1 SimGrid (3.5) unstable; urgency=low
4 * New feature to any SimGrid-based simulator: Model-Checking
5 Check SIN#1 for more details.
8 * New Model SMPI (three-interval linear regression for correction factors)
9 See RR-7426, available at http://hal.inria.fr/inria-00527150
10 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
11 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
12 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
13 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
14 * See also src/smpi/README
19 - Tracing API changes: TRACE_start and TRACE_end should not be called
20 by user-code. They are automatically called by simulators created
21 with SimDAG, MSG and SMPI if the toolkit is compiled with
22 tracing_enabled=ON. Categories declaration and utilization remain the
23 same for MSG and SimDag.
24 - A function was added to the tracing API to declare categories with
26 - TRACE_category_with_color (char *category, char *color)
27 where color must be in the following format
28 "%f %f %f", red, green, blue
29 and red, green, blue are float values in the interval [0, 1]
30 - User can specify NULL as color parameter, or continue calling
32 On that case, the tracing system will define random colors
33 - The following command-line options are supported:
34 --cfg=tracing/filename:msg.trace
35 --cfg=tracing:1 (activate tracing, needed to use others)
36 --cfg=tracing/platform:1 (categorized resource use)
37 --cfg=tracing/uncategorized:1 (uncategorized resource use)
38 --cfg=tracing/msg/task:1 (task creation)
39 --cfg=tracing/msg/process:1 (process creation, migration)
40 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
41 --cfg=tracing/smpi:1 (SMPI interface tracing)
42 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
43 - examples of examples/msg/tracing updated
45 - DAXLoader and DOTLoader functions can generate tasks with categories
46 - A new function to attribute a category to SD tasks:
47 TRACE_sd_set_task_category (SD_task_t task, char *category)
48 * Tracing the MPI interface implemented by SMPI
49 - Collective operations are traced with states
50 - Point-to-Point operations are traced with states/links
51 - Tracing activated by a parameter "-trace filename" passed
52 to smpirun during execution (considering that simgrid
53 is compiled with tracing enabled)
54 - To run the simulation with gdb, the simulator
55 accepts --cfg=tracing/smpi:1 to trace SMPI
56 - tesh files to check if smpi tracing is ok
57 - See examples/smpi/NAS/DT-trace
58 * GTNetS tracing re-worked
59 - adaptation to the tracing system of GTNets to cope
60 with modifications regarding the fullduplex mode
61 - new tesh files to check if gtnets tracing is ok
64 * Asynchronous communications through the functions:
65 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
66 * New function: MSG_load_platform_script()
67 to make possible using a lua script instead of XML files to set up platforms
68 * New function: MSG_set_function
69 to associate functions to processes, used when bypassing the parser
70 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
72 Platforms: Add some more examples in examples/platforms
73 * Grid'5000: see www.grid5000.fr
74 * *_30000_hosts.xml: various huge files [mainly scalability testing]
77 * Change the XML format. This is a very important modification. SimGrid
78 3.5 introduces a new hierarchical format based on the notion of
79 Autonomous Systems. Compatibility with old format is ensured
80 through the perl script provided in the install bin directory
81 bin/simgrid_update_xml.
82 It is now possible to build platforms with specific routing
83 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
84 connect several platforms together. We will try to provide soon
85 set of realistic platforms exploiting these properties (have a
86 look at examples/platforms/ for the moment).
87 * Take the opportunity of the XML format change to be a good XML citizen:
88 rename link:ctn to link_ctn and similar changes (also dealed with by
90 * Add a new routing scheme (rule-based) using regular expressions. It
91 enables to have an extremely low memory footprint when the
92 underlying routing is simple and can be compactly described. You
93 need to have libpcre4-dev (perl regular expressions) installed if
94 you want to use this routing scheme.
95 * Revive the cluster TAG and allow to easily and efficiently (both in
96 term of memory and speed) connect clusters together. Have a look
97 at teshsuite/simdag/platforms/ to see how this can be done. With
98 this tag, you can create clusters with thousands of tasks at no
99 cost (have a look at examples/platforms/).
100 Note: clusters are implemented as ASes, so there is no need for an
101 enclosing AS tag if you have only one cluster in your platform.
102 * Add new generic functions in the public interface that allows the user
103 to call SURF 'create_resource' methods from your code (same
104 functionality as the XML bypass mechanism but with a much lighter
106 * Add a new model (enabled through command line --cfg=network/model:SMPI)
107 that uses a piecewise linear approximation to produce better
108 results when exchanging small messages.
109 * Add a new parameter to handle correctly full duplex link and account
110 for interferences between uplink and downlink communications
111 (activate with --cfg=fullduplex:1).
114 * Rename the SD_READY (all dependencies are satisfied and task is
115 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
116 dependencies are satisfied) state.
117 This prevents a confusion between the notion of "ready to schedule"
118 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
119 simulated" (SD_RUNNABLE) used by the simulation kernel.
120 * Change the way a task is considered as ready. Instead of removing
121 dependencies when a task is done, a counter is decreased. This way,
122 it is always possible to reach ancestors thanks to the
123 SD_taks_get_parents function (even after the end of the simulation.)
124 * Change the return type of SD_Simulate from (SD_task_t*) into
125 xbt_dynar_t. This function was in handling a dynar internally and
126 converted it into a NULL terminated array for historical reasons.
127 * New function SD_dotload(char*) to load a DAG described in dot
128 format. This loader and the corresponding examples require the
129 installation of the graphviz library.
130 * Fix a bug in the management of tasks of size 0 in the surf network
131 models. This problem was only visible with SIMDAG and you should
132 thus disregard results produced with earlier versions if you
133 relied on this feature (some tasks were blocked because of this).
134 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
135 even though your applications were doing only point-to-point
136 communications and sequential computations. Now you can really use any
137 model you want (of course, if you create real parallel tasks, which are
138 not implemented in most models beside ptaskL07, this will abort).
139 * Add an example that schedules a DAX on an heterogeneous platform
140 using a Min-Min strategy.
141 * New function SD_workstation_get_current_task() that returns the kind
142 of task currently running on a workstation in the sequential access
144 * Raise some warnings when unexecuted tasks remains at the end of the
145 simulation. This is usually caused by cycles in the DAG.
148 * New function: SIMIX_process_set_function() called by MSG_set_function
149 * Change the underlying waiting queue in semaphores so that a process
150 can wait on several of them simultaneously (as in waitany).
151 * Fix the way to handle tokens in semaphores so that all access patterns
152 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
153 * kill the dirty pimple SIMIX_message_sizes_output()
154 Please use (proper) visualization instead
157 * New data container: setset (set of sets of elements)
158 * New module: mmalloc (mapped malloc, allowing to have several
159 independent segments of malloc)
160 * New function: xbt_dict_cursor_set_data()
161 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
162 * New function: xbt_dynar_is_empty()
163 * New function: xbt_fifo_get_last_item()
164 * Fix xbt_dynar_shrink(): use the right element size.
165 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
166 uninitialized areas during expand.
167 * Fix semaphores: previous implementation was severely broken.
168 * Use library init/fini functions for our initialization.
169 - you can use logs and other feature as soon as you want in your
170 code (even before the xbt_init / MSG_init)
171 - xbt_exit is now a no-op and produce a warning when used.
174 * Port GRAS to new SIMIX mechanisms. This allows gras users to
175 benefit from the latest improvement to the simulation kernel.
176 * Kill measurement sockets for now. If you rely on them, sorry. This
177 release is not for you. This feature will be reintroduced in the
178 future, but we cannot delay the release any further.
179 * New function: gras_msgtype_get_name().
180 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
181 * Fix (at last) the pmm example: it should not randomly fail anymore.
183 Build chain: bug fixes and overall polishing
184 * Cmake is now stable enough. Hence, we killed the autotools.
185 * Port to windows ( TM :)
186 * Fix the 'make install' target.
187 No need to use 'make install-simgrid' anymore
188 * Introduce a 'make dist' target compiling a *source* archive
189 'make package' compiles a binary archive
190 * Compile java files only on need
191 * Add --cd and --setenv command line options to tesh
192 * Out of source builds are not fully supported yet, but we are close
193 * Enable supernovae and optimization flags by default for our users
196 * Add layer to set up environment directly from lua, without XML.
197 * The effect of gras_stub_generator can be achieved through
198 lua too (check examples/gras/console/ping_generator.lua)
200 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
202 SimGrid (3.4.1) stable; urgency=low
204 The "Polishing easter eggs is probably a good idea" release.
205 This is a bug fixes release only.
209 * Fix a bug preventing the tasks from begin garbage collected.
212 * Fix a bug occuring when a host involved in a communication fails.
213 This was not detected properly by the other peer involved in the
214 communication. Now, it's reported as a network error.
217 * Warn the user about loop dependencies in data flow of DAX files
218 * Obey the control-flow dependencies of DAX files
221 * Add option "enable_smpi" allowing to not compile SMPI.
222 Probably useful for the (Mac) users experiencing a build error here
223 * Improve the detection of lua5.1 and ruby1.8
225 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
227 SimGrid (3.4) stable; urgency=low
229 The "Easter in Cargese" release. Also known as (major changes):
231 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
232 ~> bindings were greatly improved
233 ~> new tracing infrastructure for better visualization introduced
235 * the "Welcome to configury modernity" release.
236 ~> we switched from autotools to cmake, and improved our cdash
239 A more detailled list of changes follow (full detail in svn log).
241 Java Bindings: Various Cleanups
242 * (install java-gcj-compat-dev on debian-like to use them)
243 * Remove put/get: no need to export deprecated interface in Java
244 Use send/receive instead.
245 * Cleanup the examples and add a README per directory
246 * Remove example autoDestination (that's the only way to go now)
247 * Remove example explicitDestination (was a plain copy of basic)
248 * Make JniException a runtime exception, so that there is no need to
249 declare the fact that you may encounter such a beast. I guess that
250 nobody will ever want to survive such error.
251 * Create specific errors for each MSG case of failure:
252 host failure, transfer failure, timeout, task cancelled
253 * Cleanup the exceptions that may get thrown by each function
254 * Other internal cleanups in Java bindings. Performance still bad :/
255 Ruby and Lua Bindings: create them
256 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
257 * That's new and great, you should try them out.
258 Same functionalities than Java bindings, only even less polished
260 * Kill the useless "rate" argument of SD_task_get_execution_time()
261 Everyone used to provide -1 as a value, it was not used, and the
262 semantic of a possible use wasn't even clear.
263 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
264 either as comm costs or compute costs to mean that there is no
265 such thing for that specific task.
266 * Add a SD_task_set_name() function
267 * Fix SD_task_unschedule() on typed tasks
268 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
269 * In DAX loader, accept useless 'level' attributes to <job> since
270 LIGO DAGs have them (seem to be to ease graphical representation).
272 * Add an example masterslave_mailbox.c using send/receive and not
273 the deprecated put/get interface.
274 * Kill the MSG_paje_output() function. It's a noop since 2 years.
275 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
277 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
278 (declare MSG_USE_DEPRECATED to still have the old name)
279 * Add a MSG_task_set_data() function
280 * About trace replay (see examples/msg/actions):
282 - Allow to work with splitted trace files for each process
283 Give the specific trace file as argument of each process,
284 and call MSG_action_trace_run(NULL)
285 You can still have one merged file for all processes.
286 - Fix implementation of collective operations
287 * Allow task_execute() on 0-sized tasks (closes #10063)
289 * This is the first release of SimGrid where SMPI is not considered
290 beta anymore (even if some corners should still be improved)
291 * Port over the new SIMIX_network submodule (internal refactoring)
292 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
293 * Implement more missing elements of the standard:
295 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
297 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
298 MPI_Scan() MPI_Get_processor_name()
299 - Added implementation of missing case for Alltoall (warning: it's
300 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
301 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
302 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
303 SMPI_MPI_Allgatherv()
305 - MPI_Waitsome() was broken
306 - Allow relative includes in smpicc
307 - Command line cfg argument 'reference_speed' was ignored...
308 - Some functions did not properly lead to auto-benching of user code
309 - smpicc passes -O2 by default (just like openmpi one)
311 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
312 * Bug fixes about timeouts during communications
313 * add SIMIX_message_sizes_output() as a pimple to write to file the
314 amount of messages per size. Use gnuplot to get histogram.
315 Pimple because that's the only user-visible function of simix,
316 defined directly in xbt.h (irk, sorry)
318 - Add a SIMIX_sem_get_capacity() function
319 - Fix interactions with processe resume/suspende
320 - release_forever() was stupidly broken
321 - Fix SIMIX_display_process_status() for processes in a semaphore
322 - Make SIMIX_sem_block_onto() user-visible
323 * Refactoring context stuff:
324 - Use pseudo-OOP for better modularity
325 - reimplement SIMIX_process_kill() without process_schedule() so
326 that the latter can take as invariant that it is called from
328 - Merge context_start into context_new for sake of simplicity
330 * Add a Vivaldi network model, coded live during SUD'10 ;)
331 * Rename configuration variables to start a hierarchy:
332 o cpu_model -> cpu/model
333 o network_model -> network/model
334 o workstation_model -> workstation/model
335 * New configuration variables:
336 o network/bandwidth_factor: correction to bandwith
337 o network/latency_factor: correction to latency
338 o netwotk/weight_S: correction to the weight of competing streams
339 * Add a long description to the models, that users can see with such
340 argument on the command line: --cfg=cpu/model:help
341 * --help-models display the long description of all known models
343 * config: add the ability to set a default value after registration
344 Does not override any previously set value (e.g. from cmd line)
345 * dict: allow to have integer key and data.
346 When so, you need to use the following functions
347 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
348 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
349 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
350 In contrary to regular dicts, the key is not malloced before copy.
351 Mixing scalar and regular elements in the same dict is not tested
353 * Allow to use xbt_dynar_shrink() to expend the dynar instead
354 Tracing for Visualization:
355 * SimGrid is now instrumented in order to generate a trace file for
356 visualization analysis: to use it, need to compile SimGrid with the
357 "tracing" option enabled, and instrument the program using SimGrid with
358 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
359 (among other functions).
360 * The instrumentation only traces the platform utilization for now
361 * Documentation to use the tracing functions and how to analyze the
362 traces with the Triva tool is written.
363 * More information about: SimGrid FAQ (in the section Tracing Simulations
366 * We moved to cmake as default build system. Autotools support will
367 be dropped soon. Check the FAQ for more info about how to use it.
368 * Greatly improved our cdash/ctest interactions
369 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
370 * Added memory checking tests with valgrind; lot of memleak fixing.
371 This may be the first release of simgrid with so few memory issues
372 * Added code coverage tests.
373 Our coverage is still improvable, but at least we see it on cdash.
375 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
377 SimGrid (3.3.4) stable; urgency=low
379 The "Desktop Grid needs love too" release (also called Xmas release).
382 * Major speedup in the maxmin system solving by using lazy evaluation
383 Instead of solving completely the maxmin system at each iteration,
384 only invalidate (and recompute) the modified parts.
385 This new feature is enabled in default models but you can try to
386 turn it on with "--cfg:maxmin-selective-update=1" for other models.
387 * Cas01 IMproved as default CPU model
388 This CPU model is the same Cas01 model, but it uses the
389 maxmin-selective-update flag and a heap structure to manage
390 actions on SURF kernel.
391 It reduces the complexity to find the next action to finish and,
392 consequently, it's faster than the old Cas01.
393 This is the new default CPU model (Cas01).
394 * Rename the old Cas01 model to Cas01_fullupdate
395 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
396 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
397 * CpuTI (CPU Trace Integration)
398 A new CPU model whose objective is simulate faster when using
399 availability trace files.
400 Instead of using a full featured, over engineered maxmin system for
401 CPU modeling, this model does the pre-integration of traces files
402 to calculate the amount of CPU power available, and so, executes
403 faster than the old CPU models.
404 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
405 * Use LV08 as default network model since it gives better accuracy
406 for small messages and shouldn't change things for big ones.
407 Use --cfg=network_model:CM02 to get the previous behavior.
410 ******************************************
411 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
412 ******************************************
413 * The new CPU model may changes simulations!
414 The point is that events occurring at the exact same timestamp
415 are not scheduled in the same order with the old and new
416 version. This may be enough to completely change the execution
417 of simulations in some cases.
418 * The new network model will change simulations!
419 This new model is more realistic than the previous one, so you
420 should consider redoing your old experiments with this model.
421 Sorry for the inconvenience.
424 * Introduce the supernovae compilation mode
425 When compiled that way, the whole SimGrid (or almost) is put in a
426 single compilation unit and compiled in one shoot.
427 This is to help gcc which has difficulties to inline stuff from one
429 The speedup seem to be above 15%, althrough more tests are needed on
430 amd64 to confirm that gain.
433 * Port of MSG's mailbox on top of SIMIX network
434 The put/get mechanism was greatly simplified on the way.
437 * New SIMIX network module. Provides:
438 - Mailbox: rendez-vous mecanism to find with who you want to speak
439 - Synchronous send/recv: easier and hopefully faster since the
440 logic is handled in the maestro process directly now
441 - Asynchronous send/recv: you dreamt of it? It's here now
442 Too bad that nobody cared enough to propagate the change to MSG.
443 * Add semaphores as SIMIX synchronization mechanism.
446 * new function SD_daxload(char*) to load a DAX file
447 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
448 * Introduce typed tasks. Specify its kind and cost at creation.
449 At scheduling, just give where it should be placed, and the cost
450 for each involved resource is automatically computed.
451 Existing constructors so far (more to come of course):
452 - SD_task_create_comm_e2e() for end-to-end communication
453 - SD_task_create_comp_seq() for sequential computation
454 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
455 * new function SD_task_dump() for debuging display
456 * new function SD_task_dotty(task,FILE*) writing to file the info
457 about the task in dotty format
458 * SD_task_dependency_exists() can now cope with having one of its
459 arguments NULL. If so, it tests whether the other argument has any
461 * Add getters on list of preceding/following tasks:
462 SD_task_get_parents(task) and SD_task_get_children(task)
463 * Add getters on amount of workstations and list:
464 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
465 * Add getter on task kind: SD_task_get_kind(task)
466 * Update the start_time and finish_time of tasks on completion/failure
467 * Bugfix: Remove task from state swags when destroyed
470 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
473 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
474 dynar with one separate thread per value of the dynar.
475 * Change the prototype of xbt_thread_create(), sorry.
476 Added a boolean parameter indicating whether we want to join this
477 thread (used in SG only for now)
478 * Implement xbt_thread_join and xbt_thread_yield in SG also.
481 * GTNetS wrappers should now be usable again (and betterly tested too)
482 * Fix a major regression from 3.2 where the timeout provided to
483 MSG_task_put_with_timeout() was used as absolute time before which
484 the comm should be done.
485 * Start to fix the <cluster> tag.
486 - Internal links should be good now (beside of the loopback, which
487 use the private link instead)
488 - paths to the external world is still rather broken
489 - the <route:multi> tag is just broken. Actually that's brain-dead.
490 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
491 to make it less stupid
492 ** Check your platform with teshsuite/simdag/platforms/flatifier **
493 * Fix a source-level compatibility glitch from 3.2: after defining
494 MSG_USE_DEPRECATED, you can use the old name
495 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
496 * Allow to compile from the SVN with automake 1.11
497 * Fix some problems when using the "start_time" tag in deployment XMLs.
498 * Fix #8569: XBT/synchro.h has redundant declarations
499 * Fix #8563: MSG return values and exceptions
500 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
501 * Integrate patch #8636: Obey DESTDIR when installing documentation.
502 Thanks to Robson Peixoto.
503 * Fix a vicious bug in dictionaries inducing that some elements were
504 not freed on xbt_dict_free()
506 Portability report of this version:
507 * Main portability targets:
508 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
510 - mac leopard on i386
511 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
512 but nothing critical.
513 * Other platforms: windows, AIX and others were not tested for this release
515 Timing report of this version:
516 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
517 scenario parameters). From 8h to a few seconds in desktop grid settings.
518 * Supernovae brings about 25% speedup on i386.
520 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
522 SimGrid (3.3.3) stable; urgency=low
524 The "Need for Speed" release.
526 The timings done to validate the 3.3.2 were faulty.
527 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
529 The problem was a conversion from a manually handled vector to
530 xbt_dynar_t on the critical path.
531 xbt_dynar_foreach calls functions, inducing stack management crap.
533 We inlined these functions and xbt_dynar_foreach is now breath taking.
534 We also inlined xbt_swag_belong on the way.
536 Here are some approximate speedup measurements (on master/slaves
537 simulations lasting between 10s and 20s each):
538 3.3.1 -> 3.3.2: about same performance
539 3.3.2 -> 3.3.3: 40% speedup
540 3.3.1 -> 3.3.3: 40% speedup
541 3.3.1 with inline patch -> 3.3.3: 30% speedup
543 Our reading is that the refactoring which occurred in 3.3.2 made us
544 suffer much more from the xbt_dynar_foreach low performance, but
545 once we solved this, this refactoring proved to be very performance
546 effective. From the 40% speedup, somehow, 10% are due to the
547 inlining and 30% to the refactoring.
549 That's a pitty that gcc cannot inline functions placed in other files
550 alone. We have to choose between:
551 - break the encapsulation (by putting private data structures and
552 accessors in headers files to help gcc)
553 - live with low performance
554 - switch to a decent compiler such as icc (not quite possible).
556 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
558 SimGrid (3.3.2) stable; urgency=low
560 The "Simplicity does not preceed complexity, but follows it" release.
562 The main contributors of this release were (lexical order):
563 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
566 * Extract the routing logic into its own object.
567 (was dupplicated in network.c and workstation_LV07.c;
568 Allows to implement other ways of storing that info)
569 => kill now useless network_card concept
570 - Use dynar to represent routes (instead of void** + int*)
571 - kill link_set (use surf_network_model->resource_set instead)
572 - Add a command-line option to choose the routing schema to use
573 - Add three new models:
574 * Floyd (shortest path computed at initialization)
575 * Dijikstra (shortest path recomputed all the time)
576 * Cached Dijikstra (shortest path computed on need)
577 All these models where contributed by Silas De Munck, and are
578 described in his ICCS09 paper.
580 * Simplify model declaration
581 (less redirections, less function to write when defining a model)
582 - Factorize stuff between models:
585 surf_model_resource_set(model)
586 surf_model_resource_by_name(model, name)
587 - Unify the types of models in s_surf_model_t (using an union)
588 - Embeed fields of common_public directly into s_surf_model_t
589 - Rename model methods:
590 action_free ~> action_unref
591 action_change_state ~> action_state_set
592 action_get_state ~> action_state_get
593 - Change model methods into functions :
594 (model)->common_public->action_use ~> surf_action_ref
596 * Implement a generic resource; use it as ancestor to specific ones
597 (allows to kill duplicated code in models)
598 Drawback: timer command don't need no name nor properties;
599 workstation_CLM03 don't need no properties
600 (but I guess we can live with those few bytes wasted)
602 * Improve the action object model
603 - implement a constructor avoiding dupplicated code about field
604 initialization in generic_action part.
606 * Kill the SDP model: it has an external dependency, is deprecated
607 in flavor of modern lmm models, and didn't compile since a while
610 * Relocation of the context module from XBT to SIMIX.
611 (the context were decoupled from the simix processes, duplicating a lot of code)
612 => a lot of code was factorized
613 - less overhead is introduced during scheduling
614 - simpler API for the context factory
615 - the logic for process creation,destruction and manipulation was simplified
616 * Simplification of the s_smx_process_t data structure.
617 => accesing the simix level data associated to a process is faster now,
618 and the code is a lot more readable.
621 * Implement some more MPI primitives:
622 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
623 -implementation: Bcast: flat or 2-ary tree (default),
626 Allreduce: Reduce then Bcast
627 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
628 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
629 Alltoallv: flat tree, like ompi
631 * Add support for optimized collectives (Bcast is now binomial by default)
632 * Port smpirun and smpicc to OS X
635 * Kill SD_link_get_properties: hard to maintain and makes very little sense
636 Shout out if you used it.
639 * Display the list of still queued messages in SG mode when existing
643 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
644 * Add xbt_graph_node_get_outedges() [Silas De Munck]
645 * Add xbt_str_from_file(FILE*)
646 * Add xbt_dict_get_key achieving a linear reverse search
647 * Remove the context module
649 Portability report of this version:
650 * Main portability targets:
651 - Linux(debian)/x86/context
652 - Linux(debian)/x86/pthreads
653 - Linux(debian)/amd64/context
654 - Linux(debian)/amd64/pthreads
655 On these, we still have the eratic breakages of gras/pmm and
656 amok/saturate_sg reported in previous version. We still think
657 that the tests are the cause of the fault, not the tested code.
659 - Mac OSX Leopard/x86/context
660 Still false negative in tesh autotesting.
661 Smpi still fails, but this time because readlink does not accept -f
662 Everything seems to work properly beside of that.
665 - AIX version 5.3 (only tested contexts this time)
666 Smpi still fails there because mktemp is not installed.
667 Everything seems to work properly beside of that.
669 I managed to compile it for the first time, but several breakages.
670 Won't delay the release for this exotic platform.
672 * Windows: it's still lagging behind. If you want to help, please
675 Timing report of this version:
676 This version seem to be more than 5% faster than 3.3.1 (on linux
677 64bits with contextes). The gain is less than expected, we are
678 investigating this for next release.
680 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
682 SimGrid (3.3.1) stable; urgency=low
685 * Implement a --cfg-help to show existing configuration variables
686 * Build chain do not require doxygen in maintainer mode
689 * fix a bug on struct sizeof computation, which prevented the
690 exchange of arrays of structs in some conditions
691 - added a regression test about this in datadesc_usage
692 * Allow the exchange of 0-long dynamic vectors.
693 - for that, use -1 as indicator of dynamic size instead of 0
694 - This implied to change any size from unsigned long to long,
695 reducing a bit communication abilities, but I guess that with
696 64bits being quite common, this is more than enough.
697 - This also induce a protocol change, thus bumping network protocol
698 version from 0 to 1 (if we have external users, we have to get
699 clean on that point too ;)
700 - added two regression tests about this in datadesc_usage
701 * Be more verbose when propagating local exceptions
702 This helps debugging.
703 * Display the status of simulated processes when receiving SIGINT in
707 * Allow to control the simulation from a trace file.
708 New functions MSG_action_register() and MSG_action_trace_run()
709 The first one allows to associate a function execution to each
710 kind of action while the second one parses a trace file and
711 triggers the corresponding actions within the system.
712 For now, only a toy example is provided in examples/msg/actions
713 * Add an exemple of process migration in examples/msg/migration
714 * Fix a bug in task exchange which broke MSG_task_get_sender()
715 Add a teshsuite regression test for that.
716 [Bug: if MSG_task_get_sender() is called after sender exit,
718 * Fix a bug which prevented suspend/resume to work properly
719 * Display the status of simulated processes when receiving SIGINT
720 This fixes a regression of v3.3. due to the introduction of SIMIX
721 * Bug fixing in failure management:
722 - trace could not start by a failure at time 0
723 - failure during communications were not working
726 * Add SIMIX_process_set_name() to change the name of the current
727 process in the log messages.
728 * Store smx_hosts in a dict since we only retrieve them by name
729 * Move the configuration infrastructure to surf
732 * Move the configuration infrastructure to surf
735 * Massive internal cleanups:
736 - Store internal structures on processes instead of hosts (allows
737 to have more than one process per host, in addition of being more
739 - Cleanup the initialization/finalization process
740 - Kill a whole bunch of unneeded synchronization:
741 processes run in exclusive manner within the simulator
742 - Move queues from global tables to process data fields
744 - now accept -platform and -hostfile arguments
745 - Pass the right rank value to processes according to the hostfile
746 * Compile the examples by default, and use them as regression tests
747 * Implement MPI_Wtime()
748 * Change the reference speed to a command line option
751 * TCP_gamma can now be specified as command line option using
752 --cfg=TCP_gamma:10000000.0
753 * Change the --surf-path cmd line option into --cfg=path:
756 * Also include strbuff from xbt.h public header
757 * xbt_ex_display(): do not free the exception after displaying
758 This allows to do more with the given exception afterward.
759 Users should call xbt_ex_free() themselves.
763 Portability report of this version:
764 * Main portability targets:
765 - Linux(debian)/x86/context
766 - Linux(debian)/x86/pthreads
767 - Linux(debian)/amd64/context
768 - Linux(debian)/amd64/pthreads
769 These targets fail about 1/10 of times on gras/pmm, but we believe
770 that this is because of the test, not because of simgrid.
771 amok/saturate_sg fails even more rarely, and the test may not be
774 - Mac OSX Leopard/x86/context
775 The test suite still spits tons of errors because some obscure
776 force prevents us from removing the temporary directories
777 arguing that they still contain some metadata I've never heard of.
778 Smpi fails because seq is not installed.
779 Everything seems to work properly beside of that.
782 - AIX version 5.3 (both contexts and pthread)
783 Smpi still fails there because mktemp is not installed.
784 XML inclusions seems rosty on AIX.
786 * Windows: it's still lagging behind. If you want to help, please
789 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
791 SimGrid (3.3) stable; urgency=high
795 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
798 * Introduce the SIMIX module: factorize code between MSG and GRAS.
801 Until now, GRAS were using MSG as an interface to SURF. It was
802 quite difficult because both interface have several differences
803 (MSG channels vs GRAS sockets were the most notable point).
805 This also opens the gate to SMPI (which should occur soon) and speed
806 up simulations by to 40% (even if it were not the main goal).
808 **************************************
809 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
810 **************************************
811 The point is that events occuring at the exact same timestamp are
812 not scheduled in the same order with the old and new version. This
813 may be enough to completely change the execution of simulations in
814 some cases. Sorry for the inconvenience.
816 * Cleanup and upgrade the XML format to push further scalability
817 issues (check http://hal.inria.fr/inria-00256883/ for more info)
819 * Improve the testing infrastructure with tesh. Now a very large part of
820 the code is tested not only by being run but also by checking that the
821 output match an expected output [Mt].
823 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
824 is really less memory-demanding, which should allow you to use
825 larger files in SimGrid [AL].
827 * Inform valgrind about our contextes, so that it becomes usable
828 with the default (and more effecient) version of SimGrid
829 [contributed by Sékou Diakite, many thanks]
832 * Introduce a listener thread in charge of receiving incomming
833 messages from the network. It allows to overlap communication and
834 computation but most notably, it removes some stupid deadlocks due
835 to the fact that so far, a process could not send and receive at
836 the same time. This made most non trivial communication schema
838 * Convert the PIDs from long int to int to match the MSG ones (and
840 * New function: gras_agent_spawn() to launch a new process on
841 current host. Only working in simulation for now. [Mt]
842 * New function: gras_os_hostport() returning a constant form (ie,
843 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
846 * Make the backtrace of exceptions more human readable [Mt]
847 * New module: xbt/str [Mt]
848 a ton of string utility functions (split, join, printf to a newly
849 allocated buffer, trim, etc)
850 * New module: xbt/hash [Mt]
851 SHA1 hashing algorithm (more to come if needed)
852 * New module: xbt/synchro [Mt]
853 synchronization tools (mutex and conditions) working the same way
854 in simulation and in real life (mainly useful for GRAS, but not
856 * New module: xbt/queue [Mt]
857 classical producer/consumer synchronization scheme
858 * xbt_dynar_new_sync() creates a synchronized dynar. All access
859 (using the classical functions will get serialized) [Mt]
860 * Make dictionary internal table dynamic. No need to specify its size
861 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
863 * Make sure the log channels are organized as a tree under windows
864 (because of ANSI C compatibility issue, any channel were child of
868 * Cleaned many thing in surf and fixed a few bugs [AL].
869 * Add a nice command line configuration mechanism to compose models [AL].
870 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
871 the previous one (KCCFLN05). It relies on something that looks like
872 a max-min sharing mechanism but cannot be written as such. A new solver
873 was thus designed [AL].
874 * Add a new solver to lmm. Based on Lagrange optimization and
875 gradient-based descent, it enables to efficiently maximise systems s.a
877 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
879 This solver enables to propose two new network models for TCP Reno and
880 TCP Vegas based on Low's work. These models still need to be fully
881 tested though [Pedro Velho].
884 * Bug fix in SD_simulate. Now the time bound given as argument is
886 * Use the new parallel task model (ptask_L07) as default.
887 * Use the SURF command line configuration mechanism.
888 * 0-size tasks (for synchronization) should now work.
890 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
892 SimGrid (3.2) stable; urgency=high
896 We still experience issues on this platform, but we believe that at
899 GRAS API BREAKAGE (for simplification purpose, sorry):
900 * the gras_msgtype_by_name is not used anymore. Instead of
901 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
902 you can write (and must)
903 gras_msg_send(toserver, "request", &request);
904 - If you still want to pass a gras_msgtype_t to the function (to cache
905 the type and avoid the lookup time), use the gras_msg_send_() variant.
906 - Impacted functions:
907 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
908 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
909 * The callbacks are now expected to return 0 when everything went well
910 (just like the main() function)
912 GRAS new features and improvements:
913 * New module mecanism where user code can use per process globals [Mt]
914 This is similar to gras_userdata_*() functions, but for libraries. It
915 factorize some code developped over and over in the examples and AMOK.
916 It has still to be documented and used (only amok/peermanagement is
918 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
919 loss when they were small enough to fit into the buffer and sent quickly
920 enough so that they can all get received in one shoot.
921 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
922 if not found. Use the *_or_null() variant for the old semantic.
923 * In gras_msg_handle, do not discard messages without callback.
924 They are probably messages to be explicitly awaited later (ie, proofs of
925 mis-synchronization in userland since they are sent before being awaited)
927 * gras_socket_meas_send/recv: semantic changed!
928 The numerical arguments used to be (1) the total amount of data to send
929 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
930 messages. This was need for the fool willing to send more than MAXINT
931 bytes on quite fat pipes.
934 * Do really rename the hostmanagement module to peermanagement. [Mt]
935 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
936 but this is rather new and this was documented in the module
937 documentation (poor excuses, I admit)
938 * Bandwidth measurement semantic changed! This follows the changes to
939 gras_socket_meas_send/recv explained above.
942 * A sequential mode has been added to the workstations. When a workstation
943 is in sequential mode, it can execute only one task, and the other tasks
944 are waiting in a FIFO. [Christophe Thiery]
947 * The KCCFLN05 workstation model now handles parallel tasks. It is the
948 model for SIMDAG. [Christophe Thiery]
949 * Bug fix in the maxmin solver: Some values were close to 0 instead of
950 equal to 0, which caused some bad behaviors in
951 saturated_constraint_set_update. I now use a threshold mechanism like in
955 * When running manually src/testall, you select specific units [Mt]
956 testall is the result of our cunit mecanism, and should replace all
957 the scripty thingy around since bash don't run easily on billware.
959 * A mallocator system has been added. [Christophe Thiery]
960 Mallocators allow you to recycle your unused objects instead of freeing them
961 and allocating new ones.
963 Documentation update:
964 * FAQ reworking + New FAQs:
965 - "Valgrind spits tons of errors!" [Mt]
966 - "How to repport bugs" [Mt]
967 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
968 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
969 - Communication time measurement within MSG [AL]
970 - I experience weird communication times when I change the latency [AL]
973 - an introduction to the framework and to the used communication model
974 - an initiatic tour introducing the most proheminent features:
976 . Lesson 0: Installing GRAS
977 . Lesson 1: Setting up your own project
978 o Part 2: Message passing
979 . Lesson 2: Exchanging simple messages
980 . Lesson 3: Passing arguments to the processes (in SG)
981 . Lesson 4: Attaching callbacks to messages
982 . Lesson 5: Using globals in processes
983 . Lesson 6: Logging informations properly
984 . Lesson 7: Using internal timers
985 . Lesson 8: Handling errors through exceptions
986 . Lesson 9: Exchanging simple data
987 . Lesson 10: Remote Procedure Calling (RPC)
988 . Lesson 11: Explicitely waiting for messages
989 . Recapping of message passing features in GRAS
990 - A HOWTO section containing:
991 o HOWTO design a GRAS application
992 More are due, of course. They will come latter. In the meanwhile, you can
993 check the examples which are still here.
995 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
997 SimGrid (3.1) stable; urgency=high
1001 There was a stack corruption somewhere, visible only when optimizing
1002 with these versions. [Vince]
1005 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1006 module that enabled to play with Directed Acyclic Graphs. It is built
1007 directly on top of SURF and provides an API rather close to the old
1008 SG. Some old codes using SG are currently under rewrite to check that
1009 all needful functions are provided. [Christophe Thiery]
1012 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1013 extension of the classical CLM03 model that gracefully handles
1014 failures. This is now the default model for MSG and GRAS. It doesn't
1015 handle parallel tasks yet though. [AL]
1016 * Bug fix: Weights were not correctly set in the network part.
1017 WARNING: This may have resulted in incorrect results with simulations
1018 where there are more than one flow on a given link. [AL]
1021 * After a (long ?) discussion on simgrid-devel, we have decided that the
1022 convention we had on units was stupid. That is why it has been decided
1023 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1024 WARNING : This means that all previous platform files will not work as
1025 such with this version! A warning is issued to ask users to update
1027 A conversion script can be found in the contrib module of the CVS, under
1028 the name contrib/platform_generation/surfxml_update.pl [MQ]
1031 * Bug fix: Processes were started in reverse order, wrt deployment file.
1032 WARNING: if your code relies on this bug, please fix it. [AL]
1033 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1034 being executed on two different locations. [AL]
1035 * Bug fix: Failures are now better supported thanks to Derrick's tests
1036 (there was many failure situations I hadn't thought of and that weren't
1037 correctly handled). [AL]
1038 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1042 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1043 Exception occurring on server-side are propagated back to client (!).
1045 API CHANGE: the callback changed their prototype. Change:
1046 int my_handler(gras_socket_t expeditor, void *payload_data) {
1048 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1049 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1051 * New! function: gras_msg_handleall to deal with all messages arriving
1052 within a given period.
1053 * New! function: gras_socket_server_range to get a server socket in a
1054 range of port numbers (ease to avoid port number conflicts) [MQ]
1055 * New! gras processes display their backtrace when they get a SIGUSR1
1056 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1057 Sweet to debug RL processes [MQ]
1061 - Do not force experiment sizes to be expressed in kb, or it becomes
1062 impossible to measure the latency this way (needs one byte-long tests)
1063 WARNING: this changes the amok_bw_* function semantic. [MQ]
1064 - Implements the link saturation stuff. [MQ]
1065 * Peer management module:
1066 New! module factorizing code that we wrote over and over [MQ].
1069 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1070 - Test units are placed directly into the library code, they get extracted
1071 automatically and placed into the src/testall binary.
1072 - Convert most of the XBT tests to this system.
1073 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1074 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1075 live in the same C code. This required to change a little bit the API
1076 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1077 * New module: sparse graph structure with basic algorithms (this is work
1078 in progress and the API is not considered to be frozen yet). [AL]
1079 * Display more information on backtraces: source line & function names are
1080 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1081 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1082 * Reworked a little bit some #include statements to load only required
1083 headers. Some user code that relied on SimGrid to include stdlib or
1084 stdio may need to include it by themselves. [AL]
1085 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1087 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1088 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1089 but I guess that almost nobody used those parts. [MQ]
1091 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1093 SimGrid (3.0.1) stable; urgency=low
1096 * Unfortunately, I had missed 5 misnamed functions:
1097 xbt_fifo_item_t xbt_fifo_newitem(void);
1098 void xbt_fifo_freeitem(xbt_fifo_item_t);
1099 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1100 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1101 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1102 They're now deprecated. Please use their new versions:
1103 xbt_fifo_item_t xbt_fifo_new_item(void);
1104 void xbt_fifo_free_item(xbt_fifo_item_t);
1105 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1106 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1107 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1109 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1110 * Documentation: xbt_log module unmercifully reworked [MQ]
1111 * Bugfix: there was a problem with the ending of contexts with
1112 the pthread backend. It caused some weird deadlock or behavior
1113 depending on the pthread implementation. [AL]
1114 * Bugfix: get the exceptions raised in the simulator repport where
1115 and why they come from when they are not catched in time [AL, MQ]
1118 * Bugfix: Do repport the error when two non-connected hosts try to
1119 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1122 * Add additionnal checkings on communications. Assert that two
1123 communicating hosts are connected by a set of links... [AL]
1126 * Add additionnal checkings on channel values in communication [AL]
1127 * New: MSG_task_get_source to see on which host a task was generated [HC]
1128 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1129 the number of tasks waiting to be received on channel and sent
1131 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1132 waits for the first task coming from a given host.. [AL]
1134 GRAS new functionnalities: [MQ]
1135 * Enhance the parsing macro to allow the size of multidimentional objects
1136 to be given thru annotations.
1137 * New example (and documentation): Matrix Multiplication a la RPC
1138 (as when I was young!) and fix a bunch of bugs found on the way.
1140 GRAS performance improvements: [MQ]
1142 * Reduce the amount of cbps creation/destruction by making it static to
1143 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1146 * Change libdata to a set so that we can search for stuff by ID (and thus
1147 reduce the insane amount of dict lookups)
1150 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1151 RL communication on top of SG since it's so uneffective.
1152 It may also be used for inter-thread communication in RL, one day.
1153 * Use gras_datadesc_copy() to exchange messages on top of SG
1155 - improve message exchange performance on top of SG
1156 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1157 * Don't exchange on the network the size of the used part of buffer,
1158 instead, specify the possible buffer size to read().
1160 - reduces the amount of read/write calls (one pair per exchange)
1161 - reduces the amount of exchanged data (the size)
1162 - allows to retrieve all arrived data on receiver side, if we don't need
1163 it right now (subsequent read will peek the buffer)
1164 - allows the receiver to proceed with the begining of the stream before
1165 everything is arrived
1166 - make it possible to build an iov transport (using readv/writev)
1168 - take care of the data with non-stable storage (like stacked data),
1170 * If possible, TCP send uses vector I/O (when writev() is here)
1171 - Don't use it for receive since we send data sizes and data on the
1172 same stream, so we wouldn't be able to chain large amount of chunks
1173 before having to flush the stuff to read the size.
1174 * Rework the transport plugin mecanism to simplify it and reduce the
1175 amount of pointer dereferencement when searching for the right function
1178 * I guess that now, we do almost as few system calls as possible while
1179 doing as few data copy as possible.
1181 To improve it further, we could try to send all the sizes first and then
1182 all the data (to use iov on receiving size), but it's only a partial
1183 solution: when you have 2 dimensional data, the sizes of the second
1184 dimension is data of the first dimension, so you need 3 streams.
1186 I'm not sure the potential performance gains justify the coding burden.
1188 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1190 SimGrid (3.00) stable; urgency=high
1193 * New! Give the possibility to hijack the surf parser and thus bypass
1194 MSG_create_environment and MSG_launch_application. Have a look at
1195 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1197 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1199 SimGrid (2.96) unstable; urgency=low
1204 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1205 This deprecates the xbt_error_t mecanisms.
1206 It modifies (simplifies) all XBT and GRAS API.
1207 MSG API keeps unchanged (exceptions raised by XBT are catched from
1208 within MSG and masked with existing error handling facilities)
1211 * New! Add a FATPIPE model. [AL]
1212 * New! Add a parallel task model. [AL]
1213 * New! Add automatically a loopback interface (in the default
1214 network model) if none was precised.
1217 * Bugfix: MSG_process_resume now works with the current running process.
1219 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1220 * Modification of MSG_task_get_compute_duration. Once a task has been
1221 processed, the value returned by this function is now equal to 0. [AL]
1222 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1223 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1224 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1225 of a m_task. For now, MSG_task_cancel only works with computation
1227 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1228 of the processor (in Mflop/s) regardless of the current load on the
1230 * API Change: use proper naming convention for MSG_getClock and
1231 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1233 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1234 This function changes the priority of a computation task. This priority
1235 doesn't affect the transfer rate. A priority of 2 will make a task
1236 receive two times more cpu power than the other ones. This function
1237 has been added to suit the needs of Nguyen The Loc and hasn't been that
1238 much tested yet. So if it fails, please report it and send me your code.
1240 * API Change: removed all functions and types that were marked "deprecated"
1241 since many months. Renamed MSG_global_init_args to MSG_global_init.
1243 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1245 SimGrid (2.95) unstable; urgency=low
1248 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1249 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1250 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1253 * Bugfix: Allow absolute paths for platform description files [MQ]
1254 * Bugfix: do free the variables after use. Leads to drastic performance
1256 * Implement max_duration (ie, timeouts) on resources [AL]
1259 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1261 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1262 the needed support (timeouts) [AL]
1265 * Implement measurement sockets. You can now get the bandwidth between two
1266 hosts thanks to AMOK (see below). [MQ]
1267 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1268 dynar over the network (yeah) [MQ]
1269 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1270 * simple/static token-ring example. [Alexandre Colucci and MQ]
1271 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1272 in place before [MQ]
1275 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1276 type parsing. "short" and "long" modifiers were also ignored; other
1277 modifier (such as reference level) are still ignored. [MQ]
1278 * Embeed the buffer size within the buffer itself on SG. [MQ]
1279 That way, send() are atomic and cannot get intermixed anymore (at least
1280 the ones which are less than 100k; bigger messages still have the issue)
1281 * Array size pushed by the field, not by the field type (or each
1282 and every long int will push stuff to the cbps) [MQ]
1283 * use select() to sleep since it allows to portably sleep less than one
1286 GRAS (minor cleanups)
1287 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1290 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1291 * type category 'ignored' killed as it was never used and were difficult
1293 * whether a type can cycle or not is now a flag, leaving room for more
1294 flags (as "ignored", if we feel the need one day ;)
1295 * Rename raw sockets to measurement sockets since "raw" has another
1296 meaning in networking community.
1299 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1300 and offers features not belonging to GRAS but that most applications
1301 need. One day, it may be a set of plugins loadable at runtime.
1302 * New module: bandwidth
1303 bandwidth measurement between arbitrary nodes running this module. [MQ]
1305 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1307 SimGrid (2.94) unstable; urgency=low
1309 The first beta release of SimGrid 3 !
1313 * Update the main page and the FAQ. Adding references to gforge.
1316 * Add a gras_os_getpid function.
1319 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1320 * Extend the logs so that they also print PID, hostname, date, ... if
1322 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1323 and kill the old version which were in testsuite/
1324 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1328 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1329 for users. Instead, each programming environment (GRAS, MSG,...) use it
1330 when needed to provide such a feature to users.
1331 Don't shortcut the mecanism or you will also shortcut the virtualization
1332 you need on the simulator.
1336 * Cleanups in configury with regard to compile optimization/warning flags.
1337 Also add -fno-loop-optimize to any powerpc since it's the optimization
1338 killing gcc (< 3.4.0).
1339 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1340 needed by MSG examples complications
1341 * Borrow configury beautifications from PHP
1344 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1345 warning (thanks loris for stumbling into this one).
1346 * Bugfix: stop loading private headers (gras_config.h) from the public
1350 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1351 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1352 RL would imply a DNS resolver, which is *hard* to do in a portable way
1353 (and therefore delayed).
1354 * Implement a real timer mecanism and use it in timing macros. This allows
1355 to avoid rounding errors and get a 0.000005 sec precision in timing
1356 macros. While I was at it, various cleanups:
1357 - allow to declare more than one timed section per file (fix a stupid bug)
1358 - move some private declaration to the right place
1359 - merge conditional execution and timing macros into emulation module
1360 - document the module
1361 - make sure the module cleanups its mess on gras_exit
1362 * Documentation improvements:
1363 - (new) how to compile applications using GRAS
1364 - (new) emulation support (timing macros)
1366 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1368 SimGrid (2.93) unstable; urgency=low
1370 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1373 - Use Paje properly where used. Still to be sanitized properly.
1374 - Portability fix: Add an implementation of the contexts using pthread
1378 - Add xbt_procname(): returns the name of the current process.
1379 Use it to show the current process's name in all logging.
1381 - fix detection of older flex version and the reaction, since we do
1382 depend on modern ones (we use lex_destroy)
1383 - Better separation of SG and RL in the libs: remove all simulation code
1384 from libgras. As a result, this lib is now only 200k when stripped.
1385 Some of the xbt modules may also be duplicated (two sets and such) and
1386 should be cleaned/killed before SG3.
1387 - Insist on using xlC on AIX because of weird problems involving gcc there.
1388 - Cleanup the make remote stuff. This is now done by scripts
1389 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1390 mainly for our private use, but we're working on changing them to user
1393 - Bugfix: flush the socket on close only if there is some *output*.
1394 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1396 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1399 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1401 SimGrid (2.92) unstable; urgency=low
1403 Alpha 3 on the path to SimGrid 3
1407 - New! First try of benchmarking macros.
1408 - New! First try so that gras_stub_generator generate deployment and
1409 remote compilation helpers.
1411 - Bugfix: Initialization fix in msg_test.
1415 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1417 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1419 - API Change: gras_os_sleep to take the amount of seconds as a double.
1420 Accepting an int was error prone since it was the only location where
1421 seconds were coded as such. It leaded to damn rounding errors.
1422 - Bugfix: Hard to belive that timers ever worked before this.
1424 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1426 SimGrid (2.91) unstable; urgency=low
1428 Alpha 2 on the path to SimGrid 3
1432 - Bug fix in the lmm_solver.
1434 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1435 through the function MSG_paje_output.
1436 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1437 - It is possible to bound the rate of a communication in MSG with
1438 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1439 to write it in the changelog).
1440 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1445 - Various cleanups to the autotools stuff
1446 - Begin to move Gras examples to examples/gras/
1447 - Let make distcheck work again (yeah!)
1449 - documentation overhauled using doxygen.
1450 gtk-doc-tools is dead in SimGrid now.
1451 - Automatically extract all existing logging categories, and add the list
1452 to the documentation (long standing one, to say the less)
1454 - Cleanup the known architecture table. Reorder the entries to group what
1455 should be, and use a more consistent naming scheme.
1456 (some of the test dataset are still to be regenerated)
1457 - New! Allow library to register globals on each process just as userdata
1459 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1460 lookup time (for now).
1461 Use it in msg and trp.
1462 This cleans a lot the internals and helps enforcing privacy of the
1463 headers between the gras components.
1464 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1465 - Bugfix: gras_os_time was delirious in RL.
1466 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1468 - Reenable GRAS now that it works.
1470 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1472 SimGrid (2.90) unstable; urgency=low
1474 Alpha 1 on the path to SimGrid 3
1476 * It is a long time since the last release of SimGrid. I'm sorry about
1477 that but as I had told you, I was rewriting a lot of things. I apologize
1478 to those who had been reporting bugs to me and that I had not answered.
1479 If your bug is still in the new version, please tell me. Here is a
1480 summary of the main changes.
1482 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1483 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1484 portability, speed, and a lot more but you'll figure it out later.
1485 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1486 I wanted to release the new MSG as soon as possible and I have
1487 broken GRAS, which is the reason why, for now, only MSG is fully
1488 functional. A laconic description of these projects is available
1489 in the documentation.
1491 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1492 kernel optimized for our needs (called SURF but only the developers
1493 should use it). Hence, MSG is now roughly 30 times faster and I think
1494 that by rewriting a little bit MSG, I could event speed it up a little
1495 bit more. Beside the gain in speed, it is also much easier to encode a
1496 new platform model with SURF than it was with SG. More to come...
1498 * REVOLUTION 3: I have tried to change a little as possible the API of
1499 MSG but a few things really had to disappear. The main differences
1500 with the previous version are :
1501 1) no more m_links_t and the corresponding functions. Platforms are
1502 directly read from a XML description and cannot be hard-coded
1503 anymore. The same format is used for application deployment
1504 description. The new format is described in the documentation.
1505 Have a look in tools/platform_generation. There is a tiny script
1506 that converts from the old platform format to the new one. Concerning
1507 the application deployment format, parsing the old one is tricky.
1508 I think most of you should however be able to convert your files. If
1509 it is really an issue, I can write a C code that does the conversion.
1511 2) the toolbox tbx does not exist anymore. We now have a library
1512 with much more data-structures but without the hash-tables (we have
1513 dictionaries that are much faster).
1515 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1517 *****************************************************************************
1518 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1519 * don't feel like converting the oldies. (Mt) *
1520 *****************************************************************************
1523 Version 2.90: "the long awaited one"
1524 - Finished rewriting and debugging MSG. Rewrote the documentation.
1525 - disable GRAS for now since it needs to be ported to the newest SG
1528 - Finish the port to windows (using mingw32 for cross-compile)
1531 - Main loop and datastructures of SURF. A cpu resource object is
1532 functional. Surf can thus be used to create cpu's with variable
1533 performance on which you can execute some actions.
1535 2004-11-15 Martin Quinson
1536 - Port to ARM. Simply added the alignment and size descriptions. Should
1537 work, but the ARM machines are so slow that I didn't had the opportunity
1538 to 'make check' over there yet.
1540 2004-11-15 Arnaud Legrand
1541 - Trace manager now written. It uses a heap structure and is therefore
1542 expected to be efficient. It may however be speeded up (particularly
1543 when many events occur at the same date) by using red and black
1544 trees. One day maybe...
1545 - Max-min linear system solver written. It uses a sparse matrix
1546 structure taking advantage of its expected use. Most operations are
1547 O(1) and free/calloc are called as few as possible. The computation of
1548 the minimum could however be improved by using a red and black tree
1551 2004-11-03 Arnaud Legrand
1552 - Rename every gras_* function that was in xbt/ to its xbt_
1554 - Add a heap and a doubly-linked list to xbt
1555 - Added a dichotomy to the dictionaries. make check works as well before
1556 so I assume that the patch is correct. I do not know however if things
1557 run effectively faster than before now. :)
1559 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1560 SimGrid, and the version number is bumped to 2.x
1562 2004-10-29 Martin Quinson
1563 - Introduction of the remote errors.
1564 They are the result of a RMI/RPC on the remote machine.
1565 ErrCodes being scalar values, you can't get the host on which those
1566 errors did happen. Extending the error mechanism as in Gnome is possible.
1567 No idea yet whether it is a good idea.
1569 2004-10-28 Martin Quinson
1570 - Interface revolution: the Starred Structure Eradication.
1571 I used to do typedef struct {} toto_t; and then handle *toto_t.
1572 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1573 - ???_t is a valid type (builded with typedef)
1574 - s_toto_t is a structure (access to fields with .)
1575 - s_toto is a structure needing 'struct' keyword to be used
1576 - e_toto_t is an enum
1577 - toto_t is an 'object' (struct*)
1579 typedef struct s_toto {} s_toto_t, *toto_t;
1580 typedef enum {} e_toto_t;
1581 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1582 s_toto_t) is private.
1584 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1585 so that it can set the variable to NULL. It was so for dicts and sets,
1586 it changed for dynars.
1588 - Fix a bunch of memleaks in dict_remove
1589 - Fix a bug in sg/server_socket opening: it failed all the time.
1591 2004-10-07 Martin Quinson
1592 - Speed up dynar lookup operation a bit.
1594 gras_dynar_get is dead.
1596 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1597 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1598 (which gives you the address of the stored data).
1600 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1601 copy of the data using an affectation to do the job and not a memcpy.
1603 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1604 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1606 It does not really speedup the dynar test because they are
1607 setting elements all the time (and look them seldom). But the dict does
1608 far more lookup than setting.
1610 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1612 2004-10-05 Martin Quinson
1613 - Allow to (en/dis)able the cycle detection at run time.
1615 Whether we should check for cycle or not is now a property of each
1616 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1617 datadesc_cycle_unset allow to remove this property when previously set.
1619 Note that the cycle detection is off by default since it impacts the
1620 performance. Watch the data you feed GRAS with ;)
1622 This property is hereditary. Any element embedded in a structure having it
1623 set have it set for the time of this data exchange.
1625 You should set it both on sender and receiver side. If you don't set it on
1626 sender side, it will enter an endless loop. If you forget on receiver
1627 side, the cycles won't be recreated after communication.
1629 - Header reorganization.
1630 Kill gras_private.h, each submodule must load the headers it needs.
1632 2004-10-04 Martin Quinson
1633 - Interface revolution: do not try to survive to malloc failure.
1635 Now, gras_malloc and friends call gras_abort() on failure.
1636 As a conclusion, malloc_error is not a valid error anymore, and all
1637 functions for which it was the only gras_error_t return value are
1638 changed. They now return void, or there result directly.
1639 This simplify the API a lot.
1641 2004-09-29 Martin Quinson
1642 - Re-enable raw sockets.
1643 Created by gras_socket_{client,server}_ext;
1644 Used with gras_raw_{send,recv}
1647 It should allow to kill the last bits of gras first version soon.
1649 This is not completely satisfactory yet (duplicate code with
1650 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1653 - Simplify transport plugin (internal) interface by not passing any
1654 argument to _server and _client, but embedding them in the socket
1657 2004-09-28 Martin Quinson
1658 - Finish the port to AIX.
1659 autoconf was my problem (segfault within the malloc replacement
1660 function. No idea why)
1662 2004-09-16 Martin Quinson
1663 - Fix some size_t madness on 64bit architectures.
1665 2004-09-08 Martin Quinson
1666 - Reduce the number of system headers loaded, overload some more system
1667 calls (such as malloc to cast the result of the system one, and work
1669 - Fix and reintroduce the config support
1671 2004-09-07 Martin Quinson
1672 - Source code reorganization to allow Arnaud to surf all over there.
1673 - Allow to document the logging categories.
1674 - Remove all uppercase from logging categories and useless cleanup in names.
1676 2004-08-18 Martin Quinson
1677 Version 0.6.2 (protocol not changed; API changed)
1678 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1679 gras_error_t), and NULL when not found. Functions expecting a msgtype
1680 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1681 hopefully usefull message.
1682 - Portability to prehistoric sparcs again
1684 2004-08-17 Martin Quinson
1685 Version 0.6.1 (protocol not changed; ABI not changed)
1686 - prealloc some buffers to speed things up
1688 2004-08-11 Martin Quinson
1689 Version 0.6 (protocol not changed; ABI expended)
1690 - The parsing macro can deal with the references, provided that you add
1691 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1693 2004-08-09 Martin Quinson
1694 Version 0.5 (protocol not changed; ABI changed)
1695 - Allow to off turn the cycle detection code in data exchange at
1696 compilation time. It should be at run time, but I'm short of time (and
1697 the config stuff is still broken). That way, we keep dict out of the
1698 critical path, which is good because the performance is poor:
1699 - search not dichotomial yet
1700 - dynar give no way to access their content and memcpy everytime
1701 - In composed data description (struct, ref and so on), stop foolness of
1702 keeping the subtype's ID, but store the type itself. This keeps sets out
1703 of the critical path, which is good since they rely on dynar and
1704 dictionnaries. The only loose of that is that we cannot detect the
1705 redeclaration of a structure/union with another content (but I'm not sure
1706 the code detected well this error before anyway). We still can detect
1707 the redefinition discrepancy for the other types.
1708 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1709 it breaks the code because of type-punning used all over the place).
1710 This breaks on all non-gcc architectures (for now).
1712 All those changes (plus the buffer of last time) allow me to gain 2 order
1713 of magnitude on cruel tests consisting of 800000 array of integers on two
1714 level of a hierarchical structure (200 secondes -> 4 secondes)
1717 - the selector of reference must now return the type it points to, not
1718 the ID of this type.
1720 2004-08-06 Martin Quinson
1721 Version 0.4 (protocol changed; ABI not changed)
1722 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1723 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1724 - Implement a buffer transport, and use it by default (it relies on tcp in
1725 real life and on sg in simulation).
1726 That's a bit hackish since I had a new field to the structure to store
1727 its data without interfering with the subtype ones. Inheritance
1728 is tricky in C. And that's a kind of reverse inheritance with one class
1729 derivating two classes. Or maybe a game with java interfaces. Anyway,
1730 that's damn hard in C (at least).
1731 Moreover, I got tired while trying to ensure plugin separation and
1732 genericity in SG mode. MSG wants me to do weird things, so let's go for
1733 cruel hacks (temporarily of course ;).
1734 See comment in transport_private.h:71
1735 - do not include all the _interface headers in private but in the files
1736 which really need them (to cut the compilation time when they are
1739 2004-07-26 Martin Quinson
1740 Version 0.3 (protocol not changed; ABI changed)
1741 - Major overhault of the datadesc interface to simplify it:
1742 - shorted the function names:
1743 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1744 - add a trivial way to push/pop integers into the cbps without malloc.
1745 This allows to make really generic sub_type description, which simply
1746 pop their size of the stack.
1747 - add a function gras_datadesc_ref_pop_arr() which does what users want
1748 most of the time: Declare a dynamic array (which pops its size of the
1749 stack) and declare a reference to it. Poor name, but anyway.
1750 - kill the post-send callback, add a post-receive one
1752 2004-07-23 Martin Quinson
1753 Version 0.2 (protocol changed; ABI changed)
1754 - add some testing for cpbs in the test cases, and fix some more bugs.
1755 This invalidate again the little64 data file, since I cannot regenerate
1757 - remove an awfull optimization in the logging stuff, allowing me to:
1758 - understand it again
1759 - learn gcc how to check that the argument match the provided format
1760 - fix all errors revealed by gcc after that
1761 - internal keys of dict are not \0 terminated. Deal with it properly in
1762 loggings instead of segfaulting when the user want to see the logs :-/
1764 2004-07-22 Martin Quinson
1765 - Fix some stupid bug preventing cbps (callback postit) from working
1767 2004-07-21 Martin Quinson
1768 - Some documentation cleanups
1769 - remove the useless last argument of msgtype_declare
1770 - rename the Virtu functions to fit into the 'os' namespace
1771 - move headers src/include -> src/include/gras/ and stop fooling with
1772 gras -> . symbolic link
1773 - make distcheck is now successful
1775 2004-07-19 Martin Quinson
1777 - Build shared library also
1778 - Install html doc to the right location
1779 - stop removing maintainer files in make clean
1780 - build tests only on make check
1782 2004-07-13 Martin Quinson
1784 - No major issue in previous version => change versionning schema
1785 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1786 corresponding dataset.
1788 2004-07-11 Martin Quinson
1790 - Get it working with any kind of structure (we can compute the padding
1791 bytes remotely for all the architectures I have access to)
1792 - Implement the structure parsing macro (still not quite robust/complete)
1793 - Improvement to the remote testing toysuite
1795 2004-07-10 Martin Quinson
1796 [autoconf mechanism]
1797 - get ride of a bunch of deprecated macros
1798 - actually run the test for two-compliment, not only compile it :-/
1799 - test whether the structures get packed (and bail out if yes. Damn.
1800 Alignment is a serious matter)
1801 - test whether the structures get compacted (but respecting the alignment
1802 constraints of each types)
1803 - test whether the array fields of structures can straddle alignment boundaries
1805 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1806 (took me 2 hours to find that bug, looking at the wrong place)
1807 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1808 before sending/receiving and are used to compute the offsets of fields
1809 - Given that padding size depend even on compiler options, keep track of
1810 alignment and aligned_size only for the current architecture. Not a big
1811 deal since we send structure fields one after the other (seems
1813 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1814 program, let it work on linux/gcc/little32. portability todo.
1816 2004-07-08 Martin Quinson
1817 - import and improve remote compilation support from FAST
1818 - make sure make check works on half a dozen of machines out there
1820 2004-07-07 Martin Quinson
1821 Let's say it's version 0.0.3 ;)
1822 - Implement conversions (yuhu!)
1823 - Let it work on solaris (beside conversion, of course)
1824 - Stupid me, using rand() to generate the conversion datatests in not wise.
1826 2004-07-06 Martin Quinson
1827 - Let make dist work, since I'm gonna need it to compile on remote hosts
1828 - Let Tests/datadesc_usage write the architecture on which the file was
1829 generated as first byte.
1830 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1831 architecture descriptions.
1832 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1833 on those architectures.
1834 - Optimization: send/recv array of scalar in one shoot
1836 2004-07-05 Martin Quinson
1837 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1839 - Plug a whole bunch of memleaks
1840 - each process now have to call gras_{init,exit}. One day, their log
1841 settings will be separated
1842 - Continue the code factorisation between SG, RL and common in Transport.
1844 2004-07-04 Martin Quinson
1846 - Redistribution between SG and RL.
1847 We wanna have to accept in SG, so move accepted related parts of RL in
1848 the common part. (more precisely, the dynar of all known sockets is no
1849 more a static in transport.c, but part of the process_data)
1851 [gras_stub_generator]
1852 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1854 2004-07-03 Martin Quinson
1855 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1856 them all up in one shot)
1858 - Fix the ugly case of reference to dynamic array.
1859 - New (semi-public) function gras_datadesc_size to allow the messaging
1860 layer to malloc the needed space for the buffer.
1862 - gras_socket_close now expect the socket to close (and not its address to
1863 put NULL in it after it). This is because the socket passed to handlers
1864 is one of their argument (=> not writable).
1866 - propagate the interface cleanup from last week in datadesc, ie remove a
1867 superfluous level of indirection. User pass adress of variable
1868 containing data (both when sending and receiving), and not of a variable
1869 being a pointer to the data. Let's say that I like it better ;)
1870 The price for that is constructs like "int msg=*(int*)payload" in
1871 handlers, but it's a fine price, IMHO.
1873 - Let it work in RL (yuhu)
1875 2004-06-21 Martin Quinson
1877 - porting SG plugin and SG select to new standards (works almost).
1878 - plug memleaks and fix bugs around.
1881 - cleanup the prototype of data recv and force users to specify when they
1882 want to handle references to objects. Test case working even for cycles.
1883 - plug memleaks. Valgrind is perfectly ok with this.
1885 2004-06-12 Martin Quinson
1887 - cleanup the separation between plugin and main code in plugin creation
1889 2004-06-11 Martin Quinson
1891 - Reput hook for raw sockets, needed for BW experiments
1892 - kill a few lines of dead code
1893 [Data description] Interface cleanup
1894 - gras_datadesc_by_name returns the searched type or NULL.
1895 That way, no variable is needed to use a type desc once, which makes
1897 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1898 two parameters were strings (field name, type name), leading to
1900 [Dicos] Interface cleanup
1901 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1902 This is consistant with the dynar API.
1904 2004-04-21 Martin Quinson
1906 - Porting to new standards.
1908 - interface cleanup.
1909 There is no bag anymore, no need to take extra provision to mask the
1910 pointers behind "ID".
1911 Better splitup of functions between files create/exchange/convert.
1912 This is still a bit artificial since convert and receive are so
1913 interleaved, but anyway.
1915 - add a queued message list to procdata (the ones not matching criteria
1917 - factorize some more code between SG and RL wrt procdata
1919 - use gras_exit in example to track memleaks
1920 - get rid of gs_example now that GS is properly integrated into gras
1921 - update run_test to integrate the lastest tests (datadesc)
1923 - rename WARNINGn macros to WARNn since it prooved error-prone
1925 2004-04-19 Martin Quinson
1927 - register init/exit functions within gras module mechanism
1928 - send/receive function.
1929 Convertion is not implemented, but short-cutted if not needed.
1930 struct/array elements are sent one by one (instead of block-wise), but
1931 nobody really cares (yet). Get a prototype before optimizing.
1932 - tests (using a file socket) for DD send/receive on:
1933 - base types: int, float
1934 - array: fixed size, string (ie ref to dynamic string)
1935 - structure: homogeneous, heterogeneous
1936 - chained list, graph with cycle
1937 Believe it or not, valgrind is not too unhappy with the results. The
1938 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1942 - Bugfix when using a filename explicitely (instead of '-')
1944 2004-04-09 Martin Quinson
1946 - factorize more code between RL and SG in socket creation
1947 - Complete the implementation and tests of:
1949 o file (only in RL, and mainly for debugging)
1951 I lost 3 days to design a portable address resolver, and then decided
1952 that the prototype mainly have to run on my box.
1953 Addressing portability too early may be like optimizing too early :-/
1955 - use gras_init in the Tests instead of the crappy parse_log_opt
1956 (the latter function is removed)
1957 [Conditional execution]
1958 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1959 [Code reorganisation]
1960 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1961 Build examples against the RL library, since there is no way to disable
1962 its creation for now.
1964 For information, the beginning of coding on GRAS was back in june
1965 2003. I guess that every line has been rewritten at least twice since