1 SimGrid (3.6) stable; urgency=medium
3 The Summer Release, also known as the "F*! They Killed Kenny!" version
6 * Bindings now constitute their own package, separated from the main one.
7 Rationale: reduce our maintainance nightmare by reducing the module coupling
8 They will soon be released on their own on gforge.
10 svn co svn://scm.gforge.inria.fr/svn/simgrid/contrib/trunk/simgrid-java
11 svn co svn://scm.gforge.inria.fr/svn/simgrid/contrib/trunk/simgrid-ruby
13 GRAS: It is not considered as stable anymore, but experimental. Sorry.
14 * It's not quite deprecated for now because we have no replacement,
15 but it may soon become the case.
18 * New MPI functions supported: MPI_Comm_disconnect, MPI_Comm_get_name
19 * Fortran: New user-level cache variable to store the rank of the running
20 process. This improves performance by an order of magnitude.
21 * C: New coccinelle script to automatically locate and modifiy global and
22 local static variables.
23 * Improved SMPI network model with a sender-side gap to account for multiple
27 * New function MSG_comm_get_status(). MSG_comm_test() and MSG_comm_testany()
28 only say if a communication is finished, no matter whether it succeeded or
29 failed. You can call MSG_comm_get_status() to know the status of a finished
31 * New function MSG_task_dsend() to send a task and detach it. When a
32 communication is detached, you are never notified of its success or failure
33 and the memory is released automatically once it is finished. This function
34 is useful when you don't care about the end nor the success of a
36 * Change the prototypes of action replay. Sorry for inconvenience,
37 but this is really more efficient this way (and to adapt your code,
38 you just have to fix the initialization, that shouldn't be too long)
39 * Kill the braindead MSG_task_refcount_dec() function. I guess nobody
40 ever managed to do anything useful with it.
41 * New function MSG_comm_testany(). Similarly to MSG_comm_waitany(), it
42 takes a dynar of communications. It returns immediately and gives the
43 index of a finished communication (if any).
44 * New example: a basic implementation of the Chord P2P algorithm.
47 * New model for multi-core CPUs. You can now use the core attribute to
48 precise the number of cores of a host. This is a basic model. Every
49 process running on the host receives at most the power provided in
50 the DTD (throughput<=power). Total throughput of process cannot exceed
52 * New peer tag. This peer tag creates a tiny AS comprising a host and a
53 router linked by an up-link and a down-link (possibly asymmetrical).
54 This kind of pattern allows to easily build last-mile model style platforms.
55 Aggregating such patterns in a rule-based AS is thus the technique of
56 choice for modeling large peer-to-peer/volunteer computing/cloud platforms.
57 * New model for Vivaldi routing. We transformed the Vivaldi network model
58 into a Vivaldi routing model (based on the rule-based model). This allows to
59 combine Vivaldi based latencies with last-mile platforms.
62 * Introduce a new context factory "raw", highly inspirated from the
63 ucontext factory, but using manually crafted functions in assembly to
64 do the work in an efficient manner.
65 * Allow to change the used context factory at run time, not only at
66 compilation time. Use --cfg=contexts/factory:raw for maximal speed.
67 * Add an option --cfg=contexts/stacksize:N to set the stack size of the user
68 contextes at runtime (only with raw contexts or ucontexts).
69 * Completely rewrote this module to allow parallel execution of user
70 processes. Use --cfg=contexts/nthreads:N to execute user processes
71 with N parallel threads (the default is 1, meaning no parallelism).
72 * Allow to decide dynamically between sequential and parallel modes.
73 When nthreads > 1, you can use --cfg=contexts/threshold:P to run the user
74 processes in parallel only when their number is greater than or equal to P
76 * Added a check for NaN of IEEE754 infinite in the double entries of
80 * New command line option: if you pass --cfg=verbose-exit:0, SimGrid
81 won't output the state of processes when interrupted with Ctrl-C
82 * Add a new function xbt_dynar_to_array that transforms a dynar into a
83 NULL-terminated array. This may solve backward compatibility issues
84 due to the change to return type of SD_simulate. See also:
85 http://lists.gforge.inria.fr/pipermail/simgrid-user/2010-December/002206.html
86 * Add new macros with variable number of arguments.
87 - in xbt/log.h: XBT_DEBUG, XBT_VERB, XBT_INFO, etc.
88 - in xbt/asserts.h: xbt_assert
89 - in xbt/cunit.h: xbt_test_{add,fail,assert,log}
90 - in xbt/ex.h: THROWF and RETHROWF.
91 Define XBT_USE_DEPRECATED if you want to use the old numbered macros like
93 * Change xbt_die() to accept a format string with arguments, just like printf.
94 * New data structure: xbt_lib_t, like a dict but more general and with better
98 * New configuration options
99 Options triva/categorized and triva/uncategorized can be used to generate
100 graph configuration files for Triva visualization tool.
101 * Configuration option tracing/platform is renamed to tracing/categorized
102 * XBT logging makes tracing error checks easier, new root log hierarchy: instr
103 * New TRACE_user_link_variable interface:
104 User provides the name of the link and the tracing variable to attach to it
105 * the declaration of tracing categories must be done after the environment
107 * simpler tracing interface, just one way to declare categories
108 TRACE_category or TRACE_category_with_color, it is up to you
109 * links in the trace file are again identified by their names
110 * trace contains the full platform hierarchy exactly as declared using the ASes
111 * Options tracing/msg/[task|process]:1 groups the process by hosts
112 for both cases, tasks and processes must have names that are unique during the simulation
113 these options generate traces that are suited to gantt-charts, such as the space-time view of Paje
114 * The experimental option tracing/msg/volume is deprecated
115 its functionality may be reincorporated if needed
117 The tracing generates a trace file with unordered timestamped events,
118 because of the way the core simulator (surf) works. A script available
119 at the tools directory (fix-paje-trace.sh) can be used to put the events
120 in order. We have changed the tracing so it can generate ordered timestamped
121 events in the final trace, but depending on the simulator (and how much time
122 is simulated) that can lead to a huge memory utilization. It is desactivated
123 by default, but it can be activated using the --cfg=tracing/buffer:1 switch.
126 * Define a SIMGRID_VERSION macro in simgrid_config.h.
127 - We are trying hard to keep the API stable, but it may happen that
128 some things change (we're a research project after all, not a
129 nuclear plant operating system). If such things should happen, you
130 could rely on that macro to adapt.
131 - current value: 30600 for 3.06.00, aka 3.6
132 * Define macro MAKE_SIMGRID_VERSION(major, minor, patch) to help building
133 a number that can be compared with SIMGRID_VERSION.
134 * Add a build option -Denable_debug (set to ON by default): when set to OFF,
135 assertions and verbose/debug logging events are disabled at compile time.
137 -- Tue Jun 21 08:57:43 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
139 SimGrid (3.5) stable; urgency=medium
142 * New feature to any SimGrid-based simulator: Model-Checking
143 Check SIN#1 for more details.
146 * New Model SMPI (three-interval linear regression for correction factors)
147 See RR-7426, available at http://hal.inria.fr/inria-00527150
148 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
149 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
150 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
151 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
152 * See also src/smpi/README
157 - Tracing API changes: TRACE_start and TRACE_end should not be called
158 by user-code. They are automatically called by simulators created
159 with SimDAG, MSG and SMPI if the toolkit is compiled with
160 tracing_enabled=ON. Categories declaration and utilization remain the
161 same for MSG and SimDag.
162 - A function was added to the tracing API to declare categories with
164 - TRACE_category_with_color (char *category, char *color)
165 where color must be in the following format
166 "%f %f %f", red, green, blue
167 and red, green, blue are float values in the interval [0, 1]
168 - User can specify NULL as color parameter, or continue calling
170 On that case, the tracing system will define random colors
171 - The following command-line options are supported:
172 --cfg=tracing/filename:msg.trace
173 --cfg=tracing:1 (activate tracing, needed to use others)
174 --cfg=tracing/platform:1 (categorized resource use)
175 --cfg=tracing/uncategorized:1 (uncategorized resource use)
176 --cfg=tracing/msg/task:1 (task creation)
177 --cfg=tracing/msg/process:1 (process creation, migration)
178 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
179 --cfg=tracing/smpi:1 (SMPI interface tracing)
180 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
181 - examples of examples/msg/tracing updated
183 - DAXLoader and DOTLoader functions can generate tasks with categories
184 - A new function to attribute a category to SD tasks:
185 TRACE_sd_set_task_category (SD_task_t task, char *category)
186 * Tracing the MPI interface implemented by SMPI
187 - Collective operations are traced with states
188 - Point-to-Point operations are traced with states/links
189 - Tracing activated by a parameter "-trace filename" passed
190 to smpirun during execution (considering that simgrid
191 is compiled with tracing enabled)
192 - To run the simulation with gdb, the simulator
193 accepts --cfg=tracing/smpi:1 to trace SMPI
194 - tesh files to check if smpi tracing is ok
195 - See examples/smpi/NAS/DT-trace
196 * GTNetS tracing re-worked
197 - adaptation to the tracing system of GTNets to cope
198 with modifications regarding the fullduplex mode
199 - new tesh files to check if gtnets tracing is ok
202 * Asynchronous communications through the functions:
203 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
204 * New function: MSG_load_platform_script()
205 to make possible using a lua script instead of XML files to set up platforms
206 * New function: MSG_set_function
207 to associate functions to processes, used when bypassing the parser
208 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
210 Platforms: Add some more examples in examples/platforms
211 * Grid'5000: see www.grid5000.fr
212 * *_30000_hosts.xml: various huge files [mainly scalability testing]
215 * Change the XML format. This is a very important modification. SimGrid
216 3.5 introduces a new hierarchical format based on the notion of
217 Autonomous Systems. Compatibility with old format is ensured
218 through the perl script provided in the install bin directory
219 bin/simgrid_update_xml.
220 It is now possible to build platforms with specific routing
221 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
222 connect several platforms together. We will try to provide soon
223 set of realistic platforms exploiting these properties (have a
224 look at examples/platforms/ for the moment).
225 * Take the opportunity of the XML format change to be a good XML citizen:
226 rename link:ctn to link_ctn and similar changes (also dealed with by
228 * Add a new routing scheme (rule-based) using regular expressions. It
229 enables to have an extremely low memory footprint when the
230 underlying routing is simple and can be compactly described. You
231 need to have libpcre4-dev (perl regular expressions) installed if
232 you want to use this routing scheme.
233 * Revive the cluster TAG and allow to easily and efficiently (both in
234 term of memory and speed) connect clusters together. Have a look
235 at teshsuite/simdag/platforms/ to see how this can be done. With
236 this tag, you can create clusters with thousands of tasks at no
237 cost (have a look at examples/platforms/).
238 Note: clusters are implemented as ASes, so there is no need for an
239 enclosing AS tag if you have only one cluster in your platform.
240 * Add new generic functions in the public interface that allows the user
241 to call SURF 'create_resource' methods from your code (same
242 functionality as the XML bypass mechanism but with a much lighter
244 * Add a new model (enabled through command line --cfg=network/model:SMPI)
245 that uses a piecewise linear approximation to produce better
246 results when exchanging small messages.
247 * Add a new parameter to handle correctly full duplex link and account
248 for interferences between uplink and downlink communications
249 (activate with --cfg=fullduplex:1).
252 * Rename the SD_READY (all dependencies are satisfied and task is
253 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
254 dependencies are satisfied) state.
255 This prevents a confusion between the notion of "ready to schedule"
256 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
257 simulated" (SD_RUNNABLE) used by the simulation kernel.
258 * Change the way a task is considered as ready. Instead of removing
259 dependencies when a task is done, a counter is decreased. This way,
260 it is always possible to reach ancestors thanks to the
261 SD_taks_get_parents function (even after the end of the simulation.)
262 * Change the return type of SD_Simulate from (SD_task_t*) into
263 xbt_dynar_t. This function was in handling a dynar internally and
264 converted it into a NULL terminated array for historical reasons.
265 * New function SD_dotload(char*) to load a DAG described in dot
266 format. This loader and the corresponding examples require the
267 installation of the graphviz library.
268 * Fix a bug in the management of tasks of size 0 in the surf network
269 models. This problem was only visible with SIMDAG and you should
270 thus disregard results produced with earlier versions if you
271 relied on this feature (some tasks were blocked because of this).
272 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
273 even though your applications were doing only point-to-point
274 communications and sequential computations. Now you can really use any
275 model you want (of course, if you create real parallel tasks, which are
276 not implemented in most models beside ptaskL07, this will abort).
277 * Add an example that schedules a DAX on an heterogeneous platform
278 using a Min-Min strategy.
279 * New function SD_workstation_get_current_task() that returns the kind
280 of task currently running on a workstation in the sequential access
282 * Raise some warnings when unexecuted tasks remains at the end of the
283 simulation. This is usually caused by cycles in the DAG.
286 * New function: SIMIX_process_set_function() called by MSG_set_function
287 * Change the underlying waiting queue in semaphores so that a process
288 can wait on several of them simultaneously (as in waitany).
289 * Fix the way to handle tokens in semaphores so that all access patterns
290 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
291 * kill the dirty pimple SIMIX_message_sizes_output()
292 Please use (proper) visualization instead
295 * New data container: setset (set of sets of elements)
296 * New module: mmalloc (mapped malloc, allowing to have several
297 independent segments of malloc)
298 * New function: xbt_dict_cursor_set_data()
299 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
300 * New function: xbt_dynar_is_empty()
301 * New function: xbt_fifo_get_last_item()
302 * Fix xbt_dynar_shrink(): use the right element size.
303 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
304 uninitialized areas during expand.
305 * Fix semaphores: previous implementation was severely broken.
306 * Use library init/fini functions for our initialization.
307 - you can use logs and other feature as soon as you want in your
308 code (even before the xbt_init / MSG_init)
309 - xbt_exit is now a no-op and produce a warning when used.
312 * Port GRAS to new SIMIX mechanisms. This allows gras users to
313 benefit from the latest improvement to the simulation kernel.
314 * Kill measurement sockets for now. If you rely on them, sorry. This
315 release is not for you. This feature will be reintroduced in the
316 future, but we cannot delay the release any further.
317 * New function: gras_msgtype_get_name().
318 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
319 * Fix (at last) the pmm example: it should not randomly fail anymore.
321 Build chain: bug fixes and overall polishing
322 * Cmake is now stable enough. Hence, we killed the autotools.
323 * Port to windows ( TM :)
324 * Fix the 'make install' target.
325 No need to use 'make install-simgrid' anymore
326 * Introduce a 'make dist' target compiling a *source* archive
327 'make package' compiles a binary archive
328 * Compile java files only on need
329 * Add --cd and --setenv command line options to tesh
330 * Out of source builds are not fully supported yet, but we are close
331 * Enable supernovae and optimization flags by default for our users
334 * Add layer to set up environment directly from lua, without XML.
335 * The effect of gras_stub_generator can be achieved through
336 lua too (check examples/gras/console/ping_generator.lua)
338 -- Wed, 01 Dec 2010 22:09:23 +0100 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
340 SimGrid (3.4.1) stable; urgency=low
342 The "Polishing easter eggs is probably a good idea" release.
343 This is a bug fixes release only.
347 * Fix a bug preventing the tasks from begin garbage collected.
350 * Fix a bug occuring when a host involved in a communication fails.
351 This was not detected properly by the other peer involved in the
352 communication. Now, it's reported as a network error.
355 * Warn the user about loop dependencies in data flow of DAX files
356 * Obey the control-flow dependencies of DAX files
359 * Add option "enable_smpi" allowing to not compile SMPI.
360 Probably useful for the (Mac) users experiencing a build error here
361 * Improve the detection of lua5.1 and ruby1.8
363 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
365 SimGrid (3.4) stable; urgency=low
367 The "Easter in Cargese" release. Also known as (major changes):
369 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
370 ~> bindings were greatly improved
371 ~> new tracing infrastructure for better visualization introduced
373 * the "Welcome to configury modernity" release.
374 ~> we switched from autotools to cmake, and improved our cdash
377 A more detailled list of changes follow (full detail in svn log).
379 Java Bindings: Various Cleanups
380 * (install java-gcj-compat-dev on debian-like to use them)
381 * Remove put/get: no need to export deprecated interface in Java
382 Use send/receive instead.
383 * Cleanup the examples and add a README per directory
384 * Remove example autoDestination (that's the only way to go now)
385 * Remove example explicitDestination (was a plain copy of basic)
386 * Make JniException a runtime exception, so that there is no need to
387 declare the fact that you may encounter such a beast. I guess that
388 nobody will ever want to survive such error.
389 * Create specific errors for each MSG case of failure:
390 host failure, transfer failure, timeout, task cancelled
391 * Cleanup the exceptions that may get thrown by each function
392 * Other internal cleanups in Java bindings. Performance still bad :/
393 Ruby and Lua Bindings: create them
394 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
395 * That's new and great, you should try them out.
396 Same functionalities than Java bindings, only even less polished
398 * Kill the useless "rate" argument of SD_task_get_execution_time()
399 Everyone used to provide -1 as a value, it was not used, and the
400 semantic of a possible use wasn't even clear.
401 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
402 either as comm costs or compute costs to mean that there is no
403 such thing for that specific task.
404 * Add a SD_task_set_name() function
405 * Fix SD_task_unschedule() on typed tasks
406 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
407 * In DAX loader, accept useless 'level' attributes to <job> since
408 LIGO DAGs have them (seem to be to ease graphical representation).
410 * Add an example masterslave_mailbox.c using send/receive and not
411 the deprecated put/get interface.
412 * Kill the MSG_paje_output() function. It's a noop since 2 years.
413 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
415 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
416 (declare MSG_USE_DEPRECATED to still have the old name)
417 * Add a MSG_task_set_data() function
418 * About trace replay (see examples/msg/actions):
420 - Allow to work with splitted trace files for each process
421 Give the specific trace file as argument of each process,
422 and call MSG_action_trace_run(NULL)
423 You can still have one merged file for all processes.
424 - Fix implementation of collective operations
425 * Allow task_execute() on 0-sized tasks (closes #10063)
427 * This is the first release of SimGrid where SMPI is not considered
428 beta anymore (even if some corners should still be improved)
429 * Port over the new SIMIX_network submodule (internal refactoring)
430 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
431 * Implement more missing elements of the standard:
433 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
435 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
436 MPI_Scan() MPI_Get_processor_name()
437 - Added implementation of missing case for Alltoall (warning: it's
438 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
439 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
440 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
441 SMPI_MPI_Allgatherv()
443 - MPI_Waitsome() was broken
444 - Allow relative includes in smpicc
445 - Command line cfg argument 'reference_speed' was ignored...
446 - Some functions did not properly lead to auto-benching of user code
447 - smpicc passes -O2 by default (just like openmpi one)
449 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
450 * Bug fixes about timeouts during communications
451 * add SIMIX_message_sizes_output() as a pimple to write to file the
452 amount of messages per size. Use gnuplot to get histogram.
453 Pimple because that's the only user-visible function of simix,
454 defined directly in xbt.h (irk, sorry)
456 - Add a SIMIX_sem_get_capacity() function
457 - Fix interactions with processe resume/suspende
458 - release_forever() was stupidly broken
459 - Fix SIMIX_display_process_status() for processes in a semaphore
460 - Make SIMIX_sem_block_onto() user-visible
461 * Refactoring context stuff:
462 - Use pseudo-OOP for better modularity
463 - reimplement SIMIX_process_kill() without process_schedule() so
464 that the latter can take as invariant that it is called from
466 - Merge context_start into context_new for sake of simplicity
468 * Add a Vivaldi network model, coded live during SUD'10 ;)
469 * Rename configuration variables to start a hierarchy:
470 o cpu_model -> cpu/model
471 o network_model -> network/model
472 o workstation_model -> workstation/model
473 * New configuration variables:
474 o network/bandwidth_factor: correction to bandwith
475 o network/latency_factor: correction to latency
476 o netwotk/weight_S: correction to the weight of competing streams
477 * Add a long description to the models, that users can see with such
478 argument on the command line: --cfg=cpu/model:help
479 * --help-models display the long description of all known models
481 * config: add the ability to set a default value after registration
482 Does not override any previously set value (e.g. from cmd line)
483 * dict: allow to have integer key and data.
484 When so, you need to use the following functions
485 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
486 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
487 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
488 In contrary to regular dicts, the key is not malloced before copy.
489 Mixing scalar and regular elements in the same dict is not tested
491 * Allow to use xbt_dynar_shrink() to expend the dynar instead
492 Tracing for Visualization:
493 * SimGrid is now instrumented in order to generate a trace file for
494 visualization analysis: to use it, need to compile SimGrid with the
495 "tracing" option enabled, and instrument the program using SimGrid with
496 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
497 (among other functions).
498 * The instrumentation only traces the platform utilization for now
499 * Documentation to use the tracing functions and how to analyze the
500 traces with the Triva tool is written.
501 * More information about: SimGrid FAQ (in the section Tracing Simulations
504 * We moved to cmake as default build system. Autotools support will
505 be dropped soon. Check the FAQ for more info about how to use it.
506 * Greatly improved our cdash/ctest interactions
507 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
508 * Added memory checking tests with valgrind; lot of memleak fixing.
509 This may be the first release of simgrid with so few memory issues
510 * Added code coverage tests.
511 Our coverage is still improvable, but at least we see it on cdash.
513 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
515 SimGrid (3.3.4) stable; urgency=low
517 The "Desktop Grid needs love too" release (also called Xmas release).
520 * Major speedup in the maxmin system solving by using lazy evaluation
521 Instead of solving completely the maxmin system at each iteration,
522 only invalidate (and recompute) the modified parts.
523 This new feature is enabled in default models but you can try to
524 turn it on with "--cfg:maxmin-selective-update=1" for other models.
525 * Cas01 IMproved as default CPU model
526 This CPU model is the same Cas01 model, but it uses the
527 maxmin-selective-update flag and a heap structure to manage
528 actions on SURF kernel.
529 It reduces the complexity to find the next action to finish and,
530 consequently, it's faster than the old Cas01.
531 This is the new default CPU model (Cas01).
532 * Rename the old Cas01 model to Cas01_fullupdate
533 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
534 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
535 * CpuTI (CPU Trace Integration)
536 A new CPU model whose objective is simulate faster when using
537 availability trace files.
538 Instead of using a full featured, over engineered maxmin system for
539 CPU modeling, this model does the pre-integration of traces files
540 to calculate the amount of CPU power available, and so, executes
541 faster than the old CPU models.
542 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
543 * Use LV08 as default network model since it gives better accuracy
544 for small messages and shouldn't change things for big ones.
545 Use --cfg=network_model:CM02 to get the previous behavior.
548 ******************************************
549 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
550 ******************************************
551 * The new CPU model may changes simulations!
552 The point is that events occurring at the exact same timestamp
553 are not scheduled in the same order with the old and new
554 version. This may be enough to completely change the execution
555 of simulations in some cases.
556 * The new network model will change simulations!
557 This new model is more realistic than the previous one, so you
558 should consider redoing your old experiments with this model.
559 Sorry for the inconvenience.
562 * Introduce the supernovae compilation mode
563 When compiled that way, the whole SimGrid (or almost) is put in a
564 single compilation unit and compiled in one shoot.
565 This is to help gcc which has difficulties to inline stuff from one
567 The speedup seem to be above 15%, althrough more tests are needed on
568 amd64 to confirm that gain.
571 * Port of MSG's mailbox on top of SIMIX network
572 The put/get mechanism was greatly simplified on the way.
575 * New SIMIX network module. Provides:
576 - Mailbox: rendez-vous mecanism to find with who you want to speak
577 - Synchronous send/recv: easier and hopefully faster since the
578 logic is handled in the maestro process directly now
579 - Asynchronous send/recv: you dreamt of it? It's here now
580 Too bad that nobody cared enough to propagate the change to MSG.
581 * Add semaphores as SIMIX synchronization mechanism.
584 * new function SD_daxload(char*) to load a DAX file
585 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
586 * Introduce typed tasks. Specify its kind and cost at creation.
587 At scheduling, just give where it should be placed, and the cost
588 for each involved resource is automatically computed.
589 Existing constructors so far (more to come of course):
590 - SD_task_create_comm_e2e() for end-to-end communication
591 - SD_task_create_comp_seq() for sequential computation
592 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
593 * new function SD_task_dump() for debuging display
594 * new function SD_task_dotty(task,FILE*) writing to file the info
595 about the task in dotty format
596 * SD_task_dependency_exists() can now cope with having one of its
597 arguments NULL. If so, it tests whether the other argument has any
599 * Add getters on list of preceding/following tasks:
600 SD_task_get_parents(task) and SD_task_get_children(task)
601 * Add getters on amount of workstations and list:
602 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
603 * Add getter on task kind: SD_task_get_kind(task)
604 * Update the start_time and finish_time of tasks on completion/failure
605 * Bugfix: Remove task from state swags when destroyed
608 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
611 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
612 dynar with one separate thread per value of the dynar.
613 * Change the prototype of xbt_thread_create(), sorry.
614 Added a boolean parameter indicating whether we want to join this
615 thread (used in SG only for now)
616 * Implement xbt_thread_join and xbt_thread_yield in SG also.
619 * GTNetS wrappers should now be usable again (and betterly tested too)
620 * Fix a major regression from 3.2 where the timeout provided to
621 MSG_task_put_with_timeout() was used as absolute time before which
622 the comm should be done.
623 * Start to fix the <cluster> tag.
624 - Internal links should be good now (beside of the loopback, which
625 use the private link instead)
626 - paths to the external world is still rather broken
627 - the <route:multi> tag is just broken. Actually that's brain-dead.
628 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
629 to make it less stupid
630 ** Check your platform with teshsuite/simdag/platforms/flatifier **
631 * Fix a source-level compatibility glitch from 3.2: after defining
632 MSG_USE_DEPRECATED, you can use the old name
633 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
634 * Allow to compile from the SVN with automake 1.11
635 * Fix some problems when using the "start_time" tag in deployment XMLs.
636 * Fix #8569: XBT/synchro.h has redundant declarations
637 * Fix #8563: MSG return values and exceptions
638 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
639 * Integrate patch #8636: Obey DESTDIR when installing documentation.
640 Thanks to Robson Peixoto.
641 * Fix a vicious bug in dictionaries inducing that some elements were
642 not freed on xbt_dict_free()
644 Portability report of this version:
645 * Main portability targets:
646 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
648 - mac leopard on i386
649 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
650 but nothing critical.
651 * Other platforms: windows, AIX and others were not tested for this release
653 Timing report of this version:
654 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
655 scenario parameters). From 8h to a few seconds in desktop grid settings.
656 * Supernovae brings about 25% speedup on i386.
658 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
660 SimGrid (3.3.3) stable; urgency=low
662 The "Need for Speed" release.
664 The timings done to validate the 3.3.2 were faulty.
665 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
667 The problem was a conversion from a manually handled vector to
668 xbt_dynar_t on the critical path.
669 xbt_dynar_foreach calls functions, inducing stack management crap.
671 We inlined these functions and xbt_dynar_foreach is now breath taking.
672 We also inlined xbt_swag_belong on the way.
674 Here are some approximate speedup measurements (on master/slaves
675 simulations lasting between 10s and 20s each):
676 3.3.1 -> 3.3.2: about same performance
677 3.3.2 -> 3.3.3: 40% speedup
678 3.3.1 -> 3.3.3: 40% speedup
679 3.3.1 with inline patch -> 3.3.3: 30% speedup
681 Our reading is that the refactoring which occurred in 3.3.2 made us
682 suffer much more from the xbt_dynar_foreach low performance, but
683 once we solved this, this refactoring proved to be very performance
684 effective. From the 40% speedup, somehow, 10% are due to the
685 inlining and 30% to the refactoring.
687 That's a pitty that gcc cannot inline functions placed in other files
688 alone. We have to choose between:
689 - break the encapsulation (by putting private data structures and
690 accessors in headers files to help gcc)
691 - live with low performance
692 - switch to a decent compiler such as icc (not quite possible).
694 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
696 SimGrid (3.3.2) stable; urgency=low
698 The "Simplicity does not preceed complexity, but follows it" release.
700 The main contributors of this release were (lexical order):
701 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
704 * Extract the routing logic into its own object.
705 (was dupplicated in network.c and workstation_LV07.c;
706 Allows to implement other ways of storing that info)
707 => kill now useless network_card concept
708 - Use dynar to represent routes (instead of void** + int*)
709 - kill link_set (use surf_network_model->resource_set instead)
710 - Add a command-line option to choose the routing schema to use
711 - Add three new models:
712 * Floyd (shortest path computed at initialization)
713 * Dijikstra (shortest path recomputed all the time)
714 * Cached Dijikstra (shortest path computed on need)
715 All these models where contributed by Silas De Munck, and are
716 described in his ICCS09 paper.
718 * Simplify model declaration
719 (less redirections, less function to write when defining a model)
720 - Factorize stuff between models:
723 surf_model_resource_set(model)
724 surf_model_resource_by_name(model, name)
725 - Unify the types of models in s_surf_model_t (using an union)
726 - Embeed fields of common_public directly into s_surf_model_t
727 - Rename model methods:
728 action_free ~> action_unref
729 action_change_state ~> action_state_set
730 action_get_state ~> action_state_get
731 - Change model methods into functions :
732 (model)->common_public->action_use ~> surf_action_ref
734 * Implement a generic resource; use it as ancestor to specific ones
735 (allows to kill duplicated code in models)
736 Drawback: timer command don't need no name nor properties;
737 workstation_CLM03 don't need no properties
738 (but I guess we can live with those few bytes wasted)
740 * Improve the action object model
741 - implement a constructor avoiding dupplicated code about field
742 initialization in generic_action part.
744 * Kill the SDP model: it has an external dependency, is deprecated
745 in flavor of modern lmm models, and didn't compile since a while
748 * Relocation of the context module from XBT to SIMIX.
749 (the context were decoupled from the simix processes, duplicating a lot of code)
750 => a lot of code was factorized
751 - less overhead is introduced during scheduling
752 - simpler API for the context factory
753 - the logic for process creation,destruction and manipulation was simplified
754 * Simplification of the s_smx_process_t data structure.
755 => accesing the simix level data associated to a process is faster now,
756 and the code is a lot more readable.
759 * Implement some more MPI primitives:
760 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
761 -implementation: Bcast: flat or 2-ary tree (default),
764 Allreduce: Reduce then Bcast
765 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
766 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
767 Alltoallv: flat tree, like ompi
769 * Add support for optimized collectives (Bcast is now binomial by default)
770 * Port smpirun and smpicc to OS X
773 * Kill SD_link_get_properties: hard to maintain and makes very little sense
774 Shout out if you used it.
777 * Display the list of still queued messages in SG mode when existing
781 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
782 * Add xbt_graph_node_get_outedges() [Silas De Munck]
783 * Add xbt_str_from_file(FILE*)
784 * Add xbt_dict_get_key achieving a linear reverse search
785 * Remove the context module
787 Portability report of this version:
788 * Main portability targets:
789 - Linux(debian)/x86/context
790 - Linux(debian)/x86/pthreads
791 - Linux(debian)/amd64/context
792 - Linux(debian)/amd64/pthreads
793 On these, we still have the eratic breakages of gras/pmm and
794 amok/saturate_sg reported in previous version. We still think
795 that the tests are the cause of the fault, not the tested code.
797 - Mac OSX Leopard/x86/context
798 Still false negative in tesh autotesting.
799 Smpi still fails, but this time because readlink does not accept -f
800 Everything seems to work properly beside of that.
803 - AIX version 5.3 (only tested contexts this time)
804 Smpi still fails there because mktemp is not installed.
805 Everything seems to work properly beside of that.
807 I managed to compile it for the first time, but several breakages.
808 Won't delay the release for this exotic platform.
810 * Windows: it's still lagging behind. If you want to help, please
813 Timing report of this version:
814 This version seem to be more than 5% faster than 3.3.1 (on linux
815 64bits with contextes). The gain is less than expected, we are
816 investigating this for next release.
818 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
820 SimGrid (3.3.1) stable; urgency=low
823 * Implement a --cfg-help to show existing configuration variables
824 * Build chain do not require doxygen in maintainer mode
827 * fix a bug on struct sizeof computation, which prevented the
828 exchange of arrays of structs in some conditions
829 - added a regression test about this in datadesc_usage
830 * Allow the exchange of 0-long dynamic vectors.
831 - for that, use -1 as indicator of dynamic size instead of 0
832 - This implied to change any size from unsigned long to long,
833 reducing a bit communication abilities, but I guess that with
834 64bits being quite common, this is more than enough.
835 - This also induce a protocol change, thus bumping network protocol
836 version from 0 to 1 (if we have external users, we have to get
837 clean on that point too ;)
838 - added two regression tests about this in datadesc_usage
839 * Be more verbose when propagating local exceptions
840 This helps debugging.
841 * Display the status of simulated processes when receiving SIGINT in
845 * Allow to control the simulation from a trace file.
846 New functions MSG_action_register() and MSG_action_trace_run()
847 The first one allows to associate a function execution to each
848 kind of action while the second one parses a trace file and
849 triggers the corresponding actions within the system.
850 For now, only a toy example is provided in examples/msg/actions
851 * Add an exemple of process migration in examples/msg/migration
852 * Fix a bug in task exchange which broke MSG_task_get_sender()
853 Add a teshsuite regression test for that.
854 [Bug: if MSG_task_get_sender() is called after sender exit,
856 * Fix a bug which prevented suspend/resume to work properly
857 * Display the status of simulated processes when receiving SIGINT
858 This fixes a regression of v3.3. due to the introduction of SIMIX
859 * Bug fixing in failure management:
860 - trace could not start by a failure at time 0
861 - failure during communications were not working
864 * Add SIMIX_process_set_name() to change the name of the current
865 process in the log messages.
866 * Store smx_hosts in a dict since we only retrieve them by name
867 * Move the configuration infrastructure to surf
870 * Move the configuration infrastructure to surf
873 * Massive internal cleanups:
874 - Store internal structures on processes instead of hosts (allows
875 to have more than one process per host, in addition of being more
877 - Cleanup the initialization/finalization process
878 - Kill a whole bunch of unneeded synchronization:
879 processes run in exclusive manner within the simulator
880 - Move queues from global tables to process data fields
882 - now accept -platform and -hostfile arguments
883 - Pass the right rank value to processes according to the hostfile
884 * Compile the examples by default, and use them as regression tests
885 * Implement MPI_Wtime()
886 * Change the reference speed to a command line option
889 * TCP_gamma can now be specified as command line option using
890 --cfg=TCP_gamma:10000000.0
891 * Change the --surf-path cmd line option into --cfg=path:
894 * Also include strbuff from xbt.h public header
895 * xbt_ex_display(): do not free the exception after displaying
896 This allows to do more with the given exception afterward.
897 Users should call xbt_ex_free() themselves.
901 Portability report of this version:
902 * Main portability targets:
903 - Linux(debian)/x86/context
904 - Linux(debian)/x86/pthreads
905 - Linux(debian)/amd64/context
906 - Linux(debian)/amd64/pthreads
907 These targets fail about 1/10 of times on gras/pmm, but we believe
908 that this is because of the test, not because of simgrid.
909 amok/saturate_sg fails even more rarely, and the test may not be
912 - Mac OSX Leopard/x86/context
913 The test suite still spits tons of errors because some obscure
914 force prevents us from removing the temporary directories
915 arguing that they still contain some metadata I've never heard of.
916 Smpi fails because seq is not installed.
917 Everything seems to work properly beside of that.
920 - AIX version 5.3 (both contexts and pthread)
921 Smpi still fails there because mktemp is not installed.
922 XML inclusions seems rosty on AIX.
924 * Windows: it's still lagging behind. If you want to help, please
927 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
929 SimGrid (3.3) stable; urgency=high
933 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
936 * Introduce the SIMIX module: factorize code between MSG and GRAS.
939 Until now, GRAS were using MSG as an interface to SURF. It was
940 quite difficult because both interface have several differences
941 (MSG channels vs GRAS sockets were the most notable point).
943 This also opens the gate to SMPI (which should occur soon) and speed
944 up simulations by to 40% (even if it were not the main goal).
946 **************************************
947 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
948 **************************************
949 The point is that events occuring at the exact same timestamp are
950 not scheduled in the same order with the old and new version. This
951 may be enough to completely change the execution of simulations in
952 some cases. Sorry for the inconvenience.
954 * Cleanup and upgrade the XML format to push further scalability
955 issues (check http://hal.inria.fr/inria-00256883/ for more info)
957 * Improve the testing infrastructure with tesh. Now a very large part of
958 the code is tested not only by being run but also by checking that the
959 output match an expected output [Mt].
961 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
962 is really less memory-demanding, which should allow you to use
963 larger files in SimGrid [AL].
965 * Inform valgrind about our contextes, so that it becomes usable
966 with the default (and more effecient) version of SimGrid
967 [contributed by Sékou Diakite, many thanks]
970 * Introduce a listener thread in charge of receiving incomming
971 messages from the network. It allows to overlap communication and
972 computation but most notably, it removes some stupid deadlocks due
973 to the fact that so far, a process could not send and receive at
974 the same time. This made most non trivial communication schema
976 * Convert the PIDs from long int to int to match the MSG ones (and
978 * New function: gras_agent_spawn() to launch a new process on
979 current host. Only working in simulation for now. [Mt]
980 * New function: gras_os_hostport() returning a constant form (ie,
981 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
984 * Make the backtrace of exceptions more human readable [Mt]
985 * New module: xbt/str [Mt]
986 a ton of string utility functions (split, join, printf to a newly
987 allocated buffer, trim, etc)
988 * New module: xbt/hash [Mt]
989 SHA1 hashing algorithm (more to come if needed)
990 * New module: xbt/synchro [Mt]
991 synchronization tools (mutex and conditions) working the same way
992 in simulation and in real life (mainly useful for GRAS, but not
994 * New module: xbt/queue [Mt]
995 classical producer/consumer synchronization scheme
996 * xbt_dynar_new_sync() creates a synchronized dynar. All access
997 (using the classical functions will get serialized) [Mt]
998 * Make dictionary internal table dynamic. No need to specify its size
999 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
1001 * Make sure the log channels are organized as a tree under windows
1002 (because of ANSI C compatibility issue, any channel were child of
1003 root directly) [Mt].
1006 * Cleaned many thing in surf and fixed a few bugs [AL].
1007 * Add a nice command line configuration mechanism to compose models [AL].
1008 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
1009 the previous one (KCCFLN05). It relies on something that looks like
1010 a max-min sharing mechanism but cannot be written as such. A new solver
1011 was thus designed [AL].
1012 * Add a new solver to lmm. Based on Lagrange optimization and
1013 gradient-based descent, it enables to efficiently maximise systems s.a
1015 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
1017 This solver enables to propose two new network models for TCP Reno and
1018 TCP Vegas based on Low's work. These models still need to be fully
1019 tested though [Pedro Velho].
1022 * Bug fix in SD_simulate. Now the time bound given as argument is
1024 * Use the new parallel task model (ptask_L07) as default.
1025 * Use the SURF command line configuration mechanism.
1026 * 0-size tasks (for synchronization) should now work.
1028 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
1030 SimGrid (3.2) stable; urgency=high
1034 We still experience issues on this platform, but we believe that at
1035 least MSG is usable.
1037 GRAS API BREAKAGE (for simplification purpose, sorry):
1038 * the gras_msgtype_by_name is not used anymore. Instead of
1039 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
1040 you can write (and must)
1041 gras_msg_send(toserver, "request", &request);
1042 - If you still want to pass a gras_msgtype_t to the function (to cache
1043 the type and avoid the lookup time), use the gras_msg_send_() variant.
1044 - Impacted functions:
1045 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
1046 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
1047 * The callbacks are now expected to return 0 when everything went well
1048 (just like the main() function)
1050 GRAS new features and improvements:
1051 * New module mecanism where user code can use per process globals [Mt]
1052 This is similar to gras_userdata_*() functions, but for libraries. It
1053 factorize some code developped over and over in the examples and AMOK.
1054 It has still to be documented and used (only amok/peermanagement is
1056 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
1057 loss when they were small enough to fit into the buffer and sent quickly
1058 enough so that they can all get received in one shoot.
1059 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
1060 if not found. Use the *_or_null() variant for the old semantic.
1061 * In gras_msg_handle, do not discard messages without callback.
1062 They are probably messages to be explicitly awaited later (ie, proofs of
1063 mis-synchronization in userland since they are sent before being awaited)
1064 No big deal usually.
1065 * gras_socket_meas_send/recv: semantic changed!
1066 The numerical arguments used to be (1) the total amount of data to send
1067 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
1068 messages. This was need for the fool willing to send more than MAXINT
1069 bytes on quite fat pipes.
1072 * Do really rename the hostmanagement module to peermanagement. [Mt]
1073 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
1074 but this is rather new and this was documented in the module
1075 documentation (poor excuses, I admit)
1076 * Bandwidth measurement semantic changed! This follows the changes to
1077 gras_socket_meas_send/recv explained above.
1080 * A sequential mode has been added to the workstations. When a workstation
1081 is in sequential mode, it can execute only one task, and the other tasks
1082 are waiting in a FIFO. [Christophe Thiery]
1085 * The KCCFLN05 workstation model now handles parallel tasks. It is the
1086 model for SIMDAG. [Christophe Thiery]
1087 * Bug fix in the maxmin solver: Some values were close to 0 instead of
1088 equal to 0, which caused some bad behaviors in
1089 saturated_constraint_set_update. I now use a threshold mechanism like in
1093 * When running manually src/testall, you select specific units [Mt]
1094 testall is the result of our cunit mecanism, and should replace all
1095 the scripty thingy around since bash don't run easily on billware.
1097 * A mallocator system has been added. [Christophe Thiery]
1098 Mallocators allow you to recycle your unused objects instead of freeing them
1099 and allocating new ones.
1101 Documentation update:
1102 * FAQ reworking + New FAQs:
1103 - "Valgrind spits tons of errors!" [Mt]
1104 - "How to repport bugs" [Mt]
1105 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
1106 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
1107 - Communication time measurement within MSG [AL]
1108 - I experience weird communication times when I change the latency [AL]
1109 * GRAS tutorial [Mt]
1111 - an introduction to the framework and to the used communication model
1112 - an initiatic tour introducing the most proheminent features:
1114 . Lesson 0: Installing GRAS
1115 . Lesson 1: Setting up your own project
1116 o Part 2: Message passing
1117 . Lesson 2: Exchanging simple messages
1118 . Lesson 3: Passing arguments to the processes (in SG)
1119 . Lesson 4: Attaching callbacks to messages
1120 . Lesson 5: Using globals in processes
1121 . Lesson 6: Logging informations properly
1122 . Lesson 7: Using internal timers
1123 . Lesson 8: Handling errors through exceptions
1124 . Lesson 9: Exchanging simple data
1125 . Lesson 10: Remote Procedure Calling (RPC)
1126 . Lesson 11: Explicitely waiting for messages
1127 . Recapping of message passing features in GRAS
1128 - A HOWTO section containing:
1129 o HOWTO design a GRAS application
1130 More are due, of course. They will come latter. In the meanwhile, you can
1131 check the examples which are still here.
1133 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
1135 SimGrid (3.1) stable; urgency=high
1139 There was a stack corruption somewhere, visible only when optimizing
1140 with these versions. [Vince]
1143 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1144 module that enabled to play with Directed Acyclic Graphs. It is built
1145 directly on top of SURF and provides an API rather close to the old
1146 SG. Some old codes using SG are currently under rewrite to check that
1147 all needful functions are provided. [Christophe Thiery]
1150 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1151 extension of the classical CLM03 model that gracefully handles
1152 failures. This is now the default model for MSG and GRAS. It doesn't
1153 handle parallel tasks yet though. [AL]
1154 * Bug fix: Weights were not correctly set in the network part.
1155 WARNING: This may have resulted in incorrect results with simulations
1156 where there are more than one flow on a given link. [AL]
1159 * After a (long ?) discussion on simgrid-devel, we have decided that the
1160 convention we had on units was stupid. That is why it has been decided
1161 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1162 WARNING : This means that all previous platform files will not work as
1163 such with this version! A warning is issued to ask users to update
1165 A conversion script can be found in the contrib module of the CVS, under
1166 the name contrib/platform_generation/surfxml_update.pl [MQ]
1169 * Bug fix: Processes were started in reverse order, wrt deployment file.
1170 WARNING: if your code relies on this bug, please fix it. [AL]
1171 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1172 being executed on two different locations. [AL]
1173 * Bug fix: Failures are now better supported thanks to Derrick's tests
1174 (there was many failure situations I hadn't thought of and that weren't
1175 correctly handled). [AL]
1176 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1180 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1181 Exception occurring on server-side are propagated back to client (!).
1183 API CHANGE: the callback changed their prototype. Change:
1184 int my_handler(gras_socket_t expeditor, void *payload_data) {
1186 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1187 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1189 * New! function: gras_msg_handleall to deal with all messages arriving
1190 within a given period.
1191 * New! function: gras_socket_server_range to get a server socket in a
1192 range of port numbers (ease to avoid port number conflicts) [MQ]
1193 * New! gras processes display their backtrace when they get a SIGUSR1
1194 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1195 Sweet to debug RL processes [MQ]
1199 - Do not force experiment sizes to be expressed in kb, or it becomes
1200 impossible to measure the latency this way (needs one byte-long tests)
1201 WARNING: this changes the amok_bw_* function semantic. [MQ]
1202 - Implements the link saturation stuff. [MQ]
1203 * Peer management module:
1204 New! module factorizing code that we wrote over and over [MQ].
1207 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1208 - Test units are placed directly into the library code, they get extracted
1209 automatically and placed into the src/testall binary.
1210 - Convert most of the XBT tests to this system.
1211 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1212 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1213 live in the same C code. This required to change a little bit the API
1214 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1215 * New module: sparse graph structure with basic algorithms (this is work
1216 in progress and the API is not considered to be frozen yet). [AL]
1217 * Display more information on backtraces: source line & function names are
1218 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1219 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1220 * Reworked a little bit some #include statements to load only required
1221 headers. Some user code that relied on SimGrid to include stdlib or
1222 stdio may need to include it by themselves. [AL]
1223 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1225 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1226 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1227 but I guess that almost nobody used those parts. [MQ]
1229 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1231 SimGrid (3.0.1) stable; urgency=low
1234 * Unfortunately, I had missed 5 misnamed functions:
1235 xbt_fifo_item_t xbt_fifo_newitem(void);
1236 void xbt_fifo_freeitem(xbt_fifo_item_t);
1237 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1238 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1239 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1240 They're now deprecated. Please use their new versions:
1241 xbt_fifo_item_t xbt_fifo_new_item(void);
1242 void xbt_fifo_free_item(xbt_fifo_item_t);
1243 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1244 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1245 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1247 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1248 * Documentation: xbt_log module unmercifully reworked [MQ]
1249 * Bugfix: there was a problem with the ending of contexts with
1250 the pthread backend. It caused some weird deadlock or behavior
1251 depending on the pthread implementation. [AL]
1252 * Bugfix: get the exceptions raised in the simulator repport where
1253 and why they come from when they are not catched in time [AL, MQ]
1256 * Bugfix: Do repport the error when two non-connected hosts try to
1257 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1260 * Add additionnal checkings on communications. Assert that two
1261 communicating hosts are connected by a set of links... [AL]
1264 * Add additionnal checkings on channel values in communication [AL]
1265 * New: MSG_task_get_source to see on which host a task was generated [HC]
1266 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1267 the number of tasks waiting to be received on channel and sent
1269 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1270 waits for the first task coming from a given host.. [AL]
1272 GRAS new functionnalities: [MQ]
1273 * Enhance the parsing macro to allow the size of multidimentional objects
1274 to be given thru annotations.
1275 * New example (and documentation): Matrix Multiplication a la RPC
1276 (as when I was young!) and fix a bunch of bugs found on the way.
1278 GRAS performance improvements: [MQ]
1280 * Reduce the amount of cbps creation/destruction by making it static to
1281 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1284 * Change libdata to a set so that we can search for stuff by ID (and thus
1285 reduce the insane amount of dict lookups)
1288 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1289 RL communication on top of SG since it's so uneffective.
1290 It may also be used for inter-thread communication in RL, one day.
1291 * Use gras_datadesc_copy() to exchange messages on top of SG
1293 - improve message exchange performance on top of SG
1294 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1295 * Don't exchange on the network the size of the used part of buffer,
1296 instead, specify the possible buffer size to read().
1298 - reduces the amount of read/write calls (one pair per exchange)
1299 - reduces the amount of exchanged data (the size)
1300 - allows to retrieve all arrived data on receiver side, if we don't need
1301 it right now (subsequent read will peek the buffer)
1302 - allows the receiver to proceed with the begining of the stream before
1303 everything is arrived
1304 - make it possible to build an iov transport (using readv/writev)
1306 - take care of the data with non-stable storage (like stacked data),
1308 * If possible, TCP send uses vector I/O (when writev() is here)
1309 - Don't use it for receive since we send data sizes and data on the
1310 same stream, so we wouldn't be able to chain large amount of chunks
1311 before having to flush the stuff to read the size.
1312 * Rework the transport plugin mecanism to simplify it and reduce the
1313 amount of pointer dereferencement when searching for the right function
1316 * I guess that now, we do almost as few system calls as possible while
1317 doing as few data copy as possible.
1319 To improve it further, we could try to send all the sizes first and then
1320 all the data (to use iov on receiving size), but it's only a partial
1321 solution: when you have 2 dimensional data, the sizes of the second
1322 dimension is data of the first dimension, so you need 3 streams.
1324 I'm not sure the potential performance gains justify the coding burden.
1326 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1328 SimGrid (3.00) stable; urgency=high
1331 * New! Give the possibility to hijack the surf parser and thus bypass
1332 MSG_create_environment and MSG_launch_application. Have a look at
1333 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1335 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1337 SimGrid (2.96) unstable; urgency=low
1342 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1343 This deprecates the xbt_error_t mecanisms.
1344 It modifies (simplifies) all XBT and GRAS API.
1345 MSG API keeps unchanged (exceptions raised by XBT are catched from
1346 within MSG and masked with existing error handling facilities)
1349 * New! Add a FATPIPE model. [AL]
1350 * New! Add a parallel task model. [AL]
1351 * New! Add automatically a loopback interface (in the default
1352 network model) if none was precised.
1355 * Bugfix: MSG_process_resume now works with the current running process.
1357 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1358 * Modification of MSG_task_get_compute_duration. Once a task has been
1359 processed, the value returned by this function is now equal to 0. [AL]
1360 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1361 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1362 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1363 of a m_task. For now, MSG_task_cancel only works with computation
1365 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1366 of the processor (in Mflop/s) regardless of the current load on the
1368 * API Change: use proper naming convention for MSG_getClock and
1369 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1371 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1372 This function changes the priority of a computation task. This priority
1373 doesn't affect the transfer rate. A priority of 2 will make a task
1374 receive two times more cpu power than the other ones. This function
1375 has been added to suit the needs of Nguyen The Loc and hasn't been that
1376 much tested yet. So if it fails, please report it and send me your code.
1378 * API Change: removed all functions and types that were marked "deprecated"
1379 since many months. Renamed MSG_global_init_args to MSG_global_init.
1381 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1383 SimGrid (2.95) unstable; urgency=low
1386 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1387 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1388 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1391 * Bugfix: Allow absolute paths for platform description files [MQ]
1392 * Bugfix: do free the variables after use. Leads to drastic performance
1394 * Implement max_duration (ie, timeouts) on resources [AL]
1397 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1399 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1400 the needed support (timeouts) [AL]
1403 * Implement measurement sockets. You can now get the bandwidth between two
1404 hosts thanks to AMOK (see below). [MQ]
1405 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1406 dynar over the network (yeah) [MQ]
1407 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1408 * simple/static token-ring example. [Alexandre Colucci and MQ]
1409 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1410 in place before [MQ]
1413 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1414 type parsing. "short" and "long" modifiers were also ignored; other
1415 modifier (such as reference level) are still ignored. [MQ]
1416 * Embeed the buffer size within the buffer itself on SG. [MQ]
1417 That way, send() are atomic and cannot get intermixed anymore (at least
1418 the ones which are less than 100k; bigger messages still have the issue)
1419 * Array size pushed by the field, not by the field type (or each
1420 and every long int will push stuff to the cbps) [MQ]
1421 * use select() to sleep since it allows to portably sleep less than one
1424 GRAS (minor cleanups)
1425 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1428 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1429 * type category 'ignored' killed as it was never used and were difficult
1431 * whether a type can cycle or not is now a flag, leaving room for more
1432 flags (as "ignored", if we feel the need one day ;)
1433 * Rename raw sockets to measurement sockets since "raw" has another
1434 meaning in networking community.
1437 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1438 and offers features not belonging to GRAS but that most applications
1439 need. One day, it may be a set of plugins loadable at runtime.
1440 * New module: bandwidth
1441 bandwidth measurement between arbitrary nodes running this module. [MQ]
1443 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1445 SimGrid (2.94) unstable; urgency=low
1447 The first beta release of SimGrid 3 !
1451 * Update the main page and the FAQ. Adding references to gforge.
1454 * Add a gras_os_getpid function.
1457 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1458 * Extend the logs so that they also print PID, hostname, date, ... if
1460 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1461 and kill the old version which were in testsuite/
1462 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1466 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1467 for users. Instead, each programming environment (GRAS, MSG,...) use it
1468 when needed to provide such a feature to users.
1469 Don't shortcut the mecanism or you will also shortcut the virtualization
1470 you need on the simulator.
1474 * Cleanups in configury with regard to compile optimization/warning flags.
1475 Also add -fno-loop-optimize to any powerpc since it's the optimization
1476 killing gcc (< 3.4.0).
1477 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1478 needed by MSG examples complications
1479 * Borrow configury beautifications from PHP
1482 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1483 warning (thanks loris for stumbling into this one).
1484 * Bugfix: stop loading private headers (gras_config.h) from the public
1488 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1489 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1490 RL would imply a DNS resolver, which is *hard* to do in a portable way
1491 (and therefore delayed).
1492 * Implement a real timer mecanism and use it in timing macros. This allows
1493 to avoid rounding errors and get a 0.000005 sec precision in timing
1494 macros. While I was at it, various cleanups:
1495 - allow to declare more than one timed section per file (fix a stupid bug)
1496 - move some private declaration to the right place
1497 - merge conditional execution and timing macros into emulation module
1498 - document the module
1499 - make sure the module cleanups its mess on gras_exit
1500 * Documentation improvements:
1501 - (new) how to compile applications using GRAS
1502 - (new) emulation support (timing macros)
1504 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1506 SimGrid (2.93) unstable; urgency=low
1508 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1511 - Use Paje properly where used. Still to be sanitized properly.
1512 - Portability fix: Add an implementation of the contexts using pthread
1516 - Add xbt_procname(): returns the name of the current process.
1517 Use it to show the current process's name in all logging.
1519 - fix detection of older flex version and the reaction, since we do
1520 depend on modern ones (we use lex_destroy)
1521 - Better separation of SG and RL in the libs: remove all simulation code
1522 from libgras. As a result, this lib is now only 200k when stripped.
1523 Some of the xbt modules may also be duplicated (two sets and such) and
1524 should be cleaned/killed before SG3.
1525 - Insist on using xlC on AIX because of weird problems involving gcc there.
1526 - Cleanup the make remote stuff. This is now done by scripts
1527 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1528 mainly for our private use, but we're working on changing them to user
1531 - Bugfix: flush the socket on close only if there is some *output*.
1532 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1534 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1537 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1539 SimGrid (2.92) unstable; urgency=low
1541 Alpha 3 on the path to SimGrid 3
1545 - New! First try of benchmarking macros.
1546 - New! First try so that gras_stub_generator generate deployment and
1547 remote compilation helpers.
1549 - Bugfix: Initialization fix in msg_test.
1553 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1555 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1557 - API Change: gras_os_sleep to take the amount of seconds as a double.
1558 Accepting an int was error prone since it was the only location where
1559 seconds were coded as such. It leaded to damn rounding errors.
1560 - Bugfix: Hard to belive that timers ever worked before this.
1562 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1564 SimGrid (2.91) unstable; urgency=low
1566 Alpha 2 on the path to SimGrid 3
1570 - Bug fix in the lmm_solver.
1572 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1573 through the function MSG_paje_output.
1574 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1575 - It is possible to bound the rate of a communication in MSG with
1576 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1577 to write it in the changelog).
1578 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1583 - Various cleanups to the autotools stuff
1584 - Begin to move Gras examples to examples/gras/
1585 - Let make distcheck work again (yeah!)
1587 - documentation overhauled using doxygen.
1588 gtk-doc-tools is dead in SimGrid now.
1589 - Automatically extract all existing logging categories, and add the list
1590 to the documentation (long standing one, to say the less)
1592 - Cleanup the known architecture table. Reorder the entries to group what
1593 should be, and use a more consistent naming scheme.
1594 (some of the test dataset are still to be regenerated)
1595 - New! Allow library to register globals on each process just as userdata
1597 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1598 lookup time (for now).
1599 Use it in msg and trp.
1600 This cleans a lot the internals and helps enforcing privacy of the
1601 headers between the gras components.
1602 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1603 - Bugfix: gras_os_time was delirious in RL.
1604 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1606 - Reenable GRAS now that it works.
1608 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1610 SimGrid (2.90) unstable; urgency=low
1612 Alpha 1 on the path to SimGrid 3
1614 * It is a long time since the last release of SimGrid. I'm sorry about
1615 that but as I had told you, I was rewriting a lot of things. I apologize
1616 to those who had been reporting bugs to me and that I had not answered.
1617 If your bug is still in the new version, please tell me. Here is a
1618 summary of the main changes.
1620 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1621 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1622 portability, speed, and a lot more but you'll figure it out later.
1623 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1624 I wanted to release the new MSG as soon as possible and I have
1625 broken GRAS, which is the reason why, for now, only MSG is fully
1626 functional. A laconic description of these projects is available
1627 in the documentation.
1629 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1630 kernel optimized for our needs (called SURF but only the developers
1631 should use it). Hence, MSG is now roughly 30 times faster and I think
1632 that by rewriting a little bit MSG, I could event speed it up a little
1633 bit more. Beside the gain in speed, it is also much easier to encode a
1634 new platform model with SURF than it was with SG. More to come...
1636 * REVOLUTION 3: I have tried to change a little as possible the API of
1637 MSG but a few things really had to disappear. The main differences
1638 with the previous version are :
1639 1) no more m_links_t and the corresponding functions. Platforms are
1640 directly read from a XML description and cannot be hard-coded
1641 anymore. The same format is used for application deployment
1642 description. The new format is described in the documentation.
1643 Have a look in tools/platform_generation. There is a tiny script
1644 that converts from the old platform format to the new one. Concerning
1645 the application deployment format, parsing the old one is tricky.
1646 I think most of you should however be able to convert your files. If
1647 it is really an issue, I can write a C code that does the conversion.
1649 2) the toolbox tbx does not exist anymore. We now have a library
1650 with much more data-structures but without the hash-tables (we have
1651 dictionaries that are much faster).
1653 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1655 *****************************************************************************
1656 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1657 * don't feel like converting the oldies. (Mt) *
1658 *****************************************************************************
1661 Version 2.90: "the long awaited one"
1662 - Finished rewriting and debugging MSG. Rewrote the documentation.
1663 - disable GRAS for now since it needs to be ported to the newest SG
1666 - Finish the port to windows (using mingw32 for cross-compile)
1669 - Main loop and datastructures of SURF. A cpu resource object is
1670 functional. Surf can thus be used to create cpu's with variable
1671 performance on which you can execute some actions.
1673 2004-11-15 Martin Quinson
1674 - Port to ARM. Simply added the alignment and size descriptions. Should
1675 work, but the ARM machines are so slow that I didn't had the opportunity
1676 to 'make check' over there yet.
1678 2004-11-15 Arnaud Legrand
1679 - Trace manager now written. It uses a heap structure and is therefore
1680 expected to be efficient. It may however be speeded up (particularly
1681 when many events occur at the same date) by using red and black
1682 trees. One day maybe...
1683 - Max-min linear system solver written. It uses a sparse matrix
1684 structure taking advantage of its expected use. Most operations are
1685 O(1) and free/calloc are called as few as possible. The computation of
1686 the minimum could however be improved by using a red and black tree
1689 2004-11-03 Arnaud Legrand
1690 - Rename every gras_* function that was in xbt/ to its xbt_
1692 - Add a heap and a doubly-linked list to xbt
1693 - Added a dichotomy to the dictionaries. make check works as well before
1694 so I assume that the patch is correct. I do not know however if things
1695 run effectively faster than before now. :)
1697 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1698 SimGrid, and the version number is bumped to 2.x
1700 2004-10-29 Martin Quinson
1701 - Introduction of the remote errors.
1702 They are the result of a RMI/RPC on the remote machine.
1703 ErrCodes being scalar values, you can't get the host on which those
1704 errors did happen. Extending the error mechanism as in Gnome is possible.
1705 No idea yet whether it is a good idea.
1707 2004-10-28 Martin Quinson
1708 - Interface revolution: the Starred Structure Eradication.
1709 I used to do typedef struct {} toto_t; and then handle *toto_t.
1710 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1711 - ???_t is a valid type (builded with typedef)
1712 - s_toto_t is a structure (access to fields with .)
1713 - s_toto is a structure needing 'struct' keyword to be used
1714 - e_toto_t is an enum
1715 - toto_t is an 'object' (struct*)
1717 typedef struct s_toto {} s_toto_t, *toto_t;
1718 typedef enum {} e_toto_t;
1719 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1720 s_toto_t) is private.
1722 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1723 so that it can set the variable to NULL. It was so for dicts and sets,
1724 it changed for dynars.
1726 - Fix a bunch of memleaks in dict_remove
1727 - Fix a bug in sg/server_socket opening: it failed all the time.
1729 2004-10-07 Martin Quinson
1730 - Speed up dynar lookup operation a bit.
1732 gras_dynar_get is dead.
1734 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1735 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1736 (which gives you the address of the stored data).
1738 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1739 copy of the data using an affectation to do the job and not a memcpy.
1741 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1742 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1744 It does not really speedup the dynar test because they are
1745 setting elements all the time (and look them seldom). But the dict does
1746 far more lookup than setting.
1748 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1750 2004-10-05 Martin Quinson
1751 - Allow to (en/dis)able the cycle detection at run time.
1753 Whether we should check for cycle or not is now a property of each
1754 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1755 datadesc_cycle_unset allow to remove this property when previously set.
1757 Note that the cycle detection is off by default since it impacts the
1758 performance. Watch the data you feed GRAS with ;)
1760 This property is hereditary. Any element embedded in a structure having it
1761 set have it set for the time of this data exchange.
1763 You should set it both on sender and receiver side. If you don't set it on
1764 sender side, it will enter an endless loop. If you forget on receiver
1765 side, the cycles won't be recreated after communication.
1767 - Header reorganization.
1768 Kill gras_private.h, each submodule must load the headers it needs.
1770 2004-10-04 Martin Quinson
1771 - Interface revolution: do not try to survive to malloc failure.
1773 Now, gras_malloc and friends call gras_abort() on failure.
1774 As a conclusion, malloc_error is not a valid error anymore, and all
1775 functions for which it was the only gras_error_t return value are
1776 changed. They now return void, or there result directly.
1777 This simplify the API a lot.
1779 2004-09-29 Martin Quinson
1780 - Re-enable raw sockets.
1781 Created by gras_socket_{client,server}_ext;
1782 Used with gras_raw_{send,recv}
1785 It should allow to kill the last bits of gras first version soon.
1787 This is not completely satisfactory yet (duplicate code with
1788 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1791 - Simplify transport plugin (internal) interface by not passing any
1792 argument to _server and _client, but embedding them in the socket
1795 2004-09-28 Martin Quinson
1796 - Finish the port to AIX.
1797 autoconf was my problem (segfault within the malloc replacement
1798 function. No idea why)
1800 2004-09-16 Martin Quinson
1801 - Fix some size_t madness on 64bit architectures.
1803 2004-09-08 Martin Quinson
1804 - Reduce the number of system headers loaded, overload some more system
1805 calls (such as malloc to cast the result of the system one, and work
1807 - Fix and reintroduce the config support
1809 2004-09-07 Martin Quinson
1810 - Source code reorganization to allow Arnaud to surf all over there.
1811 - Allow to document the logging categories.
1812 - Remove all uppercase from logging categories and useless cleanup in names.
1814 2004-08-18 Martin Quinson
1815 Version 0.6.2 (protocol not changed; API changed)
1816 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1817 gras_error_t), and NULL when not found. Functions expecting a msgtype
1818 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1819 hopefully usefull message.
1820 - Portability to prehistoric sparcs again
1822 2004-08-17 Martin Quinson
1823 Version 0.6.1 (protocol not changed; ABI not changed)
1824 - prealloc some buffers to speed things up
1826 2004-08-11 Martin Quinson
1827 Version 0.6 (protocol not changed; ABI expended)
1828 - The parsing macro can deal with the references, provided that you add
1829 the relevant annotations (using GRAS_ANNOTE(size,field_name))
1831 2004-08-09 Martin Quinson
1832 Version 0.5 (protocol not changed; ABI changed)
1833 - Allow to off turn the cycle detection code in data exchange at
1834 compilation time. It should be at run time, but I'm short of time (and
1835 the config stuff is still broken). That way, we keep dict out of the
1836 critical path, which is good because the performance is poor:
1837 - search not dichotomial yet
1838 - dynar give no way to access their content and memcpy everytime
1839 - In composed data description (struct, ref and so on), stop foolness of
1840 keeping the subtype's ID, but store the type itself. This keeps sets out
1841 of the critical path, which is good since they rely on dynar and
1842 dictionnaries. The only loose of that is that we cannot detect the
1843 redeclaration of a structure/union with another content (but I'm not sure
1844 the code detected well this error before anyway). We still can detect
1845 the redefinition discrepancy for the other types.
1846 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
1847 it breaks the code because of type-punning used all over the place).
1848 This breaks on all non-gcc architectures (for now).
1850 All those changes (plus the buffer of last time) allow me to gain 2 order
1851 of magnitude on cruel tests consisting of 800000 array of integers on two
1852 level of a hierarchical structure (200 secondes -> 4 secondes)
1855 - the selector of reference must now return the type it points to, not
1856 the ID of this type.
1858 2004-08-06 Martin Quinson
1859 Version 0.4 (protocol changed; ABI not changed)
1860 - Allow to pass --gras-log argument to processes in simulation mode. Really.
1861 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
1862 - Implement a buffer transport, and use it by default (it relies on tcp in
1863 real life and on sg in simulation).
1864 That's a bit hackish since I had a new field to the structure to store
1865 its data without interfering with the subtype ones. Inheritance
1866 is tricky in C. And that's a kind of reverse inheritance with one class
1867 derivating two classes. Or maybe a game with java interfaces. Anyway,
1868 that's damn hard in C (at least).
1869 Moreover, I got tired while trying to ensure plugin separation and
1870 genericity in SG mode. MSG wants me to do weird things, so let's go for
1871 cruel hacks (temporarily of course ;).
1872 See comment in transport_private.h:71
1873 - do not include all the _interface headers in private but in the files
1874 which really need them (to cut the compilation time when they are
1877 2004-07-26 Martin Quinson
1878 Version 0.3 (protocol not changed; ABI changed)
1879 - Major overhault of the datadesc interface to simplify it:
1880 - shorted the function names:
1881 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
1882 - add a trivial way to push/pop integers into the cbps without malloc.
1883 This allows to make really generic sub_type description, which simply
1884 pop their size of the stack.
1885 - add a function gras_datadesc_ref_pop_arr() which does what users want
1886 most of the time: Declare a dynamic array (which pops its size of the
1887 stack) and declare a reference to it. Poor name, but anyway.
1888 - kill the post-send callback, add a post-receive one
1890 2004-07-23 Martin Quinson
1891 Version 0.2 (protocol changed; ABI changed)
1892 - add some testing for cpbs in the test cases, and fix some more bugs.
1893 This invalidate again the little64 data file, since I cannot regenerate
1895 - remove an awfull optimization in the logging stuff, allowing me to:
1896 - understand it again
1897 - learn gcc how to check that the argument match the provided format
1898 - fix all errors revealed by gcc after that
1899 - internal keys of dict are not \0 terminated. Deal with it properly in
1900 loggings instead of segfaulting when the user want to see the logs :-/
1902 2004-07-22 Martin Quinson
1903 - Fix some stupid bug preventing cbps (callback postit) from working
1905 2004-07-21 Martin Quinson
1906 - Some documentation cleanups
1907 - remove the useless last argument of msgtype_declare
1908 - rename the Virtu functions to fit into the 'os' namespace
1909 - move headers src/include -> src/include/gras/ and stop fooling with
1910 gras -> . symbolic link
1911 - make distcheck is now successful
1913 2004-07-19 Martin Quinson
1915 - Build shared library also
1916 - Install html doc to the right location
1917 - stop removing maintainer files in make clean
1918 - build tests only on make check
1920 2004-07-13 Martin Quinson
1922 - No major issue in previous version => change versionning schema
1923 - Re-enable little64 convertion test now that Abdou kindly regenerated the
1924 corresponding dataset.
1926 2004-07-11 Martin Quinson
1928 - Get it working with any kind of structure (we can compute the padding
1929 bytes remotely for all the architectures I have access to)
1930 - Implement the structure parsing macro (still not quite robust/complete)
1931 - Improvement to the remote testing toysuite
1933 2004-07-10 Martin Quinson
1934 [autoconf mechanism]
1935 - get ride of a bunch of deprecated macros
1936 - actually run the test for two-compliment, not only compile it :-/
1937 - test whether the structures get packed (and bail out if yes. Damn.
1938 Alignment is a serious matter)
1939 - test whether the structures get compacted (but respecting the alignment
1940 constraints of each types)
1941 - test whether the array fields of structures can straddle alignment boundaries
1943 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
1944 (took me 2 hours to find that bug, looking at the wrong place)
1945 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
1946 before sending/receiving and are used to compute the offsets of fields
1947 - Given that padding size depend even on compiler options, keep track of
1948 alignment and aligned_size only for the current architecture. Not a big
1949 deal since we send structure fields one after the other (seems
1951 - Add the datastructure used for IEEE paper by the PBIO guys to the test
1952 program, let it work on linux/gcc/little32. portability todo.
1954 2004-07-08 Martin Quinson
1955 - import and improve remote compilation support from FAST
1956 - make sure make check works on half a dozen of machines out there
1958 2004-07-07 Martin Quinson
1959 Let's say it's version 0.0.3 ;)
1960 - Implement conversions (yuhu!)
1961 - Let it work on solaris (beside conversion, of course)
1962 - Stupid me, using rand() to generate the conversion datatests in not wise.
1964 2004-07-06 Martin Quinson
1965 - Let make dist work, since I'm gonna need it to compile on remote hosts
1966 - Let Tests/datadesc_usage write the architecture on which the file was
1967 generated as first byte.
1968 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
1969 architecture descriptions.
1970 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
1971 on those architectures.
1972 - Optimization: send/recv array of scalar in one shoot
1974 2004-07-05 Martin Quinson
1975 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
1977 - Plug a whole bunch of memleaks
1978 - each process now have to call gras_{init,exit}. One day, their log
1979 settings will be separated
1980 - Continue the code factorisation between SG, RL and common in Transport.
1982 2004-07-04 Martin Quinson
1984 - Redistribution between SG and RL.
1985 We wanna have to accept in SG, so move accepted related parts of RL in
1986 the common part. (more precisely, the dynar of all known sockets is no
1987 more a static in transport.c, but part of the process_data)
1989 [gras_stub_generator]
1990 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
1992 2004-07-03 Martin Quinson
1993 - Create a new log channel tbx containing dict, set, log, dynar (to shut
1994 them all up in one shot)
1996 - Fix the ugly case of reference to dynamic array.
1997 - New (semi-public) function gras_datadesc_size to allow the messaging
1998 layer to malloc the needed space for the buffer.
2000 - gras_socket_close now expect the socket to close (and not its address to
2001 put NULL in it after it). This is because the socket passed to handlers
2002 is one of their argument (=> not writable).
2004 - propagate the interface cleanup from last week in datadesc, ie remove a
2005 superfluous level of indirection. User pass adress of variable
2006 containing data (both when sending and receiving), and not of a variable
2007 being a pointer to the data. Let's say that I like it better ;)
2008 The price for that is constructs like "int msg=*(int*)payload" in
2009 handlers, but it's a fine price, IMHO.
2011 - Let it work in RL (yuhu)
2013 2004-06-21 Martin Quinson
2015 - porting SG plugin and SG select to new standards (works almost).
2016 - plug memleaks and fix bugs around.
2019 - cleanup the prototype of data recv and force users to specify when they
2020 want to handle references to objects. Test case working even for cycles.
2021 - plug memleaks. Valgrind is perfectly ok with this.
2023 2004-06-12 Martin Quinson
2025 - cleanup the separation between plugin and main code in plugin creation
2027 2004-06-11 Martin Quinson
2029 - Reput hook for raw sockets, needed for BW experiments
2030 - kill a few lines of dead code
2031 [Data description] Interface cleanup
2032 - gras_datadesc_by_name returns the searched type or NULL.
2033 That way, no variable is needed to use a type desc once, which makes
2035 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
2036 two parameters were strings (field name, type name), leading to
2038 [Dicos] Interface cleanup
2039 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
2040 This is consistant with the dynar API.
2042 2004-04-21 Martin Quinson
2044 - Porting to new standards.
2046 - interface cleanup.
2047 There is no bag anymore, no need to take extra provision to mask the
2048 pointers behind "ID".
2049 Better splitup of functions between files create/exchange/convert.
2050 This is still a bit artificial since convert and receive are so
2051 interleaved, but anyway.
2053 - add a queued message list to procdata (the ones not matching criteria
2055 - factorize some more code between SG and RL wrt procdata
2057 - use gras_exit in example to track memleaks
2058 - get rid of gs_example now that GS is properly integrated into gras
2059 - update run_test to integrate the lastest tests (datadesc)
2061 - rename WARNINGn macros to WARNn since it prooved error-prone
2063 2004-04-19 Martin Quinson
2065 - register init/exit functions within gras module mechanism
2066 - send/receive function.
2067 Convertion is not implemented, but short-cutted if not needed.
2068 struct/array elements are sent one by one (instead of block-wise), but
2069 nobody really cares (yet). Get a prototype before optimizing.
2070 - tests (using a file socket) for DD send/receive on:
2071 - base types: int, float
2072 - array: fixed size, string (ie ref to dynamic string)
2073 - structure: homogeneous, heterogeneous
2074 - chained list, graph with cycle
2075 Believe it or not, valgrind is not too unhappy with the results. The
2076 cycle happily segfaults, but the others are ok. And I'm sick of pointers
2080 - Bugfix when using a filename explicitely (instead of '-')
2082 2004-04-09 Martin Quinson
2084 - factorize more code between RL and SG in socket creation
2085 - Complete the implementation and tests of:
2087 o file (only in RL, and mainly for debugging)
2089 I lost 3 days to design a portable address resolver, and then decided
2090 that the prototype mainly have to run on my box.
2091 Addressing portability too early may be like optimizing too early :-/
2093 - use gras_init in the Tests instead of the crappy parse_log_opt
2094 (the latter function is removed)
2095 [Conditional execution]
2096 - New functions: gras_if_RL/gras_if_SG (basic support for this)
2097 [Code reorganisation]
2098 - Get rid of libgrasutils.a since it makes more trouble than it solves.
2099 Build examples against the RL library, since there is no way to disable
2100 its creation for now.
2102 For information, the beginning of coding on GRAS was back in june
2103 2003. I guess that every line has been rewritten at least twice since
