaning in networking community.
1542 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1543 and offers features not belonging to GRAS but that most applications
1544 need. One day, it may be a set of plugins loadable at runtime.
1545 * New module: bandwidth
1546 bandwidth measurement between arbitrary nodes running this module. [MQ]
1548 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1550 SimGrid (2.94) unstable; urgency=low
1552 The first beta release of SimGrid 3 !
1556 * Update the main page and the FAQ. Adding references to gforge.
1559 * Add a gras_os_getpid function.
1562 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1563 * Extend the logs so that they also print PID, hostname, date, ... if
1565 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1566 and kill the old version which were in testsuite/
1567 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1571 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1572 for users. Instead, each programming environment (GRAS, MSG,...) use it
1573 when needed to provide such a feature to users.
1574 Don't shortcut the mecanism or you will also shortcut the virtualization
1575 you need on the simulator.
1579 * Cleanups in configury with regard to compile optimization/warning flags.
1580 Also add -fno-loop-optimize to any powerpc since it's the optimization
1581 killing gcc (< 3.4.0).
1582 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1583 needed by MSG examples complications
1584 * Borrow configury beautifications from PHP
1587 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1588 warning (thanks loris for stumbling into this one).
1589 * Bugfix: stop loading private headers (gras_config.h) from the public
1593 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1594 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1595 RL would imply a DNS resolver, which is *hard* to do in a portable way
1596 (and therefore delayed).
1597 * Implement a real timer mecanism and use it in timing macros. This allows
1598 to avoid rounding errors and get a 0.000005 sec precision in timing
1599 macros. While I was at it, various cleanups:
1600 - allow to declare more than one timed section per file (fix a stupid bug)
1601 - move some private declaration to the right place
1602 - merge conditional execution and timing macros into emulation module
1603 - document the module
1604 - make sure the module cleanups its mess on gras_exit
1605 * Documentation improvements:
1606 - (new) how to compile applications using GRAS
1607 - (new) emulation support (timing macros)
1609 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1611 SimGrid (2.93) unstable; urgency=low
1613 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1616 - Use Paje properly where used. Still to be sanitized properly.
1617 - Portability fix: Add an implementation of the contexts using pthread
1621 - Add xbt_procname(): returns the name of the current process.
1622 Use it to show the current process's name in all logging.
1624 - fix detection of older flex version and the reaction, since we do
1625 depend on modern ones (we use lex_destroy)
1626 - Better separation of SG and RL in the libs: remove all simulation code
1627 from libgras. As a result, this lib is now only 200k when stripped.
1628 Some of the xbt modules may also be duplicated (two sets and such) and
1629 should be cleaned/killed before SG3.
1630 - Insist on using xlC on AIX because of weird problems involving gcc there.
1631 - Cleanup the make remote stuff. This is now done by scripts
1632 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1633 mainly for our private use, but we're working on changing them to user
1636 - Bugfix: flush the socket on close only if there is some *output*.
1637 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1639 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1642 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1644 SimGrid (2.92) unstable; urgency=low
1646 Alpha 3 on the path to SimGrid 3
1650 - New! First try of benchmarking macros.
1651 - New! First try so that gras_stub_generator generate deployment and
1652 remote compilation helpers.
1654 - Bugfix: Initialization fix in msg_test.
1658 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1660 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1662 - API Change: gras_os_sleep to take the amount of seconds as a double.
1663 Accepting an int was error prone since it was the only location where
1664 seconds were coded as such. It leaded to damn rounding errors.
1665 - Bugfix: Hard to belive that timers ever worked before this.
1667 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1669 SimGrid (2.91) unstable; urgency=low
1671 Alpha 2 on the path to SimGrid 3
1675 - Bug fix in the lmm_solver.
1677 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1678 through the function MSG_paje_output.
1679 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1680 - It is possible to bound the rate of a communication in MSG with
1681 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1682 to write it in the changelog).
1683 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1688 - Various cleanups to the autotools stuff
1689 - Begin to move Gras examples to examples/gras/
1690 - Let make distcheck work again (yeah!)
1692 - documentation overhauled using doxygen.
1693 gtk-doc-tools is dead in SimGrid now.
1694 - Automatically extract all existing logging categories, and add the list
1695 to the documentation (long standing one, to say the less)
1697 - Cleanup the known architecture table. Reorder the entries to group what
1698 should be, and use a more consistent naming scheme.
1699 (some of the test dataset are still to be regenerated)
1700 - New! Allow library to register globals on each process just as userdata
1702 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1703 lookup time (for now).
1704 Use it in msg and trp.
1705 This cleans a lot the internals and helps enforcing privacy of the
1706 headers between the gras components.
1707 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1708 - Bugfix: gras_os_time was delirious in RL.
1709 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1711 - Reenable GRAS now that it works.
1713 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1715 SimGrid (2.90) unstable; urgency=low
1717 Alpha 1 on the path to SimGrid 3
1719 * It is a long time since the last release of SimGrid. I'm sorry about
1720 that but as I had told you, I was rewriting a lot of things. I apologize
1721 to those who had been reporting bugs to me and that I had not answered.
1722 If your bug is still in the new version, please tell me. Here is a
1723 summary of the main changes.
1725 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1726 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1727 portability, speed, and a lot more but you'll figure it out later.
1728 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1729 I wanted to release the new MSG as soon as possible and I have
1730 broken GRAS, which is the reason why, for now, only MSG is fully
1731 functional. A laconic description of these projects is available
1732 in the documentation.
1734 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1735 kernel optimized for our needs (called SURF but only the developers
1736 should use it). Hence, MSG is now roughly 30 times faster and I think
1737 that by rewriting a little bit MSG, I could event speed it up a little
1738 bit more. Beside the gain in speed, it is also much easier to encode a
1739 new platform model with SURF than it was with SG. More to come...
1741 * REVOLUTION 3: I have tried to change a little as possible the API of
1742 MSG but a few things really had to disappear. The main differences
1743 with the previous version are :
1744 1) no more m_links_t and the corresponding functions. Platforms are
1745 directly read from a XML description and cannot be hard-coded
1746 anymore. The same format is used for application deployment
1747 description. The new format is described in the documentation.
1748 Have a look in tools/platform_generation. There is a tiny script
1749 that converts from the old platform format to the new one. Concerning
1750 the application deployment format, parsing the old one is tricky.
1751 I think most of you should however be able to convert your files. If
1752 it is really an issue, I can write a C code that does the conversion.
1754 2) the toolbox tbx does not exist anymore. We now have a library
1755 with much more data-structures but without the hash-tables (we have
1756 dictionaries that are much faster).
1758 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1760 *****************************************************************************
1761 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1762 * don't feel like converting the oldies. (Mt) *
1763 *****************************************************************************
1766 Version 2.90: "the long awaited one"
1767 - Finished rewriting and debugging MSG. Rewrote the documentation.
1768 - disable GRAS for now since it needs to be ported to the newest SG
1771 - Finish the port to windows (using mingw32 for cross-compile)
1774 - Main loop and datastructures of SURF. A cpu resource object is
1775 functional. Surf can thus be used to create cpu's with variable
1776 performance on which you can execute some actions.
1778 2004-11-15 Martin Quinson
1779 - Port to ARM. Simply added the alignment and size descriptions. Should
1780 work, but the ARM machines are so slow that I didn't had the opportunity
1781 to 'make check' over there yet.
1783 2004-11-15 Arnaud Legrand
1784 - Trace manager now written. It uses a heap structure and is therefore
1785 expected to be efficient. It may however be speeded up (particularly
1786 when many events occur at the same date) by using red and black
1787 trees. One day maybe...
1788 - Max-min linear system solver written. It uses a sparse matrix
1789 structure taking advantage of its expected use. Most operations are
1790 O(1) and free/calloc are called as few as possible. The computation of
1791 the minimum could however be improved by using a red and black tree
1794 2004-11-03 Arnaud Legrand
1795 - Rename every gras_* function that was in xbt/ to its xbt_
1797 - Add a heap and a doubly-linked list to xbt
1798 - Added a dichotomy to the dictionaries. make check works as well before
1799 so I assume that the patch is correct. I do not know however if things
1800 run effectively faster than before now. :)
1802 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1803 SimGrid, and the version number is bumped to 2.x
1805 2004-10-29 Martin Quinson
1806 - Introduction of the remote errors.
1807 They are the result of a RMI/RPC on the remote machine.
1808 ErrCodes being scalar values, you can't get the host on which those
1809 errors did happen. Extending the error mechanism as in Gnome is possible.
1810 No idea yet whether it is a good idea.
1812 2004-10-28 Martin Quinson
1813 - Interface revolution: the Starred Structure Eradication.
1814 I used to do typedef struct {} toto_t; and then handle *toto_t.
1815 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1816 - ???_t is a valid type (builded with typedef)
1817 - s_toto_t is a structure (access to fields with .)
1818 - s_toto is a structure needing 'struct' keyword to be used
1819 - e_toto_t is an enum
1820 - toto_t is an 'object' (struct*)
1822 typedef struct s_toto {} s_toto_t, *toto_t;
1823 typedef enum {} e_toto_t;
1824 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1825 s_toto_t) is private.
1827 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1828 so that it can set the variable to NULL. It was so for dicts and sets,
1829 it changed for dynars.
1831 - Fix a bunch of memleaks in dict_remove
1832 - Fix a bug in sg/server_socket opening: it failed all the time.
1834 2004-10-07 Martin Quinson
1835 - Speed up dynar lookup operation a bit.
1837 gras_dynar_get is dead.
1839 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1840 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1841 (which gives you the address of the stored data).
1843 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1844 copy of the data using an affectation to do the job and not a memcpy.
1846 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1847 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1849 It does not really speedup the dynar test because they are
1850 setting elements all the time (and look them seldom). But the dict does
1851 far more lookup than setting.
1853 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1855 2004-10-05 Martin Quinson
1856 - Allow to (en/dis)able the cycle detection at run time.
1858 Whether we should check for cycle or not is now a property of each
1859 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1860 datadesc_cycle_unset allow to remove this property when previously set.
1862 Note that the cycle detection is off by default since it impacts the
1863 performance. Watch the data you feed GRAS with ;)
1865 This property is hereditary. Any element embedded in a structure having it
1866 set have it set for the time of this data exchange.
1868 You should set it both on sender and receiver side. If you don't set it on
1869 sender side, it will enter an endless loop. If you forget on receiver
1870 side, the cycles won't be recreated after communication.
1872 - Header reorganization.
1873 Kill gras_private.h, each submodule must load the headers it needs.
1875 2004-10-04 Martin Quinson
1876 - Interface revolution: do not try to survive to malloc failure.
1878 Now, gras_malloc and friends call gras_abort() on failure.
1879 As a conclusion, malloc_error is not a valid error anymore, and all
1880 functions for which it was the only gras_error_t return value are
1881 changed. They now return void, or there result directly.
1882 This simplify the API a lot.
1884 2004-09-29 Martin Quinson
1885 - Re-enable raw sockets.
1886 Created by gras_socket_{client,server}_ext;
1887 Used with gras_raw_{send,recv}
1890 It should allow to kill the last bits of gras first version soon.
1892 This is not completely satisfactory yet (duplicate code with
1893 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1896 - Simplify transport plugin (internal) interface by not passing any
1897 argument to _server and _client, but embedding them in the socket
1900 2004-09-28 Martin Quinson
1901 - Finish the port to AIX.
1902 autoconf was my problem (segfault within the malloc replacement
1903 function. No idea why)
1905 2004-09-16 Martin Quinson
1906 - Fix some size_t madness on 64bit architectures.
1908 2004-09-08 Martin Quinson
1909 - Reduce the number of system headers loaded, overload some more system
1910 calls (such as malloc to cast the result of the system one, and work
1912 - Fix and reintroduce the config support
1914 2004-09-07 Martin Quinson
1915 - Source code reorganization to allow Arnaud to surf all over there.
1916 - Allow to document the logging categories.
1917 - Remove all uppercase from logging categories and useless cleanup in names.
1919 2004-08-18 Martin Quinson
1920 Version 0.6.2 (protocol not changed; API changed)
1921 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1922 gras_error_t), and NULL when not found. Functions expecting a msgtype
1923 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1924 hopefully usefull message.
1925 - Portability to prehistoric sparcs again
1927 2004-08-17 Martin Quinson
1928 Version 0.6.1 (protocol not changed; ABI not changed)
1929 - prealloc some buffers to speed things up
1931 2004-08-11 Martin Quinson
1932 Version 0.6 (protocol not changed; ABI expended)
1933 - The parsing macro can deal with the references, provided that you add
1934 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1936 2004-08-09 Martin Quinson
1937 Version 0.5 (protocol not changed; ABI changed)
1938 - Allow to off turn the cycle detection code in data exchange at
1939 compilation time. It should be at run time, but I'm short of time (and
1940 the config stuff is still broken). That way, we keep dict out of the
1941 critical path, which is good because the performance is poor:
1942 - search not dichotomial yet
1943 - dynar give no way to access their content and memcpy everytime
1944 - In composed data description (struct, ref and so on), stop foolness of
1945 keeping the subtype's ID, but store the type itself. This keeps sets out
1946 of the critical path, which is good since they rely on dynar and
1947 dictionnaries. The only loose of that is that we cannot detect the
1948 redeclaration of a structure/union with another content (but I'm not sure
1949 the code detected well this error before anyway). We still can detect
1950 the redefinition discrepancy for the other types.
1951 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1952 it breaks the code because of type-punning used all over the place).
1953 This breaks on all non-gcc architectures (for now).
1955 All those changes (plus the buffer of last time) allow me to gain 2 order
1956 of magnitude on cruel tests consisting of 800000 array of integers on two
1957 level of a hierarchical structure (200 secondes -> 4 secondes)
1960 - the selector of reference must now return the type it points to, not
1961 the ID of this type.
1963 2004-08-06 Martin Quinson
1964 Version 0.4 (protocol changed; ABI not changed)
1965 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1966 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1967 - Implement a buffer transport, and use it by default (it relies on tcp in
1968 real life and on sg in simulation).
1969 That's a bit hackish since I had a new field to the structure to store
1970 its data without interfering with the subtype ones. Inheritance
1971 is tricky in C. And that's a kind of reverse inheritance with one class
1972 derivating two classes. Or maybe a game with java interfaces. Anyway,
1973 that's damn hard in C (at least).
1974 Moreover, I got tired while trying to ensure plugin separation and
1975 genericity in SG mode. MSG wants me to do weird things, so let's go for
1976 cruel hacks (temporarily of course ;).
1977 See comment in transport_private.h:71
1978 - do not include all the _interface headers in private but in the files
1979 which really need them (to cut the compilation time when they are
1982 2004-07-26 Martin Quinson
1983 Version 0.3 (protocol not changed; ABI changed)
1984 - Major overhault of the datadesc interface to simplify it:
1985 - shorted the function names:
1986 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1987 - add a trivial way to push/pop integers into the cbps without malloc.
1988 This allows to make really generic sub_type description, which simply
1989 pop their size of the stack.
1990 - add a function gras_datadesc_ref_pop_arr() which does what users want
1991 most of the time: Declare a dynamic array (which pops its size of the
1992 stack) and declare a reference to it. Poor name, but anyway.
1993 - kill the post-send callback, add a post-receive one
1995 2004-07-23 Martin Quinson
1996 Version 0.2 (protocol changed; ABI changed)
1997 - add some testing for cpbs in the test cases, and fix some more bugs.
1998 This invalidate again the little64 data file, since I cannot regenerate
2000 - remove an awfull optimization in the logging stuff, allowing me to:
2001 - understand it again
2002 - learn gcc how to check that the argument match the provided format
2003 - fix all errors revealed by gcc after that
2004 - internal keys of dict are not \0 terminated. Deal with it properly in
2005 loggings instead of segfaulting when the user want to see the logs :-/
2007 2004-07-22 Martin Quinson
2008 - Fix some stupid bug preventing cbps (callback postit) from working
2010 2004-07-21 Martin Quinson
2011 - Some documentation cleanups
2012 - remove the useless last argument of msgtype_declare
2013 - rename the Virtu functions to fit into the 'os' namespace
2014 - move headers src/include -> src/include/gras/ and stop fooling with
2015 gras -> . symbolic link
2016 - make distcheck is now successful
2018 2004-07-19 Martin Quinson
2020 - Build shared library also
2021 - Install html doc to the right location
2022 - stop removing maintainer files in make clean
2023 - build tests only on make check
2025 2004-07-13 Martin Quinson
2027 - No major issue in previous version => change versionning schema
2028 - Re-enable little64 convertion test now that Abdou kindly regenerated the
2029 corresponding dataset.
2031 2004-07-11 Martin Quinson
2033 - Get it working with any kind of structure (we can compute the padding
2034 bytes remotely for all the architectures I have access to)
2035 - Implement the structure parsing macro (still not quite robust/complete)
2036 - Improvement to the remote testing toysuite
2038 2004-07-10 Martin Quinson
2039 [autoconf mechanism]
2040 - get ride of a bunch of deprecated macros
2041 - actually run the test for two-compliment, not only compile it :-/
2042 - test whether the structures get packed (and bail out if yes. Damn.
2043 Alignment is a serious matter)
2044 - test whether the structures get compacted (but respecting the alignment
2045 constraints of each types)
2046 - test whether the array fields of structures can straddle alignment boundaries
2048 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
2049 (took me 2 hours to find that bug, looking at the wrong place)
2050 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
2051 before sending/receiving and are used to compute the offsets of fields
2052 - Given that padding size depend even on compiler options, keep track of
2053 alignment and aligned_size only for the current architecture. Not a big
2054 deal since we send structure fields one after the other (seems
2056 - Add the datastructure used for IEEE paper by the PBIO guys to the test
2057 program, let it work on linux/gcc/little32. portability todo.
2059 2004-07-08 Martin Quinson
2060 - import and improve remote compilation support from FAST
2061 - make sure make check works on half a dozen of machines out there
2063 2004-07-07 Martin Quinson
2064 Let's say it's version 0.0.3 ;)
2065 - Implement conversions (yuhu!)
2066 - Let it work on solaris (beside conversion, of course)
2067 - Stupid me, using rand() to generate the conversion datatests in not wise.
2069 2004-07-06 Martin Quinson
2070 - Let make dist work, since I'm gonna need it to compile on remote hosts
2071 - Let Tests/datadesc_usage write the architecture on which the file was
2072 generated as first byte.
2073 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
2074 architecture descriptions.
2075 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
2076 on those architectures.
2077 - Optimization: send/recv array of scalar in one shoot
2079 2004-07-05 Martin Quinson
2080 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
2082 - Plug a whole bunch of memleaks
2083 - each process now have to call gras_{init,exit}. One day, their log
2084 settings will be separated
2085 - Continue the code factorisation between SG, RL and common in Transport.
2087 2004-07-04 Martin Quinson
2089 - Redistribution between SG and RL.
2090 We wanna have to accept in SG, so move accepted related parts of RL in
2091 the common part. (more precisely, the dynar of all known sockets is no
2092 more a static in transport.c, but part of the process_data)
2094 [gras_stub_generator]
2095 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
2097 2004-07-03 Martin Quinson
2098 - Create a new log channel tbx containing dict, set, log, dynar (to shut
2099 them all up in one shot)
2101 - Fix the ugly case of reference to dynamic array.
2102 - New (semi-public) function gras_datadesc_size to allow the messaging
2103 layer to malloc the needed space for the buffer.
2105 - gras_socket_close now expect the socket to close (and not its address to
2106 put NULL in it after it). This is because the socket passed to handlers
2107 is one of their argument (=> not writable).
2109 - propagate the interface cleanup from last week in datadesc, ie remove a
2110 superfluous level of indirection. User pass adress of variable
2111 containing data (both when sending and receiving), and not of a variable
2112 being a pointer to the data. Let's say that I like it better ;)
2113 The price for that is constructs like "int msg=*(int*)payload" in
2114 handlers, but it's a fine price, IMHO.
2116 - Let it work in RL (yuhu)
2118 2004-06-21 Martin Quinson
2120 - porting SG plugin and SG select to new standards (works almost).
2121 - plug memleaks and fix bugs around.
2124 - cleanup the prototype of data recv and force users to specify when they
2125 want to handle references to objects. Test case working even for cycles.
2126 - plug memleaks. Valgrind is perfectly ok with this.
2128 2004-06-12 Martin Quinson
2130 - cleanup the separation between plugin and main code in plugin creation
2132 2004-06-11 Martin Quinson
2134 - Reput hook for raw sockets, needed for BW experiments
2135 - kill a few lines of dead code
2136 [Data description] Interface cleanup
2137 - gras_datadesc_by_name returns the searched type or NULL.
2138 That way, no variable is needed to use a type desc once, which makes
2140 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
2141 two parameters were strings (field name, type name), leading to
2143 [Dicos] Interface cleanup
2144 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
2145 This is consistant with the dynar API.
2147 2004-04-21 Martin Quinson
2149 - Porting to new standards.
2151 - interface cleanup.
2152 There is no bag anymore, no need to take extra provision to mask the
2153 pointers behind "ID".
2154 Better splitup of functions between files create/exchange/convert.
2155 This is still a bit artificial since convert and receive are so
2156 interleaved, but anyway.
2158 - add a queued message list to procdata (the ones not matching criteria
2160 - factorize some more code between SG and RL wrt procdata
2162 - use gras_exit in example to track memleaks
2163 - get rid of gs_example now that GS is properly integrated into gras
2164 - update run_test to integrate the lastest tests (datadesc)
2166 - rename WARNINGn macros to WARNn since it prooved error-prone
2168 2004-04-19 Martin Quinson
2170 - register init/exit functions within gras module mechanism
2171 - send/receive function.
2172 Convertion is not implemented, but short-cutted if not needed.
2173 struct/array elements are sent one by one (instead of block-wise), but
2174 nobody really cares (yet). Get a prototype before optimizing.
2175 - tests (using a file socket) for DD send/receive on:
2176 - base types: int, float
2177 - array: fixed size, string (ie ref to dynamic string)
2178 - structure: homogeneous, heterogeneous
2179 - chained list, graph with cycle
2180 Believe it or not, valgrind is not too unhappy with the results. The
2181 cycle happily segfaults, but the others are ok. And I'm sick of pointers
2185 - Bugfix when using a filename explicitely (instead of '-')
2187 2004-04-09 Martin Quinson
2189 - factorize more code between RL and SG in socket creation
2190 - Complete the implementation and tests of:
2192 o file (only in RL, and mainly for debugging)
2194 I lost 3 days to design a portable address resolver, and then decided
2195 that the prototype mainly have to run on my box.
2196 Addressing portability too early may be like optimizing too early :-/
2198 - use gras_init in the Tests instead of the crappy parse_log_opt
2199 (the latter function is removed)
2200 [Conditional execution]
2201 - New functions: gras_if_RL/gras_if_SG (basic support for this)
2202 [Code reorganisation]
2203 - Get rid of libgrasutils.a since it makes more trouble than it solves.
2204 Build examples against the RL library, since there is no way to disable
2205 its creation for now.
2207 For information, the beginning of coding on GRAS was back in june
2208 2003. I guess that every line has been rewritten at least twice since