1 SimGrid (3.6) unstable; urgency=low
4 * Change the prototypes of action replay. Sorry for inconvenience,
5 but this is really more efficient this way (and to adapt your code,
6 you just have to fix the initialization, that shouldn't be too long)
7 * Kill the braindead MSG_task_refcount_dec() function. I guess nobody
8 ever managed to do anything useful with it.
9 * New function MSG_comm_testany(). Similarly to MSG_comm_waitany(), it
10 takes a dynar of communications. It returns immediately and gives the
11 index of a finished communication (if any).
12 * New function MSG_comm_get_status(). MSG_comm_test() and MSG_comm_testany()
13 only say if a communication is finished, no matter whether it succeeded or
14 failed. You can call MSG_comm_get_status() to know the status of a finished
16 * New function MSG_task_dsend() to send a task and detach it. When a
17 communication is detached, you are never notified of its success or failure
18 and the memory is released automatically once it is finished. This function
19 is useful when you don't care about the end nor the success of a
20 communication you send.
23 * New model for multi-core CPUs. You can now use the core attribute to
24 precise the number of cores of a host. This is a basic model. Every
25 process running on the host receive at most the power provided in
26 the DTD (throughput<=power). Total throughput of process cannot exceed
28 * New peer tag. This peer tag creates a tiny AS comprising a host and a
29 router linked by an up-link and a down-link (possibly asymmetrical).
30 This kind of pattern allows to easily build last-mile model style platforms.
31 Aggregating such patterns in a rule-based AS is thus the technique of
32 choice for modeling large peer-to-peer/volunteer computing/cloud platforms.
33 * New model for Vivaldi routing. We transformed the Vivaldi network model
34 into a Vivaldi routing model (based on the rule-based model). This allows to
35 combine Vivaldi based latencies with last-mile platforms.
38 * Added a check for NaN of IEEE754 infinite in the double entries of
40 * Introduce a new context factory "raw", highly inspirated from the
41 ucontext factory, but using manually crafted functions in assembly to
42 do the work in an efficient manner.
43 * Allow to change the used context factory at run time, not only at
44 compilation time. Use --cfg=contexts/factory:raw for maximal speed.
45 * Add an option --cfg=contexts/stacksize:N to set the stack size of the user
46 contextes at runtime (only with raw contexts or ucontexts).
47 * Completely rewrote this module to allow parallel execution of user
48 processes. Use --cfg=contexts/nthreads:N to execute user processes
49 with N parallel threads (the default is 1, meaning no parallelism).
50 * Allow to decide dynamically between sequential and parallel modes.
51 When nthreads > 1, you can use --cfg=contexts/threshold:P to run the user
52 processes in parallel only when their number is greater than or equal to P.
55 * New command line option: if you pass --cfg=verbose-exit:0, SimGrid
56 won't output the state of processes when interrupted with Ctrl-C
57 * Add a new function xbt_dynar_to_array that transforms a dynar into a
58 NULL-terminated array. This may solve backward compatibility issues
59 due to the change to return type of SD_Simulate. See also:
60 http://lists.gforge.inria.fr/pipermail/simgrid-user/2010-December/002206.html
61 * Add new macros with variable number of arguments.
62 - in xbt/log.h: XBT_DEBUG, XBT_VERB, XBT_INFO, etc.
63 - in xbt/asserts.h: xbt_assert
64 - in xbt/cunit.h: xbt_test_{add,fail,assert,log}
65 - in xbt/ex.h: THROWF and RETHROWF.
66 Define XBT_USE_DEPRECATED if you want to use the old numbered macros like
68 * Change xbt_die() to accept a format string with arguments, just like printf.
69 * New data structure: xbt_lib_t, like a dict but more general and with better memory handling.
72 * New configuration options
73 Options triva/categorized and triva/uncategorized can be used to generate
74 graph configuration files for Triva visualization tool.
75 * Configuration option tracing/platform is renamed to tracing/categorized
76 * XBT logging makes tracing error checks easier, new root log hierarchy: instr
77 * New TRACE_user_link_variable interface:
78 User provides the name of the link and the tracing variable to attach to it
79 * the declaration of tracing categories must be done after the environment creation
80 * simpler tracing interface, just one way to declare categories
81 TRACE_category or TRACE_category_with_color, it is up to you
82 * links in the trace file are again identified by their names
83 * trace contains the full platform hierarchy exactly as declared using the ASes
84 * Options tracing/msg/[task|process]:1 groups the process by hosts
85 for both cases, tasks and processes must have names that are unique during the simulation
86 these options generate traces that are suited to gantt-charts, such as the space-time view of Paje
87 * The experimental option tracing/msg/volume is deprecated
88 its functionality may be reincorporated if needed
91 * Define a SIMGRID_VERSION macro in simgrid_config.h.
92 - We are trying hard to keep the API stable, but it may happen that
93 some things change (we're a research project after all, not a
94 nuclear plant operating system). If such things should happen, you
95 could rely on that macro to adapt.
96 - current value: 30600 for 3.06.00, aka 3.6
97 * Define macro MAKE_SIMGRID_VERSION(major, minor, patch) to help building
98 a number that can be compared with SIMGRID_VERSION.
99 * Add a build option -Denable_debug (set to ON by default): when set to OFF,
100 assertions and verbose/debug logging events are disabled at compile time.
102 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
104 SimGrid (3.5) stable; urgency=medium
107 * New feature to any SimGrid-based simulator: Model-Checking
108 Check SIN#1 for more details.
111 * New Model SMPI (three-interval linear regression for correction factors)
112 See RR-7426, available at http://hal.inria.fr/inria-00527150
113 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
114 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
115 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
116 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
117 * See also src/smpi/README
122 - Tracing API changes: TRACE_start and TRACE_end should not be called
123 by user-code. They are automatically called by simulators created
124 with SimDAG, MSG and SMPI if the toolkit is compiled with
125 tracing_enabled=ON. Categories declaration and utilization remain the
126 same for MSG and SimDag.
127 - A function was added to the tracing API to declare categories with
129 - TRACE_category_with_color (char *category, char *color)
130 where color must be in the following format
131 "%f %f %f", red, green, blue
132 and red, green, blue are float values in the interval [0, 1]
133 - User can specify NULL as color parameter, or continue calling
135 On that case, the tracing system will define random colors
136 - The following command-line options are supported:
137 --cfg=tracing/filename:msg.trace
138 --cfg=tracing:1 (activate tracing, needed to use others)
139 --cfg=tracing/platform:1 (categorized resource use)
140 --cfg=tracing/uncategorized:1 (uncategorized resource use)
141 --cfg=tracing/msg/task:1 (task creation)
142 --cfg=tracing/msg/process:1 (process creation, migration)
143 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
144 --cfg=tracing/smpi:1 (SMPI interface tracing)
145 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
146 - examples of examples/msg/tracing updated
148 - DAXLoader and DOTLoader functions can generate tasks with categories
149 - A new function to attribute a category to SD tasks:
150 TRACE_sd_set_task_category (SD_task_t task, char *category)
151 * Tracing the MPI interface implemented by SMPI
152 - Collective operations are traced with states
153 - Point-to-Point operations are traced with states/links
154 - Tracing activated by a parameter "-trace filename" passed
155 to smpirun during execution (considering that simgrid
156 is compiled with tracing enabled)
157 - To run the simulation with gdb, the simulator
158 accepts --cfg=tracing/smpi:1 to trace SMPI
159 - tesh files to check if smpi tracing is ok
160 - See examples/smpi/NAS/DT-trace
161 * GTNetS tracing re-worked
162 - adaptation to the tracing system of GTNets to cope
163 with modifications regarding the fullduplex mode
164 - new tesh files to check if gtnets tracing is ok
167 * Asynchronous communications through the functions:
168 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
169 * New function: MSG_load_platform_script()
170 to make possible using a lua script instead of XML files to set up platforms
171 * New function: MSG_set_function
172 to associate functions to processes, used when bypassing the parser
173 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
175 Platforms: Add some more examples in examples/platforms
176 * Grid'5000: see www.grid5000.fr
177 * *_30000_hosts.xml: various huge files [mainly scalability testing]
180 * Change the XML format. This is a very important modification. SimGrid
181 3.5 introduces a new hierarchical format based on the notion of
182 Autonomous Systems. Compatibility with old format is ensured
183 through the perl script provided in the install bin directory
184 bin/simgrid_update_xml.
185 It is now possible to build platforms with specific routing
186 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
187 connect several platforms together. We will try to provide soon
188 set of realistic platforms exploiting these properties (have a
189 look at examples/platforms/ for the moment).
190 * Take the opportunity of the XML format change to be a good XML citizen:
191 rename link:ctn to link_ctn and similar changes (also dealed with by
193 * Add a new routing scheme (rule-based) using regular expressions. It
194 enables to have an extremely low memory footprint when the
195 underlying routing is simple and can be compactly described. You
196 need to have libpcre4-dev (perl regular expressions) installed if
197 you want to use this routing scheme.
198 * Revive the cluster TAG and allow to easily and efficiently (both in
199 term of memory and speed) connect clusters together. Have a look
200 at teshsuite/simdag/platforms/ to see how this can be done. With
201 this tag, you can create clusters with thousands of tasks at no
202 cost (have a look at examples/platforms/).
203 Note: clusters are implemented as ASes, so there is no need for an
204 enclosing AS tag if you have only one cluster in your platform.
205 * Add new generic functions in the public interface that allows the user
206 to call SURF 'create_resource' methods from your code (same
207 functionality as the XML bypass mechanism but with a much lighter
209 * Add a new model (enabled through command line --cfg=network/model:SMPI)
210 that uses a piecewise linear approximation to produce better
211 results when exchanging small messages.
212 * Add a new parameter to handle correctly full duplex link and account
213 for interferences between uplink and downlink communications
214 (activate with --cfg=fullduplex:1).
217 * Rename the SD_READY (all dependencies are satisfied and task is
218 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
219 dependencies are satisfied) state.
220 This prevents a confusion between the notion of "ready to schedule"
221 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
222 simulated" (SD_RUNNABLE) used by the simulation kernel.
223 * Change the way a task is considered as ready. Instead of removing
224 dependencies when a task is done, a counter is decreased. This way,
225 it is always possible to reach ancestors thanks to the
226 SD_taks_get_parents function (even after the end of the simulation.)
227 * Change the return type of SD_Simulate from (SD_task_t*) into
228 xbt_dynar_t. This function was in handling a dynar internally and
229 converted it into a NULL terminated array for historical reasons.
230 * New function SD_dotload(char*) to load a DAG described in dot
231 format. This loader and the corresponding examples require the
232 installation of the graphviz library.
233 * Fix a bug in the management of tasks of size 0 in the surf network
234 models. This problem was only visible with SIMDAG and you should
235 thus disregard results produced with earlier versions if you
236 relied on this feature (some tasks were blocked because of this).
237 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
238 even though your applications were doing only point-to-point
239 communications and sequential computations. Now you can really use any
240 model you want (of course, if you create real parallel tasks, which are
241 not implemented in most models beside ptaskL07, this will abort).
242 * Add an example that schedules a DAX on an heterogeneous platform
243 using a Min-Min strategy.
244 * New function SD_workstation_get_current_task() that returns the kind
245 of task currently running on a workstation in the sequential access
247 * Raise some warnings when unexecuted tasks remains at the end of the
248 simulation. This is usually caused by cycles in the DAG.
251 * New function: SIMIX_process_set_function() called by MSG_set_function
252 * Change the underlying waiting queue in semaphores so that a process
253 can wait on several of them simultaneously (as in waitany).
254 * Fix the way to handle tokens in semaphores so that all access patterns
255 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
256 * kill the dirty pimple SIMIX_message_sizes_output()
257 Please use (proper) visualization instead
260 * New data container: setset (set of sets of elements)
261 * New module: mmalloc (mapped malloc, allowing to have several
262 independent segments of malloc)
263 * New function: xbt_dict_cursor_set_data()
264 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
265 * New function: xbt_dynar_is_empty()
266 * New function: xbt_fifo_get_last_item()
267 * Fix xbt_dynar_shrink(): use the right element size.
268 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
269 uninitialized areas during expand.
270 * Fix semaphores: previous implementation was severely broken.
271 * Use library init/fini functions for our initialization.
272 - you can use logs and other feature as soon as you want in your
273 code (even before the xbt_init / MSG_init)
274 - xbt_exit is now a no-op and produce a warning when used.
277 * Port GRAS to new SIMIX mechanisms. This allows gras users to
278 benefit from the latest improvement to the simulation kernel.
279 * Kill measurement sockets for now. If you rely on them, sorry. This
280 release is not for you. This feature will be reintroduced in the
281 future, but we cannot delay the release any further.
282 * New function: gras_msgtype_get_name().
283 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
284 * Fix (at last) the pmm example: it should not randomly fail anymore.
286 Build chain: bug fixes and overall polishing
287 * Cmake is now stable enough. Hence, we killed the autotools.
288 * Port to windows ( TM :)
289 * Fix the 'make install' target.
290 No need to use 'make install-simgrid' anymore
291 * Introduce a 'make dist' target compiling a *source* archive
292 'make package' compiles a binary archive
293 * Compile java files only on need
294 * Add --cd and --setenv command line options to tesh
295 * Out of source builds are not fully supported yet, but we are close
296 * Enable supernovae and optimization flags by default for our users
299 * Add layer to set up environment directly from lua, without XML.
300 * The effect of gras_stub_generator can be achieved through
301 lua too (check examples/gras/console/ping_generator.lua)
303 -- Wed, 01 Dec 2010 22:09:23 +0100 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
305 SimGrid (3.4.1) stable; urgency=low
307 The "Polishing easter eggs is probably a good idea" release.
308 This is a bug fixes release only.
312 * Fix a bug preventing the tasks from begin garbage collected.
315 * Fix a bug occuring when a host involved in a communication fails.
316 This was not detected properly by the other peer involved in the
317 communication. Now, it's reported as a network error.
320 * Warn the user about loop dependencies in data flow of DAX files
321 * Obey the control-flow dependencies of DAX files
324 * Add option "enable_smpi" allowing to not compile SMPI.
325 Probably useful for the (Mac) users experiencing a build error here
326 * Improve the detection of lua5.1 and ruby1.8
328 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
330 SimGrid (3.4) stable; urgency=low
332 The "Easter in Cargese" release. Also known as (major changes):
334 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
335 ~> bindings were greatly improved
336 ~> new tracing infrastructure for better visualization introduced
338 * the "Welcome to configury modernity" release.
339 ~> we switched from autotools to cmake, and improved our cdash
342 A more detailled list of changes follow (full detail in svn log).
344 Java Bindings: Various Cleanups
345 * (install java-gcj-compat-dev on debian-like to use them)
346 * Remove put/get: no need to export deprecated interface in Java
347 Use send/receive instead.
348 * Cleanup the examples and add a README per directory
349 * Remove example autoDestination (that's the only way to go now)
350 * Remove example explicitDestination (was a plain copy of basic)
351 * Make JniException a runtime exception, so that there is no need to
352 declare the fact that you may encounter such a beast. I guess that
353 nobody will ever want to survive such error.
354 * Create specific errors for each MSG case of failure:
355 host failure, transfer failure, timeout, task cancelled
356 * Cleanup the exceptions that may get thrown by each function
357 * Other internal cleanups in Java bindings. Performance still bad :/
358 Ruby and Lua Bindings: create them
359 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
360 * That's new and great, you should try them out.
361 Same functionalities than Java bindings, only even less polished
363 * Kill the useless "rate" argument of SD_task_get_execution_time()
364 Everyone used to provide -1 as a value, it was not used, and the
365 semantic of a possible use wasn't even clear.
366 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
367 either as comm costs or compute costs to mean that there is no
368 such thing for that specific task.
369 * Add a SD_task_set_name() function
370 * Fix SD_task_unschedule() on typed tasks
371 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
372 * In DAX loader, accept useless 'level' attributes to <job> since
373 LIGO DAGs have them (seem to be to ease graphical representation).
375 * Add an example masterslave_mailbox.c using send/receive and not
376 the deprecated put/get interface.
377 * Kill the MSG_paje_output() function. It's a noop since 2 years.
378 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
380 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
381 (declare MSG_USE_DEPRECATED to still have the old name)
382 * Add a MSG_task_set_data() function
383 * About trace replay (see examples/msg/actions):
385 - Allow to work with splitted trace files for each process
386 Give the specific trace file as argument of each process,
387 and call MSG_action_trace_run(NULL)
388 You can still have one merged file for all processes.
389 - Fix implementation of collective operations
390 * Allow task_execute() on 0-sized tasks (closes #10063)
392 * This is the first release of SimGrid where SMPI is not considered
393 beta anymore (even if some corners should still be improved)
394 * Port over the new SIMIX_network submodule (internal refactoring)
395 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
396 * Implement more missing elements of the standard:
398 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
400 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
401 MPI_Scan() MPI_Get_processor_name()
402 - Added implementation of missing case for Alltoall (warning: it's
403 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
404 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
405 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
406 SMPI_MPI_Allgatherv()
408 - MPI_Waitsome() was broken
409 - Allow relative includes in smpicc
410 - Command line cfg argument 'reference_speed' was ignored...
411 - Some functions did not properly lead to auto-benching of user code
412 - smpicc passes -O2 by default (just like openmpi one)
414 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
415 * Bug fixes about timeouts during communications
416 * add SIMIX_message_sizes_output() as a pimple to write to file the
417 amount of messages per size. Use gnuplot to get histogram.
418 Pimple because that's the only user-visible function of simix,
419 defined directly in xbt.h (irk, sorry)
421 - Add a SIMIX_sem_get_capacity() function
422 - Fix interactions with processe resume/suspende
423 - release_forever() was stupidly broken
424 - Fix SIMIX_display_process_status() for processes in a semaphore
425 - Make SIMIX_sem_block_onto() user-visible
426 * Refactoring context stuff:
427 - Use pseudo-OOP for better modularity
428 - reimplement SIMIX_process_kill() without process_schedule() so
429 that the latter can take as invariant that it is called from
431 - Merge context_start into context_new for sake of simplicity
433 * Add a Vivaldi network model, coded live during SUD'10 ;)
434 * Rename configuration variables to start a hierarchy:
435 o cpu_model -> cpu/model
436 o network_model -> network/model
437 o workstation_model -> workstation/model
438 * New configuration variables:
439 o network/bandwidth_factor: correction to bandwith
440 o network/latency_factor: correction to latency
441 o netwotk/weight_S: correction to the weight of competing streams
442 * Add a long description to the models, that users can see with such
443 argument on the command line: --cfg=cpu/model:help
444 * --help-models display the long description of all known models
446 * config: add the ability to set a default value after registration
447 Does not override any previously set value (e.g. from cmd line)
448 * dict: allow to have integer key and data.
449 When so, you need to use the following functions
450 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
451 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
452 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
453 In contrary to regular dicts, the key is not malloced before copy.
454 Mixing scalar and regular elements in the same dict is not tested
456 * Allow to use xbt_dynar_shrink() to expend the dynar instead
457 Tracing for Visualization:
458 * SimGrid is now instrumented in order to generate a trace file for
459 visualization analysis: to use it, need to compile SimGrid with the
460 "tracing" option enabled, and instrument the program using SimGrid with
461 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
462 (among other functions).
463 * The instrumentation only traces the platform utilization for now
464 * Documentation to use the tracing functions and how to analyze the
465 traces with the Triva tool is written.
466 * More information about: SimGrid FAQ (in the section Tracing Simulations
469 * We moved to cmake as default build system. Autotools support will
470 be dropped soon. Check the FAQ for more info about how to use it.
471 * Greatly improved our cdash/ctest interactions
472 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
473 * Added memory checking tests with valgrind; lot of memleak fixing.
474 This may be the first release of simgrid with so few memory issues
475 * Added code coverage tests.
476 Our coverage is still improvable, but at least we see it on cdash.
478 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
480 SimGrid (3.3.4) stable; urgency=low
482 The "Desktop Grid needs love too" release (also called Xmas release).
485 * Major speedup in the maxmin system solving by using lazy evaluation
486 Instead of solving completely the maxmin system at each iteration,
487 only invalidate (and recompute) the modified parts.
488 This new feature is enabled in default models but you can try to
489 turn it on with "--cfg:maxmin-selective-update=1" for other models.
490 * Cas01 IMproved as default CPU model
491 This CPU model is the same Cas01 model, but it uses the
492 maxmin-selective-update flag and a heap structure to manage
493 actions on SURF kernel.
494 It reduces the complexity to find the next action to finish and,
495 consequently, it's faster than the old Cas01.
496 This is the new default CPU model (Cas01).
497 * Rename the old Cas01 model to Cas01_fullupdate
498 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
499 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
500 * CpuTI (CPU Trace Integration)
501 A new CPU model whose objective is simulate faster when using
502 availability trace files.
503 Instead of using a full featured, over engineered maxmin system for
504 CPU modeling, this model does the pre-integration of traces files
505 to calculate the amount of CPU power available, and so, executes
506 faster than the old CPU models.
507 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
508 * Use LV08 as default network model since it gives better accuracy
509 for small messages and shouldn't change things for big ones.
510 Use --cfg=network_model:CM02 to get the previous behavior.
513 ******************************************
514 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
515 ******************************************
516 * The new CPU model may changes simulations!
517 The point is that events occurring at the exact same timestamp
518 are not scheduled in the same order with the old and new
519 version. This may be enough to completely change the execution
520 of simulations in some cases.
521 * The new network model will change simulations!
522 This new model is more realistic than the previous one, so you
523 should consider redoing your old experiments with this model.
524 Sorry for the inconvenience.
527 * Introduce the supernovae compilation mode
528 When compiled that way, the whole SimGrid (or almost) is put in a
529 single compilation unit and compiled in one shoot.
530 This is to help gcc which has difficulties to inline stuff from one
532 The speedup seem to be above 15%, althrough more tests are needed on
533 amd64 to confirm that gain.
536 * Port of MSG's mailbox on top of SIMIX network
537 The put/get mechanism was greatly simplified on the way.
540 * New SIMIX network module. Provides:
541 - Mailbox: rendez-vous mecanism to find with who you want to speak
542 - Synchronous send/recv: easier and hopefully faster since the
543 logic is handled in the maestro process directly now
544 - Asynchronous send/recv: you dreamt of it? It's here now
545 Too bad that nobody cared enough to propagate the change to MSG.
546 * Add semaphores as SIMIX synchronization mechanism.
549 * new function SD_daxload(char*) to load a DAX file
550 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
551 * Introduce typed tasks. Specify its kind and cost at creation.
552 At scheduling, just give where it should be placed, and the cost
553 for each involved resource is automatically computed.
554 Existing constructors so far (more to come of course):
555 - SD_task_create_comm_e2e() for end-to-end communication
556 - SD_task_create_comp_seq() for sequential computation
557 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
558 * new function SD_task_dump() for debuging display
559 * new function SD_task_dotty(task,FILE*) writing to file the info
560 about the task in dotty format
561 * SD_task_dependency_exists() can now cope with having one of its
562 arguments NULL. If so, it tests whether the other argument has any
564 * Add getters on list of preceding/following tasks:
565 SD_task_get_parents(task) and SD_task_get_children(task)
566 * Add getters on amount of workstations and list:
567 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
568 * Add getter on task kind: SD_task_get_kind(task)
569 * Update the start_time and finish_time of tasks on completion/failure
570 * Bugfix: Remove task from state swags when destroyed
573 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
576 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
577 dynar with one separate thread per value of the dynar.
578 * Change the prototype of xbt_thread_create(), sorry.
579 Added a boolean parameter indicating whether we want to join this
580 thread (used in SG only for now)
581 * Implement xbt_thread_join and xbt_thread_yield in SG also.
584 * GTNetS wrappers should now be usable again (and betterly tested too)
585 * Fix a major regression from 3.2 where the timeout provided to
586 MSG_task_put_with_timeout() was used as absolute time before which
587 the comm should be done.
588 * Start to fix the <cluster> tag.
589 - Internal links should be good now (beside of the loopback, which
590 use the private link instead)
591 - paths to the external world is still rather broken
592 - the <route:multi> tag is just broken. Actually that's brain-dead.
593 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
594 to make it less stupid
595 ** Check your platform with teshsuite/simdag/platforms/flatifier **
596 * Fix a source-level compatibility glitch from 3.2: after defining
597 MSG_USE_DEPRECATED, you can use the old name
598 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
599 * Allow to compile from the SVN with automake 1.11
600 * Fix some problems when using the "start_time" tag in deployment XMLs.
601 * Fix #8569: XBT/synchro.h has redundant declarations
602 * Fix #8563: MSG return values and exceptions
603 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
604 * Integrate patch #8636: Obey DESTDIR when installing documentation.
605 Thanks to Robson Peixoto.
606 * Fix a vicious bug in dictionaries inducing that some elements were
607 not freed on xbt_dict_free()
609 Portability report of this version:
610 * Main portability targets:
611 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
613 - mac leopard on i386
614 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
615 but nothing critical.
616 * Other platforms: windows, AIX and others were not tested for this release
618 Timing report of this version:
619 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
620 scenario parameters). From 8h to a few seconds in desktop grid settings.
621 * Supernovae brings about 25% speedup on i386.
623 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
625 SimGrid (3.3.3) stable; urgency=low
627 The "Need for Speed" release.
629 The timings done to validate the 3.3.2 were faulty.
630 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
632 The problem was a conversion from a manually handled vector to
633 xbt_dynar_t on the critical path.
634 xbt_dynar_foreach calls functions, inducing stack management crap.
636 We inlined these functions and xbt_dynar_foreach is now breath taking.
637 We also inlined xbt_swag_belong on the way.
639 Here are some approximate speedup measurements (on master/slaves
640 simulations lasting between 10s and 20s each):
641 3.3.1 -> 3.3.2: about same performance
642 3.3.2 -> 3.3.3: 40% speedup
643 3.3.1 -> 3.3.3: 40% speedup
644 3.3.1 with inline patch -> 3.3.3: 30% speedup
646 Our reading is that the refactoring which occurred in 3.3.2 made us
647 suffer much more from the xbt_dynar_foreach low performance, but
648 once we solved this, this refactoring proved to be very performance
649 effective. From the 40% speedup, somehow, 10% are due to the
650 inlining and 30% to the refactoring.
652 That's a pitty that gcc cannot inline functions placed in other files
653 alone. We have to choose between:
654 - break the encapsulation (by putting private data structures and
655 accessors in headers files to help gcc)
656 - live with low performance
657 - switch to a decent compiler such as icc (not quite possible).
659 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
661 SimGrid (3.3.2) stable; urgency=low
663 The "Simplicity does not preceed complexity, but follows it" release.
665 The main contributors of this release were (lexical order):
666 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
669 * Extract the routing logic into its own object.
670 (was dupplicated in network.c and workstation_LV07.c;
671 Allows to implement other ways of storing that info)
672 => kill now useless network_card concept
673 - Use dynar to represent routes (instead of void** + int*)
674 - kill link_set (use surf_network_model->resource_set instead)
675 - Add a command-line option to choose the routing schema to use
676 - Add three new models:
677 * Floyd (shortest path computed at initialization)
678 * Dijikstra (shortest path recomputed all the time)
679 * Cached Dijikstra (shortest path computed on need)
680 All these models where contributed by Silas De Munck, and are
681 described in his ICCS09 paper.
683 * Simplify model declaration
684 (less redirections, less function to write when defining a model)
685 - Factorize stuff between models:
688 surf_model_resource_set(model)
689 surf_model_resource_by_name(model, name)
690 - Unify the types of models in s_surf_model_t (using an union)
691 - Embeed fields of common_public directly into s_surf_model_t
692 - Rename model methods:
693 action_free ~> action_unref
694 action_change_state ~> action_state_set
695 action_get_state ~> action_state_get
696 - Change model methods into functions :
697 (model)->common_public->action_use ~> surf_action_ref
699 * Implement a generic resource; use it as ancestor to specific ones
700 (allows to kill duplicated code in models)
701 Drawback: timer command don't need no name nor properties;
702 workstation_CLM03 don't need no properties
703 (but I guess we can live with those few bytes wasted)
705 * Improve the action object model
706 - implement a constructor avoiding dupplicated code about field
707 initialization in generic_action part.
709 * Kill the SDP model: it has an external dependency, is deprecated
710 in flavor of modern lmm models, and didn't compile since a while
713 * Relocation of the context module from XBT to SIMIX.
714 (the context were decoupled from the simix processes, duplicating a lot of code)
715 => a lot of code was factorized
716 - less overhead is introduced during scheduling
717 - simpler API for the context factory
718 - the logic for process creation,destruction and manipulation was simplified
719 * Simplification of the s_smx_process_t data structure.
720 => accesing the simix level data associated to a process is faster now,
721 and the code is a lot more readable.
724 * Implement some more MPI primitives:
725 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
726 -implementation: Bcast: flat or 2-ary tree (default),
729 Allreduce: Reduce then Bcast
730 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
731 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
732 Alltoallv: flat tree, like ompi
734 * Add support for optimized collectives (Bcast is now binomial by default)
735 * Port smpirun and smpicc to OS X
738 * Kill SD_link_get_properties: hard to maintain and makes very little sense
739 Shout out if you used it.
742 * Display the list of still queued messages in SG mode when existing
746 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
747 * Add xbt_graph_node_get_outedges() [Silas De Munck]
748 * Add xbt_str_from_file(FILE*)
749 * Add xbt_dict_get_key achieving a linear reverse search
750 * Remove the context module
752 Portability report of this version:
753 * Main portability targets:
754 - Linux(debian)/x86/context
755 - Linux(debian)/x86/pthreads
756 - Linux(debian)/amd64/context
757 - Linux(debian)/amd64/pthreads
758 On these, we still have the eratic breakages of gras/pmm and
759 amok/saturate_sg reported in previous version. We still think
760 that the tests are the cause of the fault, not the tested code.
762 - Mac OSX Leopard/x86/context
763 Still false negative in tesh autotesting.
764 Smpi still fails, but this time because readlink does not accept -f
765 Everything seems to work properly beside of that.
768 - AIX version 5.3 (only tested contexts this time)
769 Smpi still fails there because mktemp is not installed.
770 Everything seems to work properly beside of that.
772 I managed to compile it for the first time, but several breakages.
773 Won't delay the release for this exotic platform.
775 * Windows: it's still lagging behind. If you want to help, please
778 Timing report of this version:
779 This version seem to be more than 5% faster than 3.3.1 (on linux
780 64bits with contextes). The gain is less than expected, we are
781 investigating this for next release.
783 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
785 SimGrid (3.3.1) stable; urgency=low
788 * Implement a --cfg-help to show existing configuration variables
789 * Build chain do not require doxygen in maintainer mode
792 * fix a bug on struct sizeof computation, which prevented the
793 exchange of arrays of structs in some conditions
794 - added a regression test about this in datadesc_usage
795 * Allow the exchange of 0-long dynamic vectors.
796 - for that, use -1 as indicator of dynamic size instead of 0
797 - This implied to change any size from unsigned long to long,
798 reducing a bit communication abilities, but I guess that with
799 64bits being quite common, this is more than enough.
800 - This also induce a protocol change, thus bumping network protocol
801 version from 0 to 1 (if we have external users, we have to get
802 clean on that point too ;)
803 - added two regression tests about this in datadesc_usage
804 * Be more verbose when propagating local exceptions
805 This helps debugging.
806 * Display the status of simulated processes when receiving SIGINT in
810 * Allow to control the simulation from a trace file.
811 New functions MSG_action_register() and MSG_action_trace_run()
812 The first one allows to associate a function execution to each
813 kind of action while the second one parses a trace file and
814 triggers the corresponding actions within the system.
815 For now, only a toy example is provided in examples/msg/actions
816 * Add an exemple of process migration in examples/msg/migration
817 * Fix a bug in task exchange which broke MSG_task_get_sender()
818 Add a teshsuite regression test for that.
819 [Bug: if MSG_task_get_sender() is called after sender exit,
821 * Fix a bug which prevented suspend/resume to work properly
822 * Display the status of simulated processes when receiving SIGINT
823 This fixes a regression of v3.3. due to the introduction of SIMIX
824 * Bug fixing in failure management:
825 - trace could not start by a failure at time 0
826 - failure during communications were not working
829 * Add SIMIX_process_set_name() to change the name of the current
830 process in the log messages.
831 * Store smx_hosts in a dict since we only retrieve them by name
832 * Move the configuration infrastructure to surf
835 * Move the configuration infrastructure to surf
838 * Massive internal cleanups:
839 - Store internal structures on processes instead of hosts (allows
840 to have more than one process per host, in addition of being more
842 - Cleanup the initialization/finalization process
843 - Kill a whole bunch of unneeded synchronization:
844 processes run in exclusive manner within the simulator
845 - Move queues from global tables to process data fields
847 - now accept -platform and -hostfile arguments
848 - Pass the right rank value to processes according to the hostfile
849 * Compile the examples by default, and use them as regression tests
850 * Implement MPI_Wtime()
851 * Change the reference speed to a command line option
854 * TCP_gamma can now be specified as command line option using
855 --cfg=TCP_gamma:10000000.0
856 * Change the --surf-path cmd line option into --cfg=path:
859 * Also include strbuff from xbt.h public header
860 * xbt_ex_display(): do not free the exception after displaying
861 This allows to do more with the given exception afterward.
862 Users should call xbt_ex_free() themselves.
866 Portability report of this version:
867 * Main portability targets:
868 - Linux(debian)/x86/context
869 - Linux(debian)/x86/pthreads
870 - Linux(debian)/amd64/context
871 - Linux(debian)/amd64/pthreads
872 These targets fail about 1/10 of times on gras/pmm, but we believe
873 that this is because of the test, not because of simgrid.
874 amok/saturate_sg fails even more rarely, and the test may not be
877 - Mac OSX Leopard/x86/context
878 The test suite still spits tons of errors because some obscure
879 force prevents us from removing the temporary directories
880 arguing that they still contain some metadata I've never heard of.
881 Smpi fails because seq is not installed.
882 Everything seems to work properly beside of that.
885 - AIX version 5.3 (both contexts and pthread)
886 Smpi still fails there because mktemp is not installed.
887 XML inclusions seems rosty on AIX.
889 * Windows: it's still lagging behind. If you want to help, please
892 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
894 SimGrid (3.3) stable; urgency=high
898 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
901 * Introduce the SIMIX module: factorize code between MSG and GRAS.
904 Until now, GRAS were using MSG as an interface to SURF. It was
905 quite difficult because both interface have several differences
906 (MSG channels vs GRAS sockets were the most notable point).
908 This also opens the gate to SMPI (which should occur soon) and speed
909 up simulations by to 40% (even if it were not the main goal).
911 **************************************
912 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
913 **************************************
914 The point is that events occuring at the exact same timestamp are
915 not scheduled in the same order with the old and new version. This
916 may be enough to completely change the execution of simulations in
917 some cases. Sorry for the inconvenience.
919 * Cleanup and upgrade the XML format to push further scalability
920 issues (check http://hal.inria.fr/inria-00256883/ for more info)
922 * Improve the testing infrastructure with tesh. Now a very large part of
923 the code is tested not only by being run but also by checking that the
924 output match an expected output [Mt].
926 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
927 is really less memory-demanding, which should allow you to use
928 larger files in SimGrid [AL].
930 * Inform valgrind about our contextes, so that it becomes usable
931 with the default (and more effecient) version of SimGrid
932 [contributed by Sékou Diakite, many thanks]
935 * Introduce a listener thread in charge of receiving incomming
936 messages from the network. It allows to overlap communication and
937 computation but most notably, it removes some stupid deadlocks due
938 to the fact that so far, a process could not send and receive at
939 the same time. This made most non trivial communication schema
941 * Convert the PIDs from long int to int to match the MSG ones (and
943 * New function: gras_agent_spawn() to launch a new process on
944 current host. Only working in simulation for now. [Mt]
945 * New function: gras_os_hostport() returning a constant form (ie,
946 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
949 * Make the backtrace of exceptions more human readable [Mt]
950 * New module: xbt/str [Mt]
951 a ton of string utility functions (split, join, printf to a newly
952 allocated buffer, trim, etc)
953 * New module: xbt/hash [Mt]
954 SHA1 hashing algorithm (more to come if needed)
955 * New module: xbt/synchro [Mt]
956 synchronization tools (mutex and conditions) working the same way
957 in simulation and in real life (mainly useful for GRAS, but not
959 * New module: xbt/queue [Mt]
960 classical producer/consumer synchronization scheme
961 * xbt_dynar_new_sync() creates a synchronized dynar. All access
962 (using the classical functions will get serialized) [Mt]
963 * Make dictionary internal table dynamic. No need to specify its size
964 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
966 * Make sure the log channels are organized as a tree under windows
967 (because of ANSI C compatibility issue, any channel were child of
971 * Cleaned many thing in surf and fixed a few bugs [AL].
972 * Add a nice command line configuration mechanism to compose models [AL].
973 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
974 the previous one (KCCFLN05). It relies on something that looks like
975 a max-min sharing mechanism but cannot be written as such. A new solver
976 was thus designed [AL].
977 * Add a new solver to lmm. Based on Lagrange optimization and
978 gradient-based descent, it enables to efficiently maximise systems s.a
980 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
982 This solver enables to propose two new network models for TCP Reno and
983 TCP Vegas based on Low's work. These models still need to be fully
984 tested though [Pedro Velho].
987 * Bug fix in SD_simulate. Now the time bound given as argument is
989 * Use the new parallel task model (ptask_L07) as default.
990 * Use the SURF command line configuration mechanism.
991 * 0-size tasks (for synchronization) should now work.
993 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
995 SimGrid (3.2) stable; urgency=high
999 We still experience issues on this platform, but we believe that at
1000 least MSG is usable.
1002 GRAS API BREAKAGE (for simplification purpose, sorry):
1003 * the gras_msgtype_by_name is not used anymore. Instead of
1004 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
1005 you can write (and must)
1006 gras_msg_send(toserver, "request", &request);
1007 - If you still want to pass a gras_msgtype_t to the function (to cache
1008 the type and avoid the lookup time), use the gras_msg_send_() variant.
1009 - Impacted functions:
1010 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
1011 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
1012 * The callbacks are now expected to return 0 when everything went well
1013 (just like the main() function)
1015 GRAS new features and improvements:
1016 * New module mecanism where user code can use per process globals [Mt]
1017 This is similar to gras_userdata_*() functions, but for libraries. It
1018 factorize some code developped over and over in the examples and AMOK.
1019 It has still to be documented and used (only amok/peermanagement is
1021 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
1022 loss when they were small enough to fit into the buffer and sent quickly
1023 enough so that they can all get received in one shoot.
1024 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
1025 if not found. Use the *_or_null() variant for the old semantic.
1026 * In gras_msg_handle, do not discard messages without callback.
1027 They are probably messages to be explicitly awaited later (ie, proofs of
1028 mis-synchronization in userland since they are sent before being awaited)
1029 No big deal usually.
1030 * gras_socket_meas_send/recv: semantic changed!
1031 The numerical arguments used to be (1) the total amount of data to send
1032 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
1033 messages. This was need for the fool willing to send more than MAXINT
1034 bytes on quite fat pipes.
1037 * Do really rename the hostmanagement module to peermanagement. [Mt]
1038 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
1039 but this is rather new and this was documented in the module
1040 documentation (poor excuses, I admit)
1041 * Bandwidth measurement semantic changed! This follows the changes to
1042 gras_socket_meas_send/recv explained above.
1045 * A sequential mode has been added to the workstations. When a workstation
1046 is in sequential mode, it can execute only one task, and the other tasks
1047 are waiting in a FIFO. [Christophe Thiery]
1050 * The KCCFLN05 workstation model now handles parallel tasks. It is the
1051 model for SIMDAG. [Christophe Thiery]
1052 * Bug fix in the maxmin solver: Some values were close to 0 instead of
1053 equal to 0, which caused some bad behaviors in
1054 saturated_constraint_set_update. I now use a threshold mechanism like in
1058 * When running manually src/testall, you select specific units [Mt]
1059 testall is the result of our cunit mecanism, and should replace all
1060 the scripty thingy around since bash don't run easily on billware.
1062 * A mallocator system has been added. [Christophe Thiery]
1063 Mallocators allow you to recycle your unused objects instead of freeing them
1064 and allocating new ones.
1066 Documentation update:
1067 * FAQ reworking + New FAQs:
1068 - "Valgrind spits tons of errors!" [Mt]
1069 - "How to repport bugs" [Mt]
1070 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
1071 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
1072 - Communication time measurement within MSG [AL]
1073 - I experience weird communication times when I change the latency [AL]
1074 * GRAS tutorial [Mt]
1076 - an introduction to the framework and to the used communication model
1077 - an initiatic tour introducing the most proheminent features:
1079 . Lesson 0: Installing GRAS
1080 . Lesson 1: Setting up your own project
1081 o Part 2: Message passing
1082 . Lesson 2: Exchanging simple messages
1083 . Lesson 3: Passing arguments to the processes (in SG)
1084 . Lesson 4: Attaching callbacks to messages
1085 . Lesson 5: Using globals in processes
1086 . Lesson 6: Logging informations properly
1087 . Lesson 7: Using internal timers
1088 . Lesson 8: Handling errors through exceptions
1089 . Lesson 9: Exchanging simple data
1090 . Lesson 10: Remote Procedure Calling (RPC)
1091 . Lesson 11: Explicitely waiting for messages
1092 . Recapping of message passing features in GRAS
1093 - A HOWTO section containing:
1094 o HOWTO design a GRAS application
1095 More are due, of course. They will come latter. In the meanwhile, you can
1096 check the examples which are still here.
1098 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
1100 SimGrid (3.1) stable; urgency=high
1104 There was a stack corruption somewhere, visible only when optimizing
1105 with these versions. [Vince]
1108 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1109 module that enabled to play with Directed Acyclic Graphs. It is built
1110 directly on top of SURF and provides an API rather close to the old
1111 SG. Some old codes using SG are currently under rewrite to check that
1112 all needful functions are provided. [Christophe Thiery]
1115 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1116 extension of the classical CLM03 model that gracefully handles
1117 failures. This is now the default model for MSG and GRAS. It doesn't
1118 handle parallel tasks yet though. [AL]
1119 * Bug fix: Weights were not correctly set in the network part.
1120 WARNING: This may have resulted in incorrect results with simulations
1121 where there are more than one flow on a given link. [AL]
1124 * After a (long ?) discussion on simgrid-devel, we have decided that the
1125 convention we had on units was stupid. That is why it has been decided
1126 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1127 WARNING : This means that all previous platform files will not work as
1128 such with this version! A warning is issued to ask users to update
1130 A conversion script can be found in the contrib module of the CVS, under
1131 the name contrib/platform_generation/surfxml_update.pl [MQ]
1134 * Bug fix: Processes were started in reverse order, wrt deployment file.
1135 WARNING: if your code relies on this bug, please fix it. [AL]
1136 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1137 being executed on two different locations. [AL]
1138 * Bug fix: Failures are now better supported thanks to Derrick's tests
1139 (there was many failure situations I hadn't thought of and that weren't
1140 correctly handled). [AL]
1141 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1145 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1146 Exception occurring on server-side are propagated back to client (!).
1148 API CHANGE: the callback changed their prototype. Change:
1149 int my_handler(gras_socket_t expeditor, void *payload_data) {
1151 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1152 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1154 * New! function: gras_msg_handleall to deal with all messages arriving
1155 within a given period.
1156 * New! function: gras_socket_server_range to get a server socket in a
1157 range of port numbers (ease to avoid port number conflicts) [MQ]
1158 * New! gras processes display their backtrace when they get a SIGUSR1
1159 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1160 Sweet to debug RL processes [MQ]
1164 - Do not force experiment sizes to be expressed in kb, or it becomes
1165 impossible to measure the latency this way (needs one byte-long tests)
1166 WARNING: this changes the amok_bw_* function semantic. [MQ]
1167 - Implements the link saturation stuff. [MQ]
1168 * Peer management module:
1169 New! module factorizing code that we wrote over and over [MQ].
1172 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1173 - Test units are placed directly into the library code, they get extracted
1174 automatically and placed into the src/testall binary.
1175 - Convert most of the XBT tests to this system.
1176 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1177 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1178 live in the same C code. This required to change a little bit the API
1179 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1180 * New module: sparse graph structure with basic algorithms (this is work
1181 in progress and the API is not considered to be frozen yet). [AL]
1182 * Display more information on backtraces: source line & function names are
1183 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1184 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1185 * Reworked a little bit some #include statements to load only required
1186 headers. Some user code that relied on SimGrid to include stdlib or
1187 stdio may need to include it by themselves. [AL]
1188 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1190 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1191 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1192 but I guess that almost nobody used those parts. [MQ]
1194 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1196 SimGrid (3.0.1) stable; urgency=low
1199 * Unfortunately, I had missed 5 misnamed functions:
1200 xbt_fifo_item_t xbt_fifo_newitem(void);
1201 void xbt_fifo_freeitem(xbt_fifo_item_t);
1202 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1203 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1204 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1205 They're now deprecated. Please use their new versions:
1206 xbt_fifo_item_t xbt_fifo_new_item(void);
1207 void xbt_fifo_free_item(xbt_fifo_item_t);
1208 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1209 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1210 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1212 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1213 * Documentation: xbt_log module unmercifully reworked [MQ]
1214 * Bugfix: there was a problem with the ending of contexts with
1215 the pthread backend. It caused some weird deadlock or behavior
1216 depending on the pthread implementation. [AL]
1217 * Bugfix: get the exceptions raised in the simulator repport where
1218 and why they come from when they are not catched in time [AL, MQ]
1221 * Bugfix: Do repport the error when two non-connected hosts try to
1222 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1225 * Add additionnal checkings on communications. Assert that two
1226 communicating hosts are connected by a set of links... [AL]
1229 * Add additionnal checkings on channel values in communication [AL]
1230 * New: MSG_task_get_source to see on which host a task was generated [HC]
1231 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1232 the number of tasks waiting to be received on channel and sent
1234 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1235 waits for the first task coming from a given host.. [AL]
1237 GRAS new functionnalities: [MQ]
1238 * Enhance the parsing macro to allow the size of multidimentional objects
1239 to be given thru annotations.
1240 * New example (and documentation): Matrix Multiplication a la RPC
1241 (as when I was young!) and fix a bunch of bugs found on the way.
1243 GRAS performance improvements: [MQ]
1245 * Reduce the amount of cbps creation/destruction by making it static to
1246 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1249 * Change libdata to a set so that we can search for stuff by ID (and thus
1250 reduce the insane amount of dict lookups)
1253 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1254 RL communication on top of SG since it's so uneffective.
1255 It may also be used for inter-thread communication in RL, one day.
1256 * Use gras_datadesc_copy() to exchange messages on top of SG
1258 - improve message exchange performance on top of SG
1259 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1260 * Don't exchange on the network the size of the used part of buffer,
1261 instead, specify the possible buffer size to read().
1263 - reduces the amount of read/write calls (one pair per exchange)
1264 - reduces the amount of exchanged data (the size)
1265 - allows to retrieve all arrived data on receiver side, if we don't need
1266 it right now (subsequent read will peek the buffer)
1267 - allows the receiver to proceed with the begining of the stream before
1268 everything is arrived
1269 - make it possible to build an iov transport (using readv/writev)
1271 - take care of the data with non-stable storage (like stacked data),
1273 * If possible, TCP send uses vector I/O (when writev() is here)
1274 - Don't use it for receive since we send data sizes and data on the
1275 same stream, so we wouldn't be able to chain large amount of chunks
1276 before having to flush the stuff to read the size.
1277 * Rework the transport plugin mecanism to simplify it and reduce the
1278 amount of pointer dereferencement when searching for the right function
1281 * I guess that now, we do almost as few system calls as possible while
1282 doing as few data copy as possible.
1284 To improve it further, we could try to send all the sizes first and then
1285 all the data (to use iov on receiving size), but it's only a partial
1286 solution: when you have 2 dimensional data, the sizes of the second
1287 dimension is data of the first dimension, so you need 3 streams.
1289 I'm not sure the potential performance gains justify the coding burden.
1291 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1293 SimGrid (3.00) stable; urgency=high
1296 * New! Give the possibility to hijack the surf parser and thus bypass
1297 MSG_create_environment and MSG_launch_application. Have a look at
1298 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1300 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1302 SimGrid (2.96) unstable; urgency=low
1307 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1308 This deprecates the xbt_error_t mecanisms.
1309 It modifies (simplifies) all XBT and GRAS API.
1310 MSG API keeps unchanged (exceptions raised by XBT are catched from
1311 within MSG and masked with existing error handling facilities)
1314 * New! Add a FATPIPE model. [AL]
1315 * New! Add a parallel task model. [AL]
1316 * New! Add automatically a loopback interface (in the default
1317 network model) if none was precised.
1320 * Bugfix: MSG_process_resume now works with the current running process.
1322 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1323 * Modification of MSG_task_get_compute_duration. Once a task has been
1324 processed, the value returned by this function is now equal to 0. [AL]
1325 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1326 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1327 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1328 of a m_task. For now, MSG_task_cancel only works with computation
1330 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1331 of the processor (in Mflop/s) regardless of the current load on the
1333 * API Change: use proper naming convention for MSG_getClock and
1334 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1336 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1337 This function changes the priority of a computation task. This priority
1338 doesn't affect the transfer rate. A priority of 2 will make a task
1339 receive two times more cpu power than the other ones. This function
1340 has been added to suit the needs of Nguyen The Loc and hasn't been that
1341 much tested yet. So if it fails, please report it and send me your code.
1343 * API Change: removed all functions and types that were marked "deprecated"
1344 since many months. Renamed MSG_global_init_args to MSG_global_init.
1346 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1348 SimGrid (2.95) unstable; urgency=low
1351 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1352 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1353 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1356 * Bugfix: Allow absolute paths for platform description files [MQ]
1357 * Bugfix: do free the variables after use. Leads to drastic performance
1359 * Implement max_duration (ie, timeouts) on resources [AL]
1362 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1364 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1365 the needed support (timeouts) [AL]
1368 * Implement measurement sockets. You can now get the bandwidth between two
1369 hosts thanks to AMOK (see below). [MQ]
1370 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1371 dynar over the network (yeah) [MQ]
1372 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1373 * simple/static token-ring example. [Alexandre Colucci and MQ]
1374 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1375 in place before [MQ]
1378 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1379 type parsing. "short" and "long" modifiers were also ignored; other
1380 modifier (such as reference level) are still ignored. [MQ]
1381 * Embeed the buffer size within the buffer itself on SG. [MQ]
1382 That way, send() are atomic and cannot get intermixed anymore (at least
1383 the ones which are less than 100k; bigger messages still have the issue)
1384 * Array size pushed by the field, not by the field type (or each
1385 and every long int will push stuff to the cbps) [MQ]
1386 * use select() to sleep since it allows to portably sleep less than one
1389 GRAS (minor cleanups)
1390 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1393 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1394 * type category 'ignored' killed as it was never used and were difficult
1396 * whether a type can cycle or not is now a flag, leaving room for more
1397 flags (as "ignored", if we feel the need one day ;)
1398 * Rename raw sockets to measurement sockets since "raw" has another
1399 meaning in networking community.
1402 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1403 and offers features not belonging to GRAS but that most applications
1404 need. One day, it may be a set of plugins loadable at runtime.
1405 * New module: bandwidth
1406 bandwidth measurement between arbitrary nodes running this module. [MQ]
1408 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1410 SimGrid (2.94) unstable; urgency=low
1412 The first beta release of SimGrid 3 !
1416 * Update the main page and the FAQ. Adding references to gforge.
1419 * Add a gras_os_getpid function.
1422 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1423 * Extend the logs so that they also print PID, hostname, date, ... if
1425 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1426 and kill the old version which were in testsuite/
1427 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1431 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1432 for users. Instead, each programming environment (GRAS, MSG,...) use it
1433 when needed to provide such a feature to users.
1434 Don't shortcut the mecanism or you will also shortcut the virtualization
1435 you need on the simulator.
1439 * Cleanups in configury with regard to compile optimization/warning flags.
1440 Also add -fno-loop-optimize to any powerpc since it's the optimization
1441 killing gcc (< 3.4.0).
1442 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1443 needed by MSG examples complications
1444 * Borrow configury beautifications from PHP
1447 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1448 warning (thanks loris for stumbling into this one).
1449 * Bugfix: stop loading private headers (gras_config.h) from the public
1453 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1454 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1455 RL would imply a DNS resolver, which is *hard* to do in a portable way
1456 (and therefore delayed).
1457 * Implement a real timer mecanism and use it in timing macros. This allows
1458 to avoid rounding errors and get a 0.000005 sec precision in timing
1459 macros. While I was at it, various cleanups:
1460 - allow to declare more than one timed section per file (fix a stupid bug)
1461 - move some private declaration to the right place
1462 - merge conditional execution and timing macros into emulation module
1463 - document the module
1464 - make sure the module cleanups its mess on gras_exit
1465 * Documentation improvements:
1466 - (new) how to compile applications using GRAS
1467 - (new) emulation support (timing macros)
1469 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1471 SimGrid (2.93) unstable; urgency=low
1473 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1476 - Use Paje properly where used. Still to be sanitized properly.
1477 - Portability fix: Add an implementation of the contexts using pthread
1481 - Add xbt_procname(): returns the name of the current process.
1482 Use it to show the current process's name in all logging.
1484 - fix detection of older flex version and the reaction, since we do
1485 depend on modern ones (we use lex_destroy)
1486 - Better separation of SG and RL in the libs: remove all simulation code
1487 from libgras. As a result, this lib is now only 200k when stripped.
1488 Some of the xbt modules may also be duplicated (two sets and such) and
1489 should be cleaned/killed before SG3.
1490 - Insist on using xlC on AIX because of weird problems involving gcc there.
1491 - Cleanup the make remote stuff. This is now done by scripts
1492 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1493 mainly for our private use, but we're working on changing them to user
1496 - Bugfix: flush the socket on close only if there is some *output*.
1497 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1499 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1502 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1504 SimGrid (2.92) unstable; urgency=low
1506 Alpha 3 on the path to SimGrid 3
1510 - New! First try of benchmarking macros.
1511 - New! First try so that gras_stub_generator generate deployment and
1512 remote compilation helpers.
1514 - Bugfix: Initialization fix in msg_test.
1518 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1520 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1522 - API Change: gras_os_sleep to take the amount of seconds as a double.
1523 Accepting an int was error prone since it was the only location where
1524 seconds were coded as such. It leaded to damn rounding errors.
1525 - Bugfix: Hard to belive that timers ever worked before this.
1527 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1529 SimGrid (2.91) unstable; urgency=low
1531 Alpha 2 on the path to SimGrid 3
1535 - Bug fix in the lmm_solver.
1537 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1538 through the function MSG_paje_output.
1539 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1540 - It is possible to bound the rate of a communication in MSG with
1541 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1542 to write it in the changelog).
1543 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1548 - Various cleanups to the autotools stuff
1549 - Begin to move Gras examples to examples/gras/
1550 - Let make distcheck work again (yeah!)
1552 - documentation overhauled using doxygen.
1553 gtk-doc-tools is dead in SimGrid now.
1554 - Automatically extract all existing logging categories, and add the list
1555 to the documentation (long standing one, to say the less)
1557 - Cleanup the known architecture table. Reorder the entries to group what
1558 should be, and use a more consistent naming scheme.
1559 (some of the test dataset are still to be regenerated)
1560 - New! Allow library to register globals on each process just as userdata
1562 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1563 lookup time (for now).
1564 Use it in msg and trp.
1565 This cleans a lot the internals and helps enforcing privacy of the
1566 headers between the gras components.
1567 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1568 - Bugfix: gras_os_time was delirious in RL.
1569 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1571 - Reenable GRAS now that it works.
1573 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1575 SimGrid (2.90) unstable; urgency=low
1577 Alpha 1 on the path to SimGrid 3
1579 * It is a long time since the last release of SimGrid. I'm sorry about
1580 that but as I had told you, I was rewriting a lot of things. I apologize
1581 to those who had been reporting bugs to me and that I had not answered.
1582 If your bug is still in the new version, please tell me. Here is a
1583 summary of the main changes.
1585 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1586 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1587 portability, speed, and a lot more but you'll figure it out later.
1588 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1589 I wanted to release the new MSG as soon as possible and I have
1590 broken GRAS, which is the reason why, for now, only MSG is fully
1591 functional. A laconic description of these projects is available
1592 in the documentation.
1594 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1595 kernel optimized for our needs (called SURF but only the developers
1596 should use it). Hence, MSG is now roughly 30 times faster and I think
1597 that by rewriting a little bit MSG, I could event speed it up a little
1598 bit more. Beside the gain in speed, it is also much easier to encode a
1599 new platform model with SURF than it was with SG. More to come...
1601 * REVOLUTION 3: I have tried to change a little as possible the API of
1602 MSG but a few things really had to disappear. The main differences
1603 with the previous version are :
1604 1) no more m_links_t and the corresponding functions. Platforms are
1605 directly read from a XML description and cannot be hard-coded
1606 anymore. The same format is used for application deployment
1607 description. The new format is described in the documentation.
1608 Have a look in tools/platform_generation. There is a tiny script
1609 that converts from the old platform format to the new one. Concerning
1610 the application deployment format, parsing the old one is tricky.
1611 I think most of you should however be able to convert your files. If
1612 it is really an issue, I can write a C code that does the conversion.
1614 2) the toolbox tbx does not exist anymore. We now have a library
1615 with much more data-structures but without the hash-tables (we have
1616 dictionaries that are much faster).
1618 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1620 *****************************************************************************
1621 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1622 * don't feel like converting the oldies. (Mt) *
1623 *****************************************************************************
1626 Version 2.90: "the long awaited one"
1627 - Finished rewriting and debugging MSG. Rewrote the documentation.
1628 - disable GRAS for now since it needs to be ported to the newest SG
1631 - Finish the port to windows (using mingw32 for cross-compile)
1634 - Main loop and datastructures of SURF. A cpu resource object is
1635 functional. Surf can thus be used to create cpu's with variable
1636 performance on which you can execute some actions.
1638 2004-11-15 Martin Quinson
1639 - Port to ARM. Simply added the alignment and size descriptions. Should
1640 work, but the ARM machines are so slow that I didn't had the opportunity
1641 to 'make check' over there yet.
1643 2004-11-15 Arnaud Legrand
1644 - Trace manager now written. It uses a heap structure and is therefore
1645 expected to be efficient. It may however be speeded up (particularly
1646 when many events occur at the same date) by using red and black
1647 trees. One day maybe...
1648 - Max-min linear system solver written. It uses a sparse matrix
1649 structure taking advantage of its expected use. Most operations are
1650 O(1) and free/calloc are called as few as possible. The computation of
1651 the minimum could however be improved by using a red and black tree
1654 2004-11-03 Arnaud Legrand
1655 - Rename every gras_* function that was in xbt/ to its xbt_
1657 - Add a heap and a doubly-linked list to xbt
1658 - Added a dichotomy to the dictionaries. make check works as well before
1659 so I assume that the patch is correct. I do not know however if things
1660 run effectively faster than before now. :)
1662 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1663 SimGrid, and the version number is bumped to 2.x
1665 2004-10-29 Martin Quinson
1666 - Introduction of the remote errors.
1667 They are the result of a RMI/RPC on the remote machine.
1668 ErrCodes being scalar values, you can't get the host on which those
1669 errors did happen. Extending the error mechanism as in Gnome is possible.
1670 No idea yet whether it is a good idea.
1672 2004-10-28 Martin Quinson
1673 - Interface revolution: the Starred Structure Eradication.
1674 I used to do typedef struct {} toto_t; and then handle *toto_t.
1675 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1676 - ???_t is a valid type (builded with typedef)
1677 - s_toto_t is a structure (access to fields with .)
1678 - s_toto is a structure needing 'struct' keyword to be used
1679 - e_toto_t is an enum
1680 - toto_t is an 'object' (struct*)
1682 typedef struct s_toto {} s_toto_t, *toto_t;
1683 typedef enum {} e_toto_t;
1684 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1685 s_toto_t) is private.
1687 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1688 so that it can set the variable to NULL. It was so for dicts and sets,
1689 it changed for dynars.
1691 - Fix a bunch of memleaks in dict_remove
1692 - Fix a bug in sg/server_socket opening: it failed all the time.
1694 2004-10-07 Martin Quinson
1695 - Speed up dynar lookup operation a bit.
1697 gras_dynar_get is dead.
1699 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1700 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1701 (which gives you the address of the stored data).
1703 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1704 copy of the data using an affectation to do the job and not a memcpy.
1706 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1707 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1709 It does not really speedup the dynar test because they are
1710 setting elements all the time (and look them seldom). But the dict does
1711 far more lookup than setting.
1713 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1715 2004-10-05 Martin Quinson
1716 - Allow to (en/dis)able the cycle detection at run time.
1718 Whether we should check for cycle or not is now a property of each
1719 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1720 datadesc_cycle_unset allow to remove this property when previously set.
1722 Note that the cycle detection is off by default since it impacts the
1723 performance. Watch the data you feed GRAS with ;)
1725 This property is hereditary. Any element embedded in a structure having it
1726 set have it set for the time of this data exchange.
1728 You should set it both on sender and receiver side. If you don't set it on
1729 sender side, it will enter an endless loop. If you forget on receiver
1730 side, the cycles won't be recreated after communication.
1732 - Header reorganization.
1733 Kill gras_private.h, each submodule must load the headers it needs.
1735 2004-10-04 Martin Quinson
1736 - Interface revolution: do not try to survive to malloc failure.
1738 Now, gras_malloc and friends call gras_abort() on failure.
1739 As a conclusion, malloc_error is not a valid error anymore, and all
1740 functions for which it was the only gras_error_t return value are
1741 changed. They now return void, or there result directly.
1742 This simplify the API a lot.
1744 2004-09-29 Martin Quinson
1745 - Re-enable raw sockets.
1746 Created by gras_socket_{client,server}_ext;
1747 Used with gras_raw_{send,recv}
1750 It should allow to kill the last bits of gras first version soon.
1752 This is not completely satisfactory yet (duplicate code with
1753 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1756 - Simplify transport plugin (internal) interface by not passing any
1757 argument to _server and _client, but embedding them in the socket
1760 2004-09-28 Martin Quinson
1761 - Finish the port to AIX.
1762 autoconf was my problem (segfault within the malloc replacement
1763 function. No idea why)
1765 2004-09-16 Martin Quinson
1766 - Fix some size_t madness on 64bit architectures.
1768 2004-09-08 Martin Quinson
1769 - Reduce the number of system headers loaded, overload some more system
1770 calls (such as malloc to cast the result of the system one, and work
1772 - Fix and reintroduce the config support
1774 2004-09-07 Martin Quinson
1775 - Source code reorganization to allow Arnaud to surf all over there.
1776 - Allow to document the logging categories.
1777 - Remove all uppercase from logging categories and useless cleanup in names.
1779 2004-08-18 Martin Quinson
1780 Version 0.6.2 (protocol not changed; API changed)
1781 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1782 gras_error_t), and NULL when not found. Functions expecting a msgtype
1783 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1784 hopefully usefull message.
1785 - Portability to prehistoric sparcs again
1787 2004-08-17 Martin Quinson
1788 Version 0.6.1 (protocol not changed; ABI not changed)
1789 - prealloc some buffers to speed things up
1791 2004-08-11 Martin Quinson
1792 Version 0.6 (protocol not changed; ABI expended)
1793 - The parsing macro can deal with the references, provided that you add
1794 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1796 2004-08-09 Martin Quinson
1797 Version 0.5 (protocol not changed; ABI changed)
1798 - Allow to off turn the cycle detection code in data exchange at
1799 compilation time. It should be at run time, but I'm short of time (and
1800 the config stuff is still broken). That way, we keep dict out of the
1801 critical path, which is good because the performance is poor:
1802 - search not dichotomial yet
1803 - dynar give no way to access their content and memcpy everytime
1804 - In composed data description (struct, ref and so on), stop foolness of
1805 keeping the subtype's ID, but store the type itself. This keeps sets out
1806 of the critical path, which is good since they rely on dynar and
1807 dictionnaries. The only loose of that is that we cannot detect the
1808 redeclaration of a structure/union with another content (but I'm not sure
1809 the code detected well this error before anyway). We still can detect
1810 the redefinition discrepancy for the other types.
1811 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1812 it breaks the code because of type-punning used all over the place).
1813 This breaks on all non-gcc architectures (for now).
1815 All those changes (plus the buffer of last time) allow me to gain 2 order
1816 of magnitude on cruel tests consisting of 800000 array of integers on two
1817 level of a hierarchical structure (200 secondes -> 4 secondes)
1820 - the selector of reference must now return the type it points to, not
1821 the ID of this type.
1823 2004-08-06 Martin Quinson
1824 Version 0.4 (protocol changed; ABI not changed)
1825 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1826 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1827 - Implement a buffer transport, and use it by default (it relies on tcp in
1828 real life and on sg in simulation).
1829 That's a bit hackish since I had a new field to the structure to store
1830 its data without interfering with the subtype ones. Inheritance
1831 is tricky in C. And that's a kind of reverse inheritance with one class
1832 derivating two classes. Or maybe a game with java interfaces. Anyway,
1833 that's damn hard in C (at least).
1834 Moreover, I got tired while trying to ensure plugin separation and
1835 genericity in SG mode. MSG wants me to do weird things, so let's go for
1836 cruel hacks (temporarily of course ;).
1837 See comment in transport_private.h:71
1838 - do not include all the _interface headers in private but in the files
1839 which really need them (to cut the compilation time when they are
1842 2004-07-26 Martin Quinson
1843 Version 0.3 (protocol not changed; ABI changed)
1844 - Major overhault of the datadesc interface to simplify it:
1845 - shorted the function names:
1846 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1847 - add a trivial way to push/pop integers into the cbps without malloc.
1848 This allows to make really generic sub_type description, which simply
1849 pop their size of the stack.
1850 - add a function gras_datadesc_ref_pop_arr() which does what users want
1851 most of the time: Declare a dynamic array (which pops its size of the
1852 stack) and declare a reference to it. Poor name, but anyway.
1853 - kill the post-send callback, add a post-receive one
1855 2004-07-23 Martin Quinson
1856 Version 0.2 (protocol changed; ABI changed)
1857 - add some testing for cpbs in the test cases, and fix some more bugs.
1858 This invalidate again the little64 data file, since I cannot regenerate
1860 - remove an awfull optimization in the logging stuff, allowing me to:
1861 - understand it again
1862 - learn gcc how to check that the argument match the provided format
1863 - fix all errors revealed by gcc after that
1864 - internal keys of dict are not \0 terminated. Deal with it properly in
1865 loggings instead of segfaulting when the user want to see the logs :-/
1867 2004-07-22 Martin Quinson
1868 - Fix some stupid bug preventing cbps (callback postit) from working
1870 2004-07-21 Martin Quinson
1871 - Some documentation cleanups
1872 - remove the useless last argument of msgtype_declare
1873 - rename the Virtu functions to fit into the 'os' namespace
1874 - move headers src/include -> src/include/gras/ and stop fooling with
1875 gras -> . symbolic link
1876 - make distcheck is now successful
1878 2004-07-19 Martin Quinson
1880 - Build shared library also
1881 - Install html doc to the right location
1882 - stop removing maintainer files in make clean
1883 - build tests only on make check
1885 2004-07-13 Martin Quinson
1887 - No major issue in previous version => change versionning schema
1888 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1889 corresponding dataset.
1891 2004-07-11 Martin Quinson
1893 - Get it working with any kind of structure (we can compute the padding
1894 bytes remotely for all the architectures I have access to)
1895 - Implement the structure parsing macro (still not quite robust/complete)
1896 - Improvement to the remote testing toysuite
1898 2004-07-10 Martin Quinson
1899 [autoconf mechanism]
1900 - get ride of a bunch of deprecated macros
1901 - actually run the test for two-compliment, not only compile it :-/
1902 - test whether the structures get packed (and bail out if yes. Damn.
1903 Alignment is a serious matter)
1904 - test whether the structures get compacted (but respecting the alignment
1905 constraints of each types)
1906 - test whether the array fields of structures can straddle alignment boundaries
1908 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1909 (took me 2 hours to find that bug, looking at the wrong place)
1910 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1911 before sending/receiving and are used to compute the offsets of fields
1912 - Given that padding size depend even on compiler options, keep track of
1913 alignment and aligned_size only for the current architecture. Not a big
1914 deal since we send structure fields one after the other (seems
1916 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1917 program, let it work on linux/gcc/little32. portability todo.
1919 2004-07-08 Martin Quinson
1920 - import and improve remote compilation support from FAST
1921 - make sure make check works on half a dozen of machines out there
1923 2004-07-07 Martin Quinson
1924 Let's say it's version 0.0.3 ;)
1925 - Implement conversions (yuhu!)
1926 - Let it work on solaris (beside conversion, of course)
1927 - Stupid me, using rand() to generate the conversion datatests in not wise.
1929 2004-07-06 Martin Quinson
1930 - Let make dist work, since I'm gonna need it to compile on remote hosts
1931 - Let Tests/datadesc_usage write the architecture on which the file was
1932 generated as first byte.
1933 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1934 architecture descriptions.
1935 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1936 on those architectures.
1937 - Optimization: send/recv array of scalar in one shoot
1939 2004-07-05 Martin Quinson
1940 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1942 - Plug a whole bunch of memleaks
1943 - each process now have to call gras_{init,exit}. One day, their log
1944 settings will be separated
1945 - Continue the code factorisation between SG, RL and common in Transport.
1947 2004-07-04 Martin Quinson
1949 - Redistribution between SG and RL.
1950 We wanna have to accept in SG, so move accepted related parts of RL in
1951 the common part. (more precisely, the dynar of all known sockets is no
1952 more a static in transport.c, but part of the process_data)
1954 [gras_stub_generator]
1955 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1957 2004-07-03 Martin Quinson
1958 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1959 them all up in one shot)
1961 - Fix the ugly case of reference to dynamic array.
1962 - New (semi-public) function gras_datadesc_size to allow the messaging
1963 layer to malloc the needed space for the buffer.
1965 - gras_socket_close now expect the socket to close (and not its address to
1966 put NULL in it after it). This is because the socket passed to handlers
1967 is one of their argument (=> not writable).
1969 - propagate the interface cleanup from last week in datadesc, ie remove a
1970 superfluous level of indirection. User pass adress of variable
1971 containing data (both when sending and receiving), and not of a variable
1972 being a pointer to the data. Let's say that I like it better ;)
1973 The price for that is constructs like "int msg=*(int*)payload" in
1974 handlers, but it's a fine price, IMHO.
1976 - Let it work in RL (yuhu)
1978 2004-06-21 Martin Quinson
1980 - porting SG plugin and SG select to new standards (works almost).
1981 - plug memleaks and fix bugs around.
1984 - cleanup the prototype of data recv and force users to specify when they
1985 want to handle references to objects. Test case working even for cycles.
1986 - plug memleaks. Valgrind is perfectly ok with this.
1988 2004-06-12 Martin Quinson
1990 - cleanup the separation between plugin and main code in plugin creation
1992 2004-06-11 Martin Quinson
1994 - Reput hook for raw sockets, needed for BW experiments
1995 - kill a few lines of dead code
1996 [Data description] Interface cleanup
1997 - gras_datadesc_by_name returns the searched type or NULL.
1998 That way, no variable is needed to use a type desc once, which makes
2000 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
2001 two parameters were strings (field name, type name), leading to
2003 [Dicos] Interface cleanup
2004 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
2005 This is consistant with the dynar API.
2007 2004-04-21 Martin Quinson
2009 - Porting to new standards.
2011 - interface cleanup.
2012 There is no bag anymore, no need to take extra provision to mask the
2013 pointers behind "ID".
2014 Better splitup of functions between files create/exchange/convert.
2015 This is still a bit artificial since convert and receive are so
2016 interleaved, but anyway.
2018 - add a queued message list to procdata (the ones not matching criteria
2020 - factorize some more code between SG and RL wrt procdata
2022 - use gras_exit in example to track memleaks
2023 - get rid of gs_example now that GS is properly integrated into gras
2024 - update run_test to integrate the lastest tests (datadesc)
2026 - rename WARNINGn macros to WARNn since it prooved error-prone
2028 2004-04-19 Martin Quinson
2030 - register init/exit functions within gras module mechanism
2031 - send/receive function.
2032 Convertion is not implemented, but short-cutted if not needed.
2033 struct/array elements are sent one by one (instead of block-wise), but
2034 nobody really cares (yet). Get a prototype before optimizing.
2035 - tests (using a file socket) for DD send/receive on:
2036 - base types: int, float
2037 - array: fixed size, string (ie ref to dynamic string)
2038 - structure: homogeneous, heterogeneous
2039 - chained list, graph with cycle
2040 Believe it or not, valgrind is not too unhappy with the results. The
2041 cycle happily segfaults, but the others are ok. And I'm sick of pointers
2045 - Bugfix when using a filename explicitely (instead of '-')
2047 2004-04-09 Martin Quinson
2049 - factorize more code between RL and SG in socket creation
2050 - Complete the implementation and tests of:
2052 o file (only in RL, and mainly for debugging)
2054 I lost 3 days to design a portable address resolver, and then decided
2055 that the prototype mainly have to run on my box.
2056 Addressing portability too early may be like optimizing too early :-/
2058 - use gras_init in the Tests instead of the crappy parse_log_opt
2059 (the latter function is removed)
2060 [Conditional execution]
2061 - New functions: gras_if_RL/gras_if_SG (basic support for this)
2062 [Code reorganisation]
2063 - Get rid of libgrasutils.a since it makes more trouble than it solves.
2064 Build examples against the RL library, since there is no way to disable
2065 its creation for now.
2067 For information, the beginning of coding on GRAS was back in june
2068 2003. I guess that every line has been rewritten at least twice since