1 SimGrid (3.5) unstable; urgency=low
4 * Asynchronous communications through the functions:
5 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
6 * New function: MSG_load_platform_script()
7 to make possible using a lua script instead of XML files to set up platforms
8 * New function: MSG_set_function
9 to associate functions to processes, used when bypassing the parser
10 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
13 * Add layer to set up environment directly from lua, without XML
14 * In GRAS, the effect of gras_stub_generator can be achieved through
15 lua too (check examples/gras/console/ping_generator.lua)
16 * Modify the entry point, to take into consideration when lua is used only as a platform generator
19 * Change the XML format. This is a very important modification. SimGrid
20 3.5 introduces a new hierarchical format based on the notion of
21 Autonomous Systems. Compatibility with old format is ensured
22 through the perl script provided in the install bin directory
23 bin/simgrid_update_xml.
24 It is now possible to build platforms with specific routing
25 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
26 connect several platforms together. We will try to provide soon
27 set of realistic platforms exploiting these properties (have a
28 look at examples/platforms/ for the moment).
29 * Take the opportunity to be a good XML citizen: rename link:ctn to
30 link_ctn and similar changes (also dealed with by simgrid_update_xml)
31 * Add a new routing scheme (rule-based) using regular expressions. It
32 enables to have an extremely low memory footprint when the
33 underlying routing is simple and can be compactly described. You
34 need to have libpcre4-dev (perl regular expressions) installed if
35 you want to use this routing scheme.
36 * Revive the cluster TAG and allow to easily and efficiently (both in
37 term of memory and speed) connect clusters together. Have a look
38 at teshsuite/simdag/platforms/ to see how this can be done. With
39 this tag, you can create clusters with thousands of tasks at no
40 cost (have a look at examples/platforms/).
41 Note: clusters are implemented as ASes, so there is no need for an
42 enclosing AS tag if you have only one cluster in your platform.
43 * Add new callbacks to the CPU & netwok models to create resources
44 no need to pass through the parser callbacks to do so.
45 * Add new function to create routes table via the routing model
46 * Add generic functions in the public interface that allows the user to
47 call SURF 'create_resource' methods from a lua script.
48 * Add a new model (enabled through command line --cfg=network/model:SMPI)
49 that uses a piecewise linear approximation to produce better
50 results when exchanging small messages.
51 * Add a new parameter to handle correctly full duplex link and account
52 for interferences between uplink and downlink communications
53 (activate with --cfg=fullduplex:1).
56 * New function: SIMIX_process_set_function() called by MSG_set_function
57 * Change the underlying waiting queue in semaphores so that a process
58 can wait on several of them simultaneously (as in waitany).
59 * Fix the way to handle tokens in semaphores so that all access patterns
60 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
61 * kill the dirty pimple SIMIX_message_sizes_output()
62 Please use (proper) visualization instead
65 * Fix a bug in the management of tasks of size 0 in the surf network
66 models. This problem was only visible with SIMDAG and you should
67 thus disregard results produced with earlier versions if you
68 relied on this feature (some tasks were blocked because of this).
69 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
70 even though your applications were doing only point-to-point
71 communications and sequential computations. Now you can really use any
72 model you want (of course, if you create real parallel tasks, which are
73 not implemented in most models beside ptaskL07, this will abort).
74 * Change the way a task is considered as ready. Instead of removing
75 dependencies when a task is done, a counter is decreased. This way,
76 it is always possible to reach ancestors thanks to the
77 SD_taks_get_parents function (even after the end of the simulation.)
78 * Change the level of some logs in sd_global from INFO to VERB. The
79 corresponding tests are now less verbose
80 * Change the return type of SD_Simulate from (SD_task_t*) into
81 xbt_dynar_t. This function was in handling a dynar internally and
82 converted it into a NULL terminated array for historical reasons.
83 * Rename the SD_READY (all dependencies are satisfied and task is
84 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
85 dependencies are satisfied) state.
86 This prevents a confusion between the notion of "ready to schedule"
87 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
88 simulated" (SD_RUNNABLE) used by the simulation kernel.
89 * Add an example that scheduling a DAX on an heterogeneous platform
90 using a Min-Min strategy
91 * new function SD_dotload(char*) to load a DAG described in dot
92 format. This loader and the corresponding examples require the
93 installation of the graphviz library.
94 * new function SD_workstation_get_current_task() that returns the kind
95 of task currently running on a workstation in the sequential access
97 * Task rate should be -1 by default
98 Using 0 as before caused TCP_gamma bound to be ignored.
99 This change modifies (fixes) almost every simdag timings.
100 * Raise some warnings when unexecuted remains at the end of the
101 simulation. This is usually caused by cycles in the DAG.
104 * New data container: setset (set of sets of elements)
105 * New module: mmalloc (mapped malloc, allowing to have several
106 independent segments of malloc)
107 * New function: xbt_dict_cursor_set_data()
108 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
109 * New function: xbt_dynar_is_empty()
110 * New function: xbt_fifo_get_last_item()
111 * Bug fix in xbt_dynar_shrink(): use the right element size
112 * Multiple bug fixes in semaphores which were wrongly implemented
113 * Use library init/fini functions for our initialization.
114 - you can use logs and other feature as soon as you want in your
115 code (even before the xbt_init / MSG_init)
116 - xbt_exit is now a no-op and produce a warning when used.
119 * New function: gras_msgtype_get_name()
120 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
121 * Port GRAS to new SIMIX deployments. This allows gras users to
122 benefit from the lastest improvement to the simulation kernel.
123 * Kill measurement sockets for now. If you rely on them, sorry. This
124 release is not for you. This feature will be reintroduced in the
125 future, but we cannot delay the release any further.
126 * Finally fixed the pmm example. It should not randomly fail anymore
129 * New Model SMPI (three-interval linear regression for correction factors)
130 See RR-7426, available at http://hal.inria.fr/inria-00527150
131 * Ability to use fortran MPI code (through f2c, automatically privatized)
132 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
133 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
134 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
135 * See also src/smpi/README
137 Build chain: bug fixes and overall polishing
138 * Fix the 'make install' target.
139 No need to use 'make install-simgrid' anymore
140 * Introduce a 'make dist' target compiling a *source* archive
141 'make package' compiles a binary archive
142 * Compile java files only on need
143 * Add --cd and --setenv command line options to tesh
145 * Out of source builds are not fully supported yet, but we are close
146 * Enable supernovae and optimization flags by default for our users
147 * Cmake is now stable enough so that we killed the autotools
149 Platforms: Add some more examples in examples/platforms
150 * Grid'5000: see www.grid5000.fr
151 * *_30000_hosts.xml: various huge files [mainly scalability testing]
155 - tracing API for MSG changes: TRACE_start has no
156 longer parameters, the trace file is supplied through
157 an option passed to simulator. Categories declaration
158 and utilization remain the same. The following options
160 --cfg=tracing/filename:msg.trace
161 --cfg=tracing/platform:1 (categorized resource use)
162 --cfg=tracing/msg/task:1 (task creation)
163 --cfg=tracing/msg/process:1 (process creation, migration)
164 --cfg=tracing/msg/volume:1 (volume of msg send/recv)
165 --cfg=tracing/smpi:1 (smpi interface tracing)
166 - examples of examples/msg/tracing updated
168 * Tracing the MPI interface implemented by SMPI
169 - Collective operations are traced with states
170 - Point-to-Point operations are traced with states/links
171 - Tracing activated by a parameter "-trace filename" passed
172 to smpirun during execution (considering that simgrid
173 is compiled with tracing enabled)
174 - To run the simulation with gdb, the simulator
175 accepts --cfg=tracing/smpi:1 to trace SMPI
176 - tesh files to check if smpi tracing is ok
177 - See examples/smpi/NAS/DT-trace
179 * GTNetS tracing re-worked
180 - adaptation to the tracing system of GTNets to cope
181 with modifications regarding the fullduplex mode
182 - new tesh files to check if gtnets tracing is ok
184 * New feature to any SimGrid-based simulator: Model-Checking
185 Check SIN#1 for more details.
187 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
189 SimGrid (3.4.1) stable; urgency=low
191 The "Polishing easter eggs is probably a good idea" release.
192 This is a bug fixes release only.
196 * Fix a bug preventing the tasks from begin garbage collected.
199 * Fix a bug occuring when a host involved in a communication fails.
200 This was not detected properly by the other peer involved in the
201 communication. Now, it's reported as a network error.
204 * Warn the user about loop dependencies in data flow of DAX files
205 * Obey the control-flow dependencies of DAX files
208 * Add option "enable_smpi" allowing to not compile SMPI.
209 Probably useful for the (Mac) users experiencing a build error here
210 * Improve the detection of lua5.1 and ruby1.8
212 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
214 SimGrid (3.4) stable; urgency=low
216 The "Easter in Cargese" release. Also known as (major changes):
218 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
219 ~> bindings were greatly improved
220 ~> new tracing infrastructure for better visualization introduced
222 * the "Welcome to configury modernity" release.
223 ~> we switched from autotools to cmake, and improved our cdash
226 A more detailled list of changes follow (full detail in svn log).
228 Java Bindings: Various Cleanups
229 * (install java-gcj-compat-dev on debian-like to use them)
230 * Remove put/get: no need to export deprecated interface in Java
231 Use send/receive instead.
232 * Cleanup the examples and add a README per directory
233 * Remove example autoDestination (that's the only way to go now)
234 * Remove example explicitDestination (was a plain copy of basic)
235 * Make JniException a runtime exception, so that there is no need to
236 declare the fact that you may encounter such a beast. I guess that
237 nobody will ever want to survive such error.
238 * Create specific errors for each MSG case of failure:
239 host failure, transfer failure, timeout, task cancelled
240 * Cleanup the exceptions that may get thrown by each function
241 * Other internal cleanups in Java bindings. Performance still bad :/
242 Ruby and Lua Bindings: create them
243 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
244 * That's new and great, you should try them out.
245 Same functionalities than Java bindings, only even less polished
247 * Kill the useless "rate" argument of SD_task_get_execution_time()
248 Everyone used to provide -1 as a value, it was not used, and the
249 semantic of a possible use wasn't even clear.
250 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
251 either as comm costs or compute costs to mean that there is no
252 such thing for that specific task.
253 * Add a SD_task_set_name() function
254 * Fix SD_task_unschedule() on typed tasks
255 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
256 * In DAX loader, accept useless 'level' attributes to <job> since
257 LIGO DAGs have them (seem to be to ease graphical representation).
259 * Add an example masterslave_mailbox.c using send/receive and not
260 the deprecated put/get interface.
261 * Kill the MSG_paje_output() function. It's a noop since 2 years.
262 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
264 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
265 (declare MSG_USE_DEPRECATED to still have the old name)
266 * Add a MSG_task_set_data() function
267 * About trace replay (see examples/msg/actions):
269 - Allow to work with splitted trace files for each process
270 Give the specific trace file as argument of each process,
271 and call MSG_action_trace_run(NULL)
272 You can still have one merged file for all processes.
273 - Fix implementation of collective operations
274 * Allow task_execute() on 0-sized tasks (closes #10063)
276 * This is the first release of SimGrid where SMPI is not considered
277 beta anymore (even if some corners should still be improved)
278 * Port over the new SIMIX_network submodule (internal refactoring)
279 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
280 * Implement more missing elements of the standard:
282 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
284 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
285 MPI_Scan() MPI_Get_processor_name()
286 - Added implementation of missing case for Alltoall (warning: it's
287 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
288 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
289 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
290 SMPI_MPI_Allgatherv()
292 - MPI_Waitsome() was broken
293 - Allow relative includes in smpicc
294 - Command line cfg argument 'reference_speed' was ignored...
295 - Some functions did not properly lead to auto-benching of user code
296 - smpicc passes -O2 by default (just like openmpi one)
298 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
299 * Bug fixes about timeouts during communications
300 * add SIMIX_message_sizes_output() as a pimple to write to file the
301 amount of messages per size. Use gnuplot to get histogram.
302 Pimple because that's the only user-visible function of simix,
303 defined directly in xbt.h (irk, sorry)
305 - Add a SIMIX_sem_get_capacity() function
306 - Fix interactions with processe resume/suspende
307 - release_forever() was stupidly broken
308 - Fix SIMIX_display_process_status() for processes in a semaphore
309 - Make SIMIX_sem_block_onto() user-visible
310 * Refactoring context stuff:
311 - Use pseudo-OOP for better modularity
312 - reimplement SIMIX_process_kill() without process_schedule() so
313 that the latter can take as invariant that it is called from
315 - Merge context_start into context_new for sake of simplicity
317 * Add a Vivaldi network model, coded live during SUD'10 ;)
318 * Rename configuration variables to start a hierarchy:
319 o cpu_model -> cpu/model
320 o network_model -> network/model
321 o workstation_model -> workstation/model
322 * New configuration variables:
323 o network/bandwidth_factor: correction to bandwith
324 o network/latency_factor: correction to latency
325 o netwotk/weight_S: correction to the weight of competing streams
326 * Add a long description to the models, that users can see with such
327 argument on the command line: --cfg=cpu/model:help
328 * --help-models display the long description of all known models
330 * config: add the ability to set a default value after registration
331 Does not override any previously set value (e.g. from cmd line)
332 * dict: allow to have integer key and data.
333 When so, you need to use the following functions
334 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
335 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
336 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
337 In contrary to regular dicts, the key is not malloced before copy.
338 Mixing scalar and regular elements in the same dict is not tested
340 * Allow to use xbt_dynar_shrink() to expend the dynar instead
341 Tracing for Visualization:
342 * SimGrid is now instrumented in order to generate a trace file for
343 visualization analysis: to use it, need to compile SimGrid with the
344 "tracing" option enabled, and instrument the program using SimGrid with
345 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
346 (among other functions).
347 * The instrumentation only traces the platform utilization for now
348 * Documentation to use the tracing functions and how to analyze the
349 traces with the Triva tool is written.
350 * More information about: SimGrid FAQ (in the section Tracing Simulations
353 * We moved to cmake as default build system. Autotools support will
354 be dropped soon. Check the FAQ for more info about how to use it.
355 * Greatly improved our cdash/ctest interactions
356 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
357 * Added memory checking tests with valgrind; lot of memleak fixing.
358 This may be the first release of simgrid with so few memory issues
359 * Added code coverage tests.
360 Our coverage is still improvable, but at least we see it on cdash.
362 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
364 SimGrid (3.3.4) stable; urgency=low
366 The "Desktop Grid needs love too" release (also called Xmas release).
369 * Major speedup in the maxmin system solving by using lazy evaluation
370 Instead of solving completely the maxmin system at each iteration,
371 only invalidate (and recompute) the modified parts.
372 This new feature is enabled in default models but you can try to
373 turn it on with "--cfg:maxmin-selective-update=1" for other models.
374 * Cas01 IMproved as default CPU model
375 This CPU model is the same Cas01 model, but it uses the
376 maxmin-selective-update flag and a heap structure to manage
377 actions on SURF kernel.
378 It reduces the complexity to find the next action to finish and,
379 consequently, it's faster than the old Cas01.
380 This is the new default CPU model (Cas01).
381 * Rename the old Cas01 model to Cas01_fullupdate
382 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
383 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
384 * CpuTI (CPU Trace Integration)
385 A new CPU model whose objective is simulate faster when using
386 availability trace files.
387 Instead of using a full featured, over engineered maxmin system for
388 CPU modeling, this model does the pre-integration of traces files
389 to calculate the amount of CPU power available, and so, executes
390 faster than the old CPU models.
391 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
392 * Use LV08 as default network model since it gives better accuracy
393 for small messages and shouldn't change things for big ones.
394 Use --cfg=network_model:CM02 to get the previous behavior.
397 ******************************************
398 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
399 ******************************************
400 * The new CPU model may changes simulations!
401 The point is that events occurring at the exact same timestamp
402 are not scheduled in the same order with the old and new
403 version. This may be enough to completely change the execution
404 of simulations in some cases.
405 * The new network model will change simulations!
406 This new model is more realistic than the previous one, so you
407 should consider redoing your old experiments with this model.
408 Sorry for the inconvenience.
411 * Introduce the supernovae compilation mode
412 When compiled that way, the whole SimGrid (or almost) is put in a
413 single compilation unit and compiled in one shoot.
414 This is to help gcc which has difficulties to inline stuff from one
416 The speedup seem to be above 15%, althrough more tests are needed on
417 amd64 to confirm that gain.
420 * Port of MSG's mailbox on top of SIMIX network
421 The put/get mechanism was greatly simplified on the way.
424 * New SIMIX network module. Provides:
425 - Mailbox: rendez-vous mecanism to find with who you want to speak
426 - Synchronous send/recv: easier and hopefully faster since the
427 logic is handled in the maestro process directly now
428 - Asynchronous send/recv: you dreamt of it? It's here now
429 Too bad that nobody cared enough to propagate the change to MSG.
430 * Add semaphores as SIMIX synchronization mechanism.
433 * new function SD_daxload(char*) to load a DAX file
434 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
435 * Introduce typed tasks. Specify its kind and cost at creation.
436 At scheduling, just give where it should be placed, and the cost
437 for each involved resource is automatically computed.
438 Existing constructors so far (more to come of course):
439 - SD_task_create_comm_e2e() for end-to-end communication
440 - SD_task_create_comp_seq() for sequential computation
441 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
442 * new function SD_task_dump() for debuging display
443 * new function SD_task_dotty(task,FILE*) writing to file the info
444 about the task in dotty format
445 * SD_task_dependency_exists() can now cope with having one of its
446 arguments NULL. If so, it tests whether the other argument has any
448 * Add getters on list of preceding/following tasks:
449 SD_task_get_parents(task) and SD_task_get_children(task)
450 * Add getters on amount of workstations and list:
451 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
452 * Add getter on task kind: SD_task_get_kind(task)
453 * Update the start_time and finish_time of tasks on completion/failure
454 * Bugfix: Remove task from state swags when destroyed
457 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
460 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
461 dynar with one separate thread per value of the dynar.
462 * Change the prototype of xbt_thread_create(), sorry.
463 Added a boolean parameter indicating whether we want to join this
464 thread (used in SG only for now)
465 * Implement xbt_thread_join and xbt_thread_yield in SG also.
468 * GTNetS wrappers should now be usable again (and betterly tested too)
469 * Fix a major regression from 3.2 where the timeout provided to
470 MSG_task_put_with_timeout() was used as absolute time before which
471 the comm should be done.
472 * Start to fix the <cluster> tag.
473 - Internal links should be good now (beside of the loopback, which
474 use the private link instead)
475 - paths to the external world is still rather broken
476 - the <route:multi> tag is just broken. Actually that's brain-dead.
477 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
478 to make it less stupid
479 ** Check your platform with teshsuite/simdag/platforms/flatifier **
480 * Fix a source-level compatibility glitch from 3.2: after defining
481 MSG_USE_DEPRECATED, you can use the old name
482 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
483 * Allow to compile from the SVN with automake 1.11
484 * Fix some problems when using the "start_time" tag in deployment XMLs.
485 * Fix #8569: XBT/synchro.h has redundant declarations
486 * Fix #8563: MSG return values and exceptions
487 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
488 * Integrate patch #8636: Obey DESTDIR when installing documentation.
489 Thanks to Robson Peixoto.
490 * Fix a vicious bug in dictionaries inducing that some elements were
491 not freed on xbt_dict_free()
493 Portability report of this version:
494 * Main portability targets:
495 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
497 - mac leopard on i386
498 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
499 but nothing critical.
500 * Other platforms: windows, AIX and others were not tested for this release
502 Timing report of this version:
503 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
504 scenario parameters). From 8h to a few seconds in desktop grid settings.
505 * Supernovae brings about 25% speedup on i386.
507 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
509 SimGrid (3.3.3) stable; urgency=low
511 The "Need for Speed" release.
513 The timings done to validate the 3.3.2 were faulty.
514 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
516 The problem was a conversion from a manually handled vector to
517 xbt_dynar_t on the critical path.
518 xbt_dynar_foreach calls functions, inducing stack management crap.
520 We inlined these functions and xbt_dynar_foreach is now breath taking.
521 We also inlined xbt_swag_belong on the way.
523 Here are some approximate speedup measurements (on master/slaves
524 simulations lasting between 10s and 20s each):
525 3.3.1 -> 3.3.2: about same performance
526 3.3.2 -> 3.3.3: 40% speedup
527 3.3.1 -> 3.3.3: 40% speedup
528 3.3.1 with inline patch -> 3.3.3: 30% speedup
530 Our reading is that the refactoring which occurred in 3.3.2 made us
531 suffer much more from the xbt_dynar_foreach low performance, but
532 once we solved this, this refactoring proved to be very performance
533 effective. From the 40% speedup, somehow, 10% are due to the
534 inlining and 30% to the refactoring.
536 That's a pitty that gcc cannot inline functions placed in other files
537 alone. We have to choose between:
538 - break the encapsulation (by putting private data structures and
539 accessors in headers files to help gcc)
540 - live with low performance
541 - switch to a decent compiler such as icc (not quite possible).
543 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
545 SimGrid (3.3.2) stable; urgency=low
547 The "Simplicity does not preceed complexity, but follows it" release.
549 The main contributors of this release were (lexical order):
550 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
553 * Extract the routing logic into its own object.
554 (was dupplicated in network.c and workstation_LV07.c;
555 Allows to implement other ways of storing that info)
556 => kill now useless network_card concept
557 - Use dynar to represent routes (instead of void** + int*)
558 - kill link_set (use surf_network_model->resource_set instead)
559 - Add a command-line option to choose the routing schema to use
560 - Add three new models:
561 * Floyd (shortest path computed at initialization)
562 * Dijikstra (shortest path recomputed all the time)
563 * Cached Dijikstra (shortest path computed on need)
564 All these models where contributed by Silas De Munck, and are
565 described in his ICCS09 paper.
567 * Simplify model declaration
568 (less redirections, less function to write when defining a model)
569 - Factorize stuff between models:
572 surf_model_resource_set(model)
573 surf_model_resource_by_name(model, name)
574 - Unify the types of models in s_surf_model_t (using an union)
575 - Embeed fields of common_public directly into s_surf_model_t
576 - Rename model methods:
577 action_free ~> action_unref
578 action_change_state ~> action_state_set
579 action_get_state ~> action_state_get
580 - Change model methods into functions :
581 (model)->common_public->action_use ~> surf_action_ref
583 * Implement a generic resource; use it as ancestor to specific ones
584 (allows to kill duplicated code in models)
585 Drawback: timer command don't need no name nor properties;
586 workstation_CLM03 don't need no properties
587 (but I guess we can live with those few bytes wasted)
589 * Improve the action object model
590 - implement a constructor avoiding dupplicated code about field
591 initialization in generic_action part.
593 * Kill the SDP model: it has an external dependency, is deprecated
594 in flavor of modern lmm models, and didn't compile since a while
597 * Relocation of the context module from XBT to SIMIX.
598 (the context were decoupled from the simix processes, duplicating a lot of code)
599 => a lot of code was factorized
600 - less overhead is introduced during scheduling
601 - simpler API for the context factory
602 - the logic for process creation,destruction and manipulation was simplified
603 * Simplification of the s_smx_process_t data structure.
604 => accesing the simix level data associated to a process is faster now,
605 and the code is a lot more readable.
608 * Implement some more MPI primitives:
609 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
610 -implementation: Bcast: flat or 2-ary tree (default),
613 Allreduce: Reduce then Bcast
614 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
615 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
616 Alltoallv: flat tree, like ompi
618 * Add support for optimized collectives (Bcast is now binomial by default)
619 * Port smpirun and smpicc to OS X
622 * Kill SD_link_get_properties: hard to maintain and makes very little sense
623 Shout out if you used it.
626 * Display the list of still queued messages in SG mode when existing
630 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
631 * Add xbt_graph_node_get_outedges() [Silas De Munck]
632 * Add xbt_str_from_file(FILE*)
633 * Add xbt_dict_get_key achieving a linear reverse search
634 * Remove the context module
636 Portability report of this version:
637 * Main portability targets:
638 - Linux(debian)/x86/context
639 - Linux(debian)/x86/pthreads
640 - Linux(debian)/amd64/context
641 - Linux(debian)/amd64/pthreads
642 On these, we still have the eratic breakages of gras/pmm and
643 amok/saturate_sg reported in previous version. We still think
644 that the tests are the cause of the fault, not the tested code.
646 - Mac OSX Leopard/x86/context
647 Still false negative in tesh autotesting.
648 Smpi still fails, but this time because readlink does not accept -f
649 Everything seems to work properly beside of that.
652 - AIX version 5.3 (only tested contexts this time)
653 Smpi still fails there because mktemp is not installed.
654 Everything seems to work properly beside of that.
656 I managed to compile it for the first time, but several breakages.
657 Won't delay the release for this exotic platform.
659 * Windows: it's still lagging behind. If you want to help, please
662 Timing report of this version:
663 This version seem to be more than 5% faster than 3.3.1 (on linux
664 64bits with contextes). The gain is less than expected, we are
665 investigating this for next release.
667 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
669 SimGrid (3.3.1) stable; urgency=low
672 * Implement a --cfg-help to show existing configuration variables
673 * Build chain do not require doxygen in maintainer mode
676 * fix a bug on struct sizeof computation, which prevented the
677 exchange of arrays of structs in some conditions
678 - added a regression test about this in datadesc_usage
679 * Allow the exchange of 0-long dynamic vectors.
680 - for that, use -1 as indicator of dynamic size instead of 0
681 - This implied to change any size from unsigned long to long,
682 reducing a bit communication abilities, but I guess that with
683 64bits being quite common, this is more than enough.
684 - This also induce a protocol change, thus bumping network protocol
685 version from 0 to 1 (if we have external users, we have to get
686 clean on that point too ;)
687 - added two regression tests about this in datadesc_usage
688 * Be more verbose when propagating local exceptions
689 This helps debugging.
690 * Display the status of simulated processes when receiving SIGINT in
694 * Allow to control the simulation from a trace file.
695 New functions MSG_action_register() and MSG_action_trace_run()
696 The first one allows to associate a function execution to each
697 kind of action while the second one parses a trace file and
698 triggers the corresponding actions within the system.
699 For now, only a toy example is provided in examples/msg/actions
700 * Add an exemple of process migration in examples/msg/migration
701 * Fix a bug in task exchange which broke MSG_task_get_sender()
702 Add a teshsuite regression test for that.
703 [Bug: if MSG_task_get_sender() is called after sender exit,
705 * Fix a bug which prevented suspend/resume to work properly
706 * Display the status of simulated processes when receiving SIGINT
707 This fixes a regression of v3.3. due to the introduction of SIMIX
708 * Bug fixing in failure management:
709 - trace could not start by a failure at time 0
710 - failure during communications were not working
713 * Add SIMIX_process_set_name() to change the name of the current
714 process in the log messages.
715 * Store smx_hosts in a dict since we only retrieve them by name
716 * Move the configuration infrastructure to surf
719 * Move the configuration infrastructure to surf
722 * Massive internal cleanups:
723 - Store internal structures on processes instead of hosts (allows
724 to have more than one process per host, in addition of being more
726 - Cleanup the initialization/finalization process
727 - Kill a whole bunch of unneeded synchronization:
728 processes run in exclusive manner within the simulator
729 - Move queues from global tables to process data fields
731 - now accept -platform and -hostfile arguments
732 - Pass the right rank value to processes according to the hostfile
733 * Compile the examples by default, and use them as regression tests
734 * Implement MPI_Wtime()
735 * Change the reference speed to a command line option
738 * TCP_gamma can now be specified as command line option using
739 --cfg=TCP_gamma:10000000.0
740 * Change the --surf-path cmd line option into --cfg=path:
743 * Also include strbuff from xbt.h public header
744 * xbt_ex_display(): do not free the exception after displaying
745 This allows to do more with the given exception afterward.
746 Users should call xbt_ex_free() themselves.
750 Portability report of this version:
751 * Main portability targets:
752 - Linux(debian)/x86/context
753 - Linux(debian)/x86/pthreads
754 - Linux(debian)/amd64/context
755 - Linux(debian)/amd64/pthreads
756 These targets fail about 1/10 of times on gras/pmm, but we believe
757 that this is because of the test, not because of simgrid.
758 amok/saturate_sg fails even more rarely, and the test may not be
761 - Mac OSX Leopard/x86/context
762 The test suite still spits tons of errors because some obscure
763 force prevents us from removing the temporary directories
764 arguing that they still contain some metadata I've never heard of.
765 Smpi fails because seq is not installed.
766 Everything seems to work properly beside of that.
769 - AIX version 5.3 (both contexts and pthread)
770 Smpi still fails there because mktemp is not installed.
771 XML inclusions seems rosty on AIX.
773 * Windows: it's still lagging behind. If you want to help, please
776 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
778 SimGrid (3.3) stable; urgency=high
782 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
785 * Introduce the SIMIX module: factorize code between MSG and GRAS.
788 Until now, GRAS were using MSG as an interface to SURF. It was
789 quite difficult because both interface have several differences
790 (MSG channels vs GRAS sockets were the most notable point).
792 This also opens the gate to SMPI (which should occur soon) and speed
793 up simulations by to 40% (even if it were not the main goal).
795 **************************************
796 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
797 **************************************
798 The point is that events occuring at the exact same timestamp are
799 not scheduled in the same order with the old and new version. This
800 may be enough to completely change the execution of simulations in
801 some cases. Sorry for the inconvenience.
803 * Cleanup and upgrade the XML format to push further scalability
804 issues (check http://hal.inria.fr/inria-00256883/ for more info)
806 * Improve the testing infrastructure with tesh. Now a very large part of
807 the code is tested not only by being run but also by checking that the
808 output match an expected output [Mt].
810 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
811 is really less memory-demanding, which should allow you to use
812 larger files in SimGrid [AL].
814 * Inform valgrind about our contextes, so that it becomes usable
815 with the default (and more effecient) version of SimGrid
816 [contributed by Sékou Diakite, many thanks]
819 * Introduce a listener thread in charge of receiving incomming
820 messages from the network. It allows to overlap communication and
821 computation but most notably, it removes some stupid deadlocks due
822 to the fact that so far, a process could not send and receive at
823 the same time. This made most non trivial communication schema
825 * Convert the PIDs from long int to int to match the MSG ones (and
827 * New function: gras_agent_spawn() to launch a new process on
828 current host. Only working in simulation for now. [Mt]
829 * New function: gras_os_hostport() returning a constant form (ie,
830 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
833 * Make the backtrace of exceptions more human readable [Mt]
834 * New module: xbt/str [Mt]
835 a ton of string utility functions (split, join, printf to a newly
836 allocated buffer, trim, etc)
837 * New module: xbt/hash [Mt]
838 SHA1 hashing algorithm (more to come if needed)
839 * New module: xbt/synchro [Mt]
840 synchronization tools (mutex and conditions) working the same way
841 in simulation and in real life (mainly useful for GRAS, but not
843 * New module: xbt/queue [Mt]
844 classical producer/consumer synchronization scheme
845 * xbt_dynar_new_sync() creates a synchronized dynar. All access
846 (using the classical functions will get serialized) [Mt]
847 * Make dictionary internal table dynamic. No need to specify its size
848 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
850 * Make sure the log channels are organized as a tree under windows
851 (because of ANSI C compatibility issue, any channel were child of
855 * Cleaned many thing in surf and fixed a few bugs [AL].
856 * Add a nice command line configuration mechanism to compose models [AL].
857 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
858 the previous one (KCCFLN05). It relies on something that looks like
859 a max-min sharing mechanism but cannot be written as such. A new solver
860 was thus designed [AL].
861 * Add a new solver to lmm. Based on Lagrange optimization and
862 gradient-based descent, it enables to efficiently maximise systems s.a
864 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
866 This solver enables to propose two new network models for TCP Reno and
867 TCP Vegas based on Low's work. These models still need to be fully
868 tested though [Pedro Velho].
871 * Bug fix in SD_simulate. Now the time bound given as argument is
873 * Use the new parallel task model (ptask_L07) as default.
874 * Use the SURF command line configuration mechanism.
875 * 0-size tasks (for synchronization) should now work.
877 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
879 SimGrid (3.2) stable; urgency=high
883 We still experience issues on this platform, but we believe that at
886 GRAS API BREAKAGE (for simplification purpose, sorry):
887 * the gras_msgtype_by_name is not used anymore. Instead of
888 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
889 you can write (and must)
890 gras_msg_send(toserver, "request", &request);
891 - If you still want to pass a gras_msgtype_t to the function (to cache
892 the type and avoid the lookup time), use the gras_msg_send_() variant.
893 - Impacted functions:
894 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
895 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
896 * The callbacks are now expected to return 0 when everything went well
897 (just like the main() function)
899 GRAS new features and improvements:
900 * New module mecanism where user code can use per process globals [Mt]
901 This is similar to gras_userdata_*() functions, but for libraries. It
902 factorize some code developped over and over in the examples and AMOK.
903 It has still to be documented and used (only amok/peermanagement is
905 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
906 loss when they were small enough to fit into the buffer and sent quickly
907 enough so that they can all get received in one shoot.
908 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
909 if not found. Use the *_or_null() variant for the old semantic.
910 * In gras_msg_handle, do not discard messages without callback.
911 They are probably messages to be explicitly awaited later (ie, proofs of
912 mis-synchronization in userland since they are sent before being awaited)
914 * gras_socket_meas_send/recv: semantic changed!
915 The numerical arguments used to be (1) the total amount of data to send
916 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
917 messages. This was need for the fool willing to send more than MAXINT
918 bytes on quite fat pipes.
921 * Do really rename the hostmanagement module to peermanagement. [Mt]
922 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
923 but this is rather new and this was documented in the module
924 documentation (poor excuses, I admit)
925 * Bandwidth measurement semantic changed! This follows the changes to
926 gras_socket_meas_send/recv explained above.
929 * A sequential mode has been added to the workstations. When a workstation
930 is in sequential mode, it can execute only one task, and the other tasks
931 are waiting in a FIFO. [Christophe Thiery]
934 * The KCCFLN05 workstation model now handles parallel tasks. It is the
935 model for SIMDAG. [Christophe Thiery]
936 * Bug fix in the maxmin solver: Some values were close to 0 instead of
937 equal to 0, which caused some bad behaviors in
938 saturated_constraint_set_update. I now use a threshold mechanism like in
942 * When running manually src/testall, you select specific units [Mt]
943 testall is the result of our cunit mecanism, and should replace all
944 the scripty thingy around since bash don't run easily on billware.
946 * A mallocator system has been added. [Christophe Thiery]
947 Mallocators allow you to recycle your unused objects instead of freeing them
948 and allocating new ones.
950 Documentation update:
951 * FAQ reworking + New FAQs:
952 - "Valgrind spits tons of errors!" [Mt]
953 - "How to repport bugs" [Mt]
954 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
955 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
956 - Communication time measurement within MSG [AL]
957 - I experience weird communication times when I change the latency [AL]
960 - an introduction to the framework and to the used communication model
961 - an initiatic tour introducing the most proheminent features:
963 . Lesson 0: Installing GRAS
964 . Lesson 1: Setting up your own project
965 o Part 2: Message passing
966 . Lesson 2: Exchanging simple messages
967 . Lesson 3: Passing arguments to the processes (in SG)
968 . Lesson 4: Attaching callbacks to messages
969 . Lesson 5: Using globals in processes
970 . Lesson 6: Logging informations properly
971 . Lesson 7: Using internal timers
972 . Lesson 8: Handling errors through exceptions
973 . Lesson 9: Exchanging simple data
974 . Lesson 10: Remote Procedure Calling (RPC)
975 . Lesson 11: Explicitely waiting for messages
976 . Recapping of message passing features in GRAS
977 - A HOWTO section containing:
978 o HOWTO design a GRAS application
979 More are due, of course. They will come latter. In the meanwhile, you can
980 check the examples which are still here.
982 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
984 SimGrid (3.1) stable; urgency=high
988 There was a stack corruption somewhere, visible only when optimizing
989 with these versions. [Vince]
992 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
993 module that enabled to play with Directed Acyclic Graphs. It is built
994 directly on top of SURF and provides an API rather close to the old
995 SG. Some old codes using SG are currently under rewrite to check that
996 all needful functions are provided. [Christophe Thiery]
999 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1000 extension of the classical CLM03 model that gracefully handles
1001 failures. This is now the default model for MSG and GRAS. It doesn't
1002 handle parallel tasks yet though. [AL]
1003 * Bug fix: Weights were not correctly set in the network part.
1004 WARNING: This may have resulted in incorrect results with simulations
1005 where there are more than one flow on a given link. [AL]
1008 * After a (long ?) discussion on simgrid-devel, we have decided that the
1009 convention we had on units was stupid. That is why it has been decided
1010 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1011 WARNING : This means that all previous platform files will not work as
1012 such with this version! A warning is issued to ask users to update
1014 A conversion script can be found in the contrib module of the CVS, under
1015 the name contrib/platform_generation/surfxml_update.pl [MQ]
1018 * Bug fix: Processes were started in reverse order, wrt deployment file.
1019 WARNING: if your code relies on this bug, please fix it. [AL]
1020 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1021 being executed on two different locations. [AL]
1022 * Bug fix: Failures are now better supported thanks to Derrick's tests
1023 (there was many failure situations I hadn't thought of and that weren't
1024 correctly handled). [AL]
1025 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1029 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1030 Exception occurring on server-side are propagated back to client (!).
1032 API CHANGE: the callback changed their prototype. Change:
1033 int my_handler(gras_socket_t expeditor, void *payload_data) {
1035 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1036 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1038 * New! function: gras_msg_handleall to deal with all messages arriving
1039 within a given period.
1040 * New! function: gras_socket_server_range to get a server socket in a
1041 range of port numbers (ease to avoid port number conflicts) [MQ]
1042 * New! gras processes display their backtrace when they get a SIGUSR1
1043 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1044 Sweet to debug RL processes [MQ]
1048 - Do not force experiment sizes to be expressed in kb, or it becomes
1049 impossible to measure the latency this way (needs one byte-long tests)
1050 WARNING: this changes the amok_bw_* function semantic. [MQ]
1051 - Implements the link saturation stuff. [MQ]
1052 * Peer management module:
1053 New! module factorizing code that we wrote over and over [MQ].
1056 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1057 - Test units are placed directly into the library code, they get extracted
1058 automatically and placed into the src/testall binary.
1059 - Convert most of the XBT tests to this system.
1060 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1061 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1062 live in the same C code. This required to change a little bit the API
1063 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1064 * New module: sparse graph structure with basic algorithms (this is work
1065 in progress and the API is not considered to be frozen yet). [AL]
1066 * Display more information on backtraces: source line & function names are
1067 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1068 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1069 * Reworked a little bit some #include statements to load only required
1070 headers. Some user code that relied on SimGrid to include stdlib or
1071 stdio may need to include it by themselves. [AL]
1072 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1074 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1075 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1076 but I guess that almost nobody used those parts. [MQ]
1078 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1080 SimGrid (3.0.1) stable; urgency=low
1083 * Unfortunately, I had missed 5 misnamed functions:
1084 xbt_fifo_item_t xbt_fifo_newitem(void);
1085 void xbt_fifo_freeitem(xbt_fifo_item_t);
1086 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1087 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1088 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1089 They're now deprecated. Please use their new versions:
1090 xbt_fifo_item_t xbt_fifo_new_item(void);
1091 void xbt_fifo_free_item(xbt_fifo_item_t);
1092 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1093 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1094 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1096 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1097 * Documentation: xbt_log module unmercifully reworked [MQ]
1098 * Bugfix: there was a problem with the ending of contexts with
1099 the pthread backend. It caused some weird deadlock or behavior
1100 depending on the pthread implementation. [AL]
1101 * Bugfix: get the exceptions raised in the simulator repport where
1102 and why they come from when they are not catched in time [AL, MQ]
1105 * Bugfix: Do repport the error when two non-connected hosts try to
1106 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1109 * Add additionnal checkings on communications. Assert that two
1110 communicating hosts are connected by a set of links... [AL]
1113 * Add additionnal checkings on channel values in communication [AL]
1114 * New: MSG_task_get_source to see on which host a task was generated [HC]
1115 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1116 the number of tasks waiting to be received on channel and sent
1118 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1119 waits for the first task coming from a given host.. [AL]
1121 GRAS new functionnalities: [MQ]
1122 * Enhance the parsing macro to allow the size of multidimentional objects
1123 to be given thru annotations.
1124 * New example (and documentation): Matrix Multiplication a la RPC
1125 (as when I was young!) and fix a bunch of bugs found on the way.
1127 GRAS performance improvements: [MQ]
1129 * Reduce the amount of cbps creation/destruction by making it static to
1130 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1133 * Change libdata to a set so that we can search for stuff by ID (and thus
1134 reduce the insane amount of dict lookups)
1137 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1138 RL communication on top of SG since it's so uneffective.
1139 It may also be used for inter-thread communication in RL, one day.
1140 * Use gras_datadesc_copy() to exchange messages on top of SG
1142 - improve message exchange performance on top of SG
1143 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1144 * Don't exchange on the network the size of the used part of buffer,
1145 instead, specify the possible buffer size to read().
1147 - reduces the amount of read/write calls (one pair per exchange)
1148 - reduces the amount of exchanged data (the size)
1149 - allows to retrieve all arrived data on receiver side, if we don't need
1150 it right now (subsequent read will peek the buffer)
1151 - allows the receiver to proceed with the begining of the stream before
1152 everything is arrived
1153 - make it possible to build an iov transport (using readv/writev)
1155 - take care of the data with non-stable storage (like stacked data),
1157 * If possible, TCP send uses vector I/O (when writev() is here)
1158 - Don't use it for receive since we send data sizes and data on the
1159 same stream, so we wouldn't be able to chain large amount of chunks
1160 before having to flush the stuff to read the size.
1161 * Rework the transport plugin mecanism to simplify it and reduce the
1162 amount of pointer dereferencement when searching for the right function
1165 * I guess that now, we do almost as few system calls as possible while
1166 doing as few data copy as possible.
1168 To improve it further, we could try to send all the sizes first and then
1169 all the data (to use iov on receiving size), but it's only a partial
1170 solution: when you have 2 dimensional data, the sizes of the second
1171 dimension is data of the first dimension, so you need 3 streams.
1173 I'm not sure the potential performance gains justify the coding burden.
1175 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1177 SimGrid (3.00) stable; urgency=high
1180 * New! Give the possibility to hijack the surf parser and thus bypass
1181 MSG_create_environment and MSG_launch_application. Have a look at
1182 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1184 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1186 SimGrid (2.96) unstable; urgency=low
1191 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1192 This deprecates the xbt_error_t mecanisms.
1193 It modifies (simplifies) all XBT and GRAS API.
1194 MSG API keeps unchanged (exceptions raised by XBT are catched from
1195 within MSG and masked with existing error handling facilities)
1198 * New! Add a FATPIPE model. [AL]
1199 * New! Add a parallel task model. [AL]
1200 * New! Add automatically a loopback interface (in the default
1201 network model) if none was precised.
1204 * Bugfix: MSG_process_resume now works with the current running process.
1206 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1207 * Modification of MSG_task_get_compute_duration. Once a task has been
1208 processed, the value returned by this function is now equal to 0. [AL]
1209 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1210 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1211 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1212 of a m_task. For now, MSG_task_cancel only works with computation
1214 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1215 of the processor (in Mflop/s) regardless of the current load on the
1217 * API Change: use proper naming convention for MSG_getClock and
1218 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1220 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1221 This function changes the priority of a computation task. This priority
1222 doesn't affect the transfer rate. A priority of 2 will make a task
1223 receive two times more cpu power than the other ones. This function
1224 has been added to suit the needs of Nguyen The Loc and hasn't been that
1225 much tested yet. So if it fails, please report it and send me your code.
1227 * API Change: removed all functions and types that were marked "deprecated"
1228 since many months. Renamed MSG_global_init_args to MSG_global_init.
1230 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1232 SimGrid (2.95) unstable; urgency=low
1235 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1236 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1237 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1240 * Bugfix: Allow absolute paths for platform description files [MQ]
1241 * Bugfix: do free the variables after use. Leads to drastic performance
1243 * Implement max_duration (ie, timeouts) on resources [AL]
1246 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1248 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1249 the needed support (timeouts) [AL]
1252 * Implement measurement sockets. You can now get the bandwidth between two
1253 hosts thanks to AMOK (see below). [MQ]
1254 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1255 dynar over the network (yeah) [MQ]
1256 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1257 * simple/static token-ring example. [Alexandre Colucci and MQ]
1258 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1259 in place before [MQ]
1262 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1263 type parsing. "short" and "long" modifiers were also ignored; other
1264 modifier (such as reference level) are still ignored. [MQ]
1265 * Embeed the buffer size within the buffer itself on SG. [MQ]
1266 That way, send() are atomic and cannot get intermixed anymore (at least
1267 the ones which are less than 100k; bigger messages still have the issue)
1268 * Array size pushed by the field, not by the field type (or each
1269 and every long int will push stuff to the cbps) [MQ]
1270 * use select() to sleep since it allows to portably sleep less than one
1273 GRAS (minor cleanups)
1274 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1277 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1278 * type category 'ignored' killed as it was never used and were difficult
1280 * whether a type can cycle or not is now a flag, leaving room for more
1281 flags (as "ignored", if we feel the need one day ;)
1282 * Rename raw sockets to measurement sockets since "raw" has another
1283 meaning in networking community.
1286 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1287 and offers features not belonging to GRAS but that most applications
1288 need. One day, it may be a set of plugins loadable at runtime.
1289 * New module: bandwidth
1290 bandwidth measurement between arbitrary nodes running this module. [MQ]
1292 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1294 SimGrid (2.94) unstable; urgency=low
1296 The first beta release of SimGrid 3 !
1300 * Update the main page and the FAQ. Adding references to gforge.
1303 * Add a gras_os_getpid function.
1306 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1307 * Extend the logs so that they also print PID, hostname, date, ... if
1309 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1310 and kill the old version which were in testsuite/
1311 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1315 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1316 for users. Instead, each programming environment (GRAS, MSG,...) use it
1317 when needed to provide such a feature to users.
1318 Don't shortcut the mecanism or you will also shortcut the virtualization
1319 you need on the simulator.
1323 * Cleanups in configury with regard to compile optimization/warning flags.
1324 Also add -fno-loop-optimize to any powerpc since it's the optimization
1325 killing gcc (< 3.4.0).
1326 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1327 needed by MSG examples complications
1328 * Borrow configury beautifications from PHP
1331 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1332 warning (thanks loris for stumbling into this one).
1333 * Bugfix: stop loading private headers (gras_config.h) from the public
1337 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1338 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1339 RL would imply a DNS resolver, which is *hard* to do in a portable way
1340 (and therefore delayed).
1341 * Implement a real timer mecanism and use it in timing macros. This allows
1342 to avoid rounding errors and get a 0.000005 sec precision in timing
1343 macros. While I was at it, various cleanups:
1344 - allow to declare more than one timed section per file (fix a stupid bug)
1345 - move some private declaration to the right place
1346 - merge conditional execution and timing macros into emulation module
1347 - document the module
1348 - make sure the module cleanups its mess on gras_exit
1349 * Documentation improvements:
1350 - (new) how to compile applications using GRAS
1351 - (new) emulation support (timing macros)
1353 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1355 SimGrid (2.93) unstable; urgency=low
1357 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1360 - Use Paje properly where used. Still to be sanitized properly.
1361 - Portability fix: Add an implementation of the contexts using pthread
1365 - Add xbt_procname(): returns the name of the current process.
1366 Use it to show the current process's name in all logging.
1368 - fix detection of older flex version and the reaction, since we do
1369 depend on modern ones (we use lex_destroy)
1370 - Better separation of SG and RL in the libs: remove all simulation code
1371 from libgras. As a result, this lib is now only 200k when stripped.
1372 Some of the xbt modules may also be duplicated (two sets and such) and
1373 should be cleaned/killed before SG3.
1374 - Insist on using xlC on AIX because of weird problems involving gcc there.
1375 - Cleanup the make remote stuff. This is now done by scripts
1376 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1377 mainly for our private use, but we're working on changing them to user
1380 - Bugfix: flush the socket on close only if there is some *output*.
1381 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1383 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1386 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1388 SimGrid (2.92) unstable; urgency=low
1390 Alpha 3 on the path to SimGrid 3
1394 - New! First try of benchmarking macros.
1395 - New! First try so that gras_stub_generator generate deployment and
1396 remote compilation helpers.
1398 - Bugfix: Initialization fix in msg_test.
1402 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1404 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1406 - API Change: gras_os_sleep to take the amount of seconds as a double.
1407 Accepting an int was error prone since it was the only location where
1408 seconds were coded as such. It leaded to damn rounding errors.
1409 - Bugfix: Hard to belive that timers ever worked before this.
1411 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1413 SimGrid (2.91) unstable; urgency=low
1415 Alpha 2 on the path to SimGrid 3
1419 - Bug fix in the lmm_solver.
1421 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1422 through the function MSG_paje_output.
1423 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1424 - It is possible to bound the rate of a communication in MSG with
1425 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1426 to write it in the changelog).
1427 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1432 - Various cleanups to the autotools stuff
1433 - Begin to move Gras examples to examples/gras/
1434 - Let make distcheck work again (yeah!)
1436 - documentation overhauled using doxygen.
1437 gtk-doc-tools is dead in SimGrid now.
1438 - Automatically extract all existing logging categories, and add the list
1439 to the documentation (long standing one, to say the less)
1441 - Cleanup the known architecture table. Reorder the entries to group what
1442 should be, and use a more consistent naming scheme.
1443 (some of the test dataset are still to be regenerated)
1444 - New! Allow library to register globals on each process just as userdata
1446 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1447 lookup time (for now).
1448 Use it in msg and trp.
1449 This cleans a lot the internals and helps enforcing privacy of the
1450 headers between the gras components.
1451 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1452 - Bugfix: gras_os_time was delirious in RL.
1453 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1455 - Reenable GRAS now that it works.
1457 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1459 SimGrid (2.90) unstable; urgency=low
1461 Alpha 1 on the path to SimGrid 3
1463 * It is a long time since the last release of SimGrid. I'm sorry about
1464 that but as I had told you, I was rewriting a lot of things. I apologize
1465 to those who had been reporting bugs to me and that I had not answered.
1466 If your bug is still in the new version, please tell me. Here is a
1467 summary of the main changes.
1469 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1470 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1471 portability, speed, and a lot more but you'll figure it out later.
1472 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1473 I wanted to release the new MSG as soon as possible and I have
1474 broken GRAS, which is the reason why, for now, only MSG is fully
1475 functional. A laconic description of these projects is available
1476 in the documentation.
1478 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1479 kernel optimized for our needs (called SURF but only the developers
1480 should use it). Hence, MSG is now roughly 30 times faster and I think
1481 that by rewriting a little bit MSG, I could event speed it up a little
1482 bit more. Beside the gain in speed, it is also much easier to encode a
1483 new platform model with SURF than it was with SG. More to come...
1485 * REVOLUTION 3: I have tried to change a little as possible the API of
1486 MSG but a few things really had to disappear. The main differences
1487 with the previous version are :
1488 1) no more m_links_t and the corresponding functions. Platforms are
1489 directly read from a XML description and cannot be hard-coded
1490 anymore. The same format is used for application deployment
1491 description. The new format is described in the documentation.
1492 Have a look in tools/platform_generation. There is a tiny script
1493 that converts from the old platform format to the new one. Concerning
1494 the application deployment format, parsing the old one is tricky.
1495 I think most of you should however be able to convert your files. If
1496 it is really an issue, I can write a C code that does the conversion.
1498 2) the toolbox tbx does not exist anymore. We now have a library
1499 with much more data-structures but without the hash-tables (we have
1500 dictionaries that are much faster).
1502 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1504 *****************************************************************************
1505 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1506 * don't feel like converting the oldies. (Mt) *
1507 *****************************************************************************
1510 Version 2.90: "the long awaited one"
1511 - Finished rewriting and debugging MSG. Rewrote the documentation.
1512 - disable GRAS for now since it needs to be ported to the newest SG
1515 - Finish the port to windows (using mingw32 for cross-compile)
1518 - Main loop and datastructures of SURF. A cpu resource object is
1519 functional. Surf can thus be used to create cpu's with variable
1520 performance on which you can execute some actions.
1522 2004-11-15 Martin Quinson
1523 - Port to ARM. Simply added the alignment and size descriptions. Should
1524 work, but the ARM machines are so slow that I didn't had the opportunity
1525 to 'make check' over there yet.
1527 2004-11-15 Arnaud Legrand
1528 - Trace manager now written. It uses a heap structure and is therefore
1529 expected to be efficient. It may however be speeded up (particularly
1530 when many events occur at the same date) by using red and black
1531 trees. One day maybe...
1532 - Max-min linear system solver written. It uses a sparse matrix
1533 structure taking advantage of its expected use. Most operations are
1534 O(1) and free/calloc are called as few as possible. The computation of
1535 the minimum could however be improved by using a red and black tree
1538 2004-11-03 Arnaud Legrand
1539 - Rename every gras_* function that was in xbt/ to its xbt_
1541 - Add a heap and a doubly-linked list to xbt
1542 - Added a dichotomy to the dictionaries. make check works as well before
1543 so I assume that the patch is correct. I do not know however if things
1544 run effectively faster than before now. :)
1546 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1547 SimGrid, and the version number is bumped to 2.x
1549 2004-10-29 Martin Quinson
1550 - Introduction of the remote errors.
1551 They are the result of a RMI/RPC on the remote machine.
1552 ErrCodes being scalar values, you can't get the host on which those
1553 errors did happen. Extending the error mechanism as in Gnome is possible.
1554 No idea yet whether it is a good idea.
1556 2004-10-28 Martin Quinson
1557 - Interface revolution: the Starred Structure Eradication.
1558 I used to do typedef struct {} toto_t; and then handle *toto_t.
1559 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1560 - ???_t is a valid type (builded with typedef)
1561 - s_toto_t is a structure (access to fields with .)
1562 - s_toto is a structure needing 'struct' keyword to be used
1563 - e_toto_t is an enum
1564 - toto_t is an 'object' (struct*)
1566 typedef struct s_toto {} s_toto_t, *toto_t;
1567 typedef enum {} e_toto_t;
1568 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1569 s_toto_t) is private.
1571 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1572 so that it can set the variable to NULL. It was so for dicts and sets,
1573 it changed for dynars.
1575 - Fix a bunch of memleaks in dict_remove
1576 - Fix a bug in sg/server_socket opening: it failed all the time.
1578 2004-10-07 Martin Quinson
1579 - Speed up dynar lookup operation a bit.
1581 gras_dynar_get is dead.
1583 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1584 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1585 (which gives you the address of the stored data).
1587 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1588 copy of the data using an affectation to do the job and not a memcpy.
1590 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1591 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1593 It does not really speedup the dynar test because they are
1594 setting elements all the time (and look them seldom). But the dict does
1595 far more lookup than setting.
1597 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1599 2004-10-05 Martin Quinson
1600 - Allow to (en/dis)able the cycle detection at run time.
1602 Whether we should check for cycle or not is now a property of each
1603 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1604 datadesc_cycle_unset allow to remove this property when previously set.
1606 Note that the cycle detection is off by default since it impacts the
1607 performance. Watch the data you feed GRAS with ;)
1609 This property is hereditary. Any element embedded in a structure having it
1610 set have it set for the time of this data exchange.
1612 You should set it both on sender and receiver side. If you don't set it on
1613 sender side, it will enter an endless loop. If you forget on receiver
1614 side, the cycles won't be recreated after communication.
1616 - Header reorganization.
1617 Kill gras_private.h, each submodule must load the headers it needs.
1619 2004-10-04 Martin Quinson
1620 - Interface revolution: do not try to survive to malloc failure.
1622 Now, gras_malloc and friends call gras_abort() on failure.
1623 As a conclusion, malloc_error is not a valid error anymore, and all
1624 functions for which it was the only gras_error_t return value are
1625 changed. They now return void, or there result directly.
1626 This simplify the API a lot.
1628 2004-09-29 Martin Quinson
1629 - Re-enable raw sockets.
1630 Created by gras_socket_{client,server}_ext;
1631 Used with gras_raw_{send,recv}
1634 It should allow to kill the last bits of gras first version soon.
1636 This is not completely satisfactory yet (duplicate code with
1637 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1640 - Simplify transport plugin (internal) interface by not passing any
1641 argument to _server and _client, but embedding them in the socket
1644 2004-09-28 Martin Quinson
1645 - Finish the port to AIX.
1646 autoconf was my problem (segfault within the malloc replacement
1647 function. No idea why)
1649 2004-09-16 Martin Quinson
1650 - Fix some size_t madness on 64bit architectures.
1652 2004-09-08 Martin Quinson
1653 - Reduce the number of system headers loaded, overload some more system
1654 calls (such as malloc to cast the result of the system one, and work
1656 - Fix and reintroduce the config support
1658 2004-09-07 Martin Quinson
1659 - Source code reorganization to allow Arnaud to surf all over there.
1660 - Allow to document the logging categories.
1661 - Remove all uppercase from logging categories and useless cleanup in names.
1663 2004-08-18 Martin Quinson
1664 Version 0.6.2 (protocol not changed; API changed)
1665 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1666 gras_error_t), and NULL when not found. Functions expecting a msgtype
1667 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1668 hopefully usefull message.
1669 - Portability to prehistoric sparcs again
1671 2004-08-17 Martin Quinson
1672 Version 0.6.1 (protocol not changed; ABI not changed)
1673 - prealloc some buffers to speed things up
1675 2004-08-11 Martin Quinson
1676 Version 0.6 (protocol not changed; ABI expended)
1677 - The parsing macro can deal with the references, provided that you add
1678 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1680 2004-08-09 Martin Quinson
1681 Version 0.5 (protocol not changed; ABI changed)
1682 - Allow to off turn the cycle detection code in data exchange at
1683 compilation time. It should be at run time, but I'm short of time (and
1684 the config stuff is still broken). That way, we keep dict out of the
1685 critical path, which is good because the performance is poor:
1686 - search not dichotomial yet
1687 - dynar give no way to access their content and memcpy everytime
1688 - In composed data description (struct, ref and so on), stop foolness of
1689 keeping the subtype's ID, but store the type itself. This keeps sets out
1690 of the critical path, which is good since they rely on dynar and
1691 dictionnaries. The only loose of that is that we cannot detect the
1692 redeclaration of a structure/union with another content (but I'm not sure
1693 the code detected well this error before anyway). We still can detect
1694 the redefinition discrepancy for the other types.
1695 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1696 it breaks the code because of type-punning used all over the place).
1697 This breaks on all non-gcc architectures (for now).
1699 All those changes (plus the buffer of last time) allow me to gain 2 order
1700 of magnitude on cruel tests consisting of 800000 array of integers on two
1701 level of a hierarchical structure (200 secondes -> 4 secondes)
1704 - the selector of reference must now return the type it points to, not
1705 the ID of this type.
1707 2004-08-06 Martin Quinson
1708 Version 0.4 (protocol changed; ABI not changed)
1709 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1710 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1711 - Implement a buffer transport, and use it by default (it relies on tcp in
1712 real life and on sg in simulation).
1713 That's a bit hackish since I had a new field to the structure to store
1714 its data without interfering with the subtype ones. Inheritance
1715 is tricky in C. And that's a kind of reverse inheritance with one class
1716 derivating two classes. Or maybe a game with java interfaces. Anyway,
1717 that's damn hard in C (at least).
1718 Moreover, I got tired while trying to ensure plugin separation and
1719 genericity in SG mode. MSG wants me to do weird things, so let's go for
1720 cruel hacks (temporarily of course ;).
1721 See comment in transport_private.h:71
1722 - do not include all the _interface headers in private but in the files
1723 which really need them (to cut the compilation time when they are
1726 2004-07-26 Martin Quinson
1727 Version 0.3 (protocol not changed; ABI changed)
1728 - Major overhault of the datadesc interface to simplify it:
1729 - shorted the function names:
1730 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1731 - add a trivial way to push/pop integers into the cbps without malloc.
1732 This allows to make really generic sub_type description, which simply
1733 pop their size of the stack.
1734 - add a function gras_datadesc_ref_pop_arr() which does what users want
1735 most of the time: Declare a dynamic array (which pops its size of the
1736 stack) and declare a reference to it. Poor name, but anyway.
1737 - kill the post-send callback, add a post-receive one
1739 2004-07-23 Martin Quinson
1740 Version 0.2 (protocol changed; ABI changed)
1741 - add some testing for cpbs in the test cases, and fix some more bugs.
1742 This invalidate again the little64 data file, since I cannot regenerate
1744 - remove an awfull optimization in the logging stuff, allowing me to:
1745 - understand it again
1746 - learn gcc how to check that the argument match the provided format
1747 - fix all errors revealed by gcc after that
1748 - internal keys of dict are not \0 terminated. Deal with it properly in
1749 loggings instead of segfaulting when the user want to see the logs :-/
1751 2004-07-22 Martin Quinson
1752 - Fix some stupid bug preventing cbps (callback postit) from working
1754 2004-07-21 Martin Quinson
1755 - Some documentation cleanups
1756 - remove the useless last argument of msgtype_declare
1757 - rename the Virtu functions to fit into the 'os' namespace
1758 - move headers src/include -> src/include/gras/ and stop fooling with
1759 gras -> . symbolic link
1760 - make distcheck is now successful
1762 2004-07-19 Martin Quinson
1764 - Build shared library also
1765 - Install html doc to the right location
1766 - stop removing maintainer files in make clean
1767 - build tests only on make check
1769 2004-07-13 Martin Quinson
1771 - No major issue in previous version => change versionning schema
1772 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1773 corresponding dataset.
1775 2004-07-11 Martin Quinson
1777 - Get it working with any kind of structure (we can compute the padding
1778 bytes remotely for all the architectures I have access to)
1779 - Implement the structure parsing macro (still not quite robust/complete)
1780 - Improvement to the remote testing toysuite
1782 2004-07-10 Martin Quinson
1783 [autoconf mechanism]
1784 - get ride of a bunch of deprecated macros
1785 - actually run the test for two-compliment, not only compile it :-/
1786 - test whether the structures get packed (and bail out if yes. Damn.
1787 Alignment is a serious matter)
1788 - test whether the structures get compacted (but respecting the alignment
1789 constraints of each types)
1790 - test whether the array fields of structures can straddle alignment boundaries
1792 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1793 (took me 2 hours to find that bug, looking at the wrong place)
1794 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1795 before sending/receiving and are used to compute the offsets of fields
1796 - Given that padding size depend even on compiler options, keep track of
1797 alignment and aligned_size only for the current architecture. Not a big
1798 deal since we send structure fields one after the other (seems
1800 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1801 program, let it work on linux/gcc/little32. portability todo.
1803 2004-07-08 Martin Quinson
1804 - import and improve remote compilation support from FAST
1805 - make sure make check works on half a dozen of machines out there
1807 2004-07-07 Martin Quinson
1808 Let's say it's version 0.0.3 ;)
1809 - Implement conversions (yuhu!)
1810 - Let it work on solaris (beside conversion, of course)
1811 - Stupid me, using rand() to generate the conversion datatests in not wise.
1813 2004-07-06 Martin Quinson
1814 - Let make dist work, since I'm gonna need it to compile on remote hosts
1815 - Let Tests/datadesc_usage write the architecture on which the file was
1816 generated as first byte.
1817 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1818 architecture descriptions.
1819 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1820 on those architectures.
1821 - Optimization: send/recv array of scalar in one shoot
1823 2004-07-05 Martin Quinson
1824 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1826 - Plug a whole bunch of memleaks
1827 - each process now have to call gras_{init,exit}. One day, their log
1828 settings will be separated
1829 - Continue the code factorisation between SG, RL and common in Transport.
1831 2004-07-04 Martin Quinson
1833 - Redistribution between SG and RL.
1834 We wanna have to accept in SG, so move accepted related parts of RL in
1835 the common part. (more precisely, the dynar of all known sockets is no
1836 more a static in transport.c, but part of the process_data)
1838 [gras_stub_generator]
1839 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1841 2004-07-03 Martin Quinson
1842 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1843 them all up in one shot)
1845 - Fix the ugly case of reference to dynamic array.
1846 - New (semi-public) function gras_datadesc_size to allow the messaging
1847 layer to malloc the needed space for the buffer.
1849 - gras_socket_close now expect the socket to close (and not its address to
1850 put NULL in it after it). This is because the socket passed to handlers
1851 is one of their argument (=> not writable).
1853 - propagate the interface cleanup from last week in datadesc, ie remove a
1854 superfluous level of indirection. User pass adress of variable
1855 containing data (both when sending and receiving), and not of a variable
1856 being a pointer to the data. Let's say that I like it better ;)
1857 The price for that is constructs like "int msg=*(int*)payload" in
1858 handlers, but it's a fine price, IMHO.
1860 - Let it work in RL (yuhu)
1862 2004-06-21 Martin Quinson
1864 - porting SG plugin and SG select to new standards (works almost).
1865 - plug memleaks and fix bugs around.
1868 - cleanup the prototype of data recv and force users to specify when they
1869 want to handle references to objects. Test case working even for cycles.
1870 - plug memleaks. Valgrind is perfectly ok with this.
1872 2004-06-12 Martin Quinson
1874 - cleanup the separation between plugin and main code in plugin creation
1876 2004-06-11 Martin Quinson
1878 - Reput hook for raw sockets, needed for BW experiments
1879 - kill a few lines of dead code
1880 [Data description] Interface cleanup
1881 - gras_datadesc_by_name returns the searched type or NULL.
1882 That way, no variable is needed to use a type desc once, which makes
1884 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
1885 two parameters were strings (field name, type name), leading to
1887 [Dicos] Interface cleanup
1888 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
1889 This is consistant with the dynar API.
1891 2004-04-21 Martin Quinson
1893 - Porting to new standards.
1895 - interface cleanup.
1896 There is no bag anymore, no need to take extra provision to mask the
1897 pointers behind "ID".
1898 Better splitup of functions between files create/exchange/convert.
1899 This is still a bit artificial since convert and receive are so
1900 interleaved, but anyway.
1902 - add a queued message list to procdata (the ones not matching criteria
1904 - factorize some more code between SG and RL wrt procdata
1906 - use gras_exit in example to track memleaks
1907 - get rid of gs_example now that GS is properly integrated into gras
1908 - update run_test to integrate the lastest tests (datadesc)
1910 - rename WARNINGn macros to WARNn since it prooved error-prone
1912 2004-04-19 Martin Quinson
1914 - register init/exit functions within gras module mechanism
1915 - send/receive function.
1916 Convertion is not implemented, but short-cutted if not needed.
1917 struct/array elements are sent one by one (instead of block-wise), but
1918 nobody really cares (yet). Get a prototype before optimizing.
1919 - tests (using a file socket) for DD send/receive on:
1920 - base types: int, float
1921 - array: fixed size, string (ie ref to dynamic string)
1922 - structure: homogeneous, heterogeneous
1923 - chained list, graph with cycle
1924 Believe it or not, valgrind is not too unhappy with the results. The
1925 cycle happily segfaults, but the others are ok. And I'm sick of pointers
1929 - Bugfix when using a filename explicitely (instead of '-')
1931 2004-04-09 Martin Quinson
1933 - factorize more code between RL and SG in socket creation
1934 - Complete the implementation and tests of:
1936 o file (only in RL, and mainly for debugging)
1938 I lost 3 days to design a portable address resolver, and then decided
1939 that the prototype mainly have to run on my box.
1940 Addressing portability too early may be like optimizing too early :-/
1942 - use gras_init in the Tests instead of the crappy parse_log_opt
1943 (the latter function is removed)
1944 [Conditional execution]
1945 - New functions: gras_if_RL/gras_if_SG (basic support for this)
1946 [Code reorganisation]
1947 - Get rid of libgrasutils.a since it makes more trouble than it solves.
1948 Build examples against the RL library, since there is no way to disable
1949 its creation for now.
1951 For information, the beginning of coding on GRAS was back in june
1952 2003. I guess that every line has been rewritten at least twice since