1 SimGrid (3.7) NOT RELEASED; urgency=low
3 The "sprint cleanups (before spring)" release.
6 * We can specify the smpi latency/bandwidth factor with command line
7 add --cfg=smpi/bw_factor:"threshold0:value0;...;thresholdN:valueN"
8 or add --cfg=smpi/lat_factor:"threshold0:value0;...;thresholdN:valueN"
9 You can also use the "config tag" from platform file by adding this line
10 <prop id="smpi/bw_factor" value="threshold0:value0;...;thresholdN:valueN"></prop>
11 (see "example/platforms/tag_config.xml" to use "config tag").
12 Note that the command line supersedes the platform file configuration.
13 * Change the correction factors used in LMM model, according to
14 the latest experiments described in INRIA RR-7821.
15 Accuracy should be improved this way.
16 (TODO: this is not true in the code, yet)
17 * Use the partial invalidation optimization by default for the
18 network too. Should produce the exact same results, only faster.
19 * Major cleanup in surf to merge models and split some optimization
20 mechanisms from the core of the models. As a result you can now
21 specify which model to use (e.g., --cfg=network/model:LV08
22 --cfg=cpu/model:Cas01) and which optimization mode to use
23 (e.g., --cfg=network/optim:lazy --cfg=cpu/optim:TI).
24 Incompatible combinations should err at initialization. See
25 --help-models for the list of all models and optimization modes.
26 * The CLM03 workstation model was dropped for simplicity because it
27 used the deprecated CM02 network model. Use default instead.
28 * Rename the TCP_gamma configuration option to network/TCP_gamma
29 * Rename the coordinates configuration option to
30 network/coordinates, and document it
31 * Use now crosstraffic keyword instead of the terribly misleading
32 fullduplex keyword. It is activated by default now in the current
33 default model, use --cfg=network/crosstraffic:0 to turn it off.
34 * Experimental: new configuration option --cfg=surf/nthreads:N to run
35 analytical models in parallel with N threads (default is 1).
36 * Ongoing refactoring the model parsing to make XML files optional
37 (more work is needed here to fully achieve this goal)
40 * Deprecate the use of m_channel_t mecanism like MSG_task_{get,put}
41 functions and friends. This interface was considered as
42 deprecated since over 2 years, it's time to inform our users that it is.
43 Switch to MSG_task_{send,recv} instead, or compile Simgrid command line
44 'cmake -Dcustom_flags="-DMSG_USE_DEPRECATED" .' if you really need to
45 use these (crappy) functions in your code.
46 These functions will be removed at some point. Stop using them now.
47 * Deprecate MSG_get_host_{table,number}
48 Implement MSG_hosts_as_dynar() instead.
49 * Implement MSG_processes_as_dynar() (Closes gforge #13642)
52 * Stabilize the parallel execution mode of user contexts
53 * Introduce configuration variables to control parallel execution:
54 - contexts/synchro: Synchronization mode to use when running
55 contexts in parallel (either futex, posix or busy_wait)
56 - contexts/parallel_threshold: Minimal number of user contexts
57 to be run in parallel (raw contexts only)
60 * Introduce a parallel mode for the models (see surf/nthreads
61 configuration item). In our tests, running the models in parallel
62 never lead to any speedups because they are so fast that the gain
63 of computing each model in parallel does not amortizes the
64 synchronization costs, even when ultra fast futexes are used.
65 This is released anyway because YMMV.
68 * Performance boost by using a swag internally to compute the set of
69 tasks that are finished and should constitute the return value of
73 * Enable it by default now that it is considered rather stable.
76 * Documentation of the tracing functions.
77 * Performance gains when tracing categorized/uncategorized resource
78 utilization by avoiding calls to get route when updating resource
79 variables. LMM constraints are being used instead.
80 * API changed to set task categories. Use MSG_task_set_category instead
81 of TRACE_msg_set_task_category, and SD_task_set_category instead
82 of TRACE_sd_set_task_category. They only work if ENABLE_TRACING is ON.
83 * Bugfix for graphicator, routes not correctly obtained, memory leaks
84 * Examples for link user variables added (see at examples/msg/tracing/)
85 * Deprecated function TRACE_msg_set_process_category completely removed
86 * Trace header updated according to the latest Paje file format
87 * Tracing network lazy updates, no longer obligate users to use full updates
90 * Improve the API of Lua MSG bindings, using the Lua spirit.
91 * Each simulated process now lives in its own Lua world (globals are
92 automatically duplicated). It helps writing simulators. This is also the
93 first step towards running real distributed Lua programs with SimGrid.
94 * Add a Chord example in Lua, equivalent to the MSG one.
97 * Mallocators: allow value NULL for the reset function.
99 - New function xbt_dict_new_homogeneous(void(*)(void*)) to
100 create homogeneous dictionaries, where all the elements share the
101 same free function. Non homogeneous dictionaries will be
102 deprecated in the next release.
103 - Dicts of scalar elements (xbt_dicti_*) are deprecated.
104 - Multi-level dictionaries are deprecated.
106 - Cleanups and simplifications to make it maintainable again.
107 - Exotic features (such as memalign and valloc) were removed.
108 - The metadata were extended and improved so that the
109 model-checker becomes able to explore and inspect the heaps.
110 - This may induce a performance drop when enable_model-checking is
111 ON in cmake (even if it's not used in the simulation), but it is
112 necessary at this point to get MC working.
114 Turn model-checking OFF if simulation performance matters to you.
116 -- $date Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
118 SimGrid (3.6.2) stable; urgency=low
120 The "Not coding new stuff allows to polish old things" release.
123 * New bindings to the NS3 packet level simulator (experimental)
124 * Use the raw (efficient) execution contexts instead of the sysv
125 (portable) ones when possible.
126 * libpcre is now mandatory in any cases since not using it led to
127 severe performance loss and possibly other issues
128 * Update the XML platforms:
129 - G5K: include the latest machine in Nancy
130 - GridPP and LCG: new platforms
131 * Documentation was partially updated, at least (more to come)
133 Bug fixes, cosmetics and small improvements
134 * Free terminated processes before the end of the simulation to avoid
135 exhausting the memory of users having very dynamic amount of
137 * Bug fix and cosmetics about canceling non-running tasks
138 * Bug fix about the dot loader's issues when using libcgraph
141 * Create an installer for windows with nsis (amd64 and win32)
142 - Add an hello world project to illustrate simgrid project creation.
143 - Embed libpcre into the Simgrid installer to avoid
144 its compilation burden
145 * The raw execution contexts should work on Apple now
146 * Port to Windows 64 bits
147 - Sysv contexts now have an implementation for this arch
148 - GRAS communication features now support this arch
149 * Drop support for borland compiler on windows
150 - this code was not maintained, and we kinda depend on gcc nowadays
151 * Fix portability issues on kfreebsd/gnu: build error about semaphores
152 * Fix portability issue on unstable ubuntu: linker became picky on
155 -- Wed Oct 5 15:51:01 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
158 SimGrid (3.6.1) stable; urgency=low
160 The "Oops, we broke Macs too" release
163 * Fixed contexts detection so that raw ones are used when possible
164 * On Mac, do not use Posix Ucontexts with gcc v4.[1-5] since this
165 leads to a strange error, with user code segfaulting sometimes when
166 the generated code is not perfectly aligned (which is not
167 controllable from the user side, depends on the amount of code)
170 * New macro: CATCH_ANONYMOUS, which is like CATCH(e) but without argument.
172 -- Mon Jun 27 13:59:03 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
174 SimGrid (3.6) stable; urgency=medium
176 The Summer Release, also known as the "OMG! They Killed Kenny!" version
179 * Bindings now constitute their own package, separated from the main one.
180 Rationale: reduce our maintenance nightmare by reducing the module coupling
181 They will soon be released on their own on gforge.
183 svn co svn://scm.gforge.inria.fr/svn/simgrid/contrib/trunk/simgrid-java
184 svn co svn://scm.gforge.inria.fr/svn/simgrid/contrib/trunk/simgrid-ruby
186 GRAS: It is not considered as stable anymore, but experimental. Sorry.
187 * It's not quite deprecated for now because we have no replacement,
188 but it may soon become the case.
191 * New MPI functions supported: MPI_Comm_disconnect, MPI_Comm_get_name
192 * Fortran: New user-level cache variable to store the rank of the running
193 process. This improves performance by an order of magnitude.
194 * C: New coccinelle script to automatically locate and modify global and
195 local static variables.
196 * Improved SMPI network model with a sender-side gap to account for multiple
200 * New function MSG_comm_get_status(). MSG_comm_test() and MSG_comm_testany()
201 only say if a communication is finished, no matter whether it succeeded or
202 failed. You can call MSG_comm_get_status() to know the status of a finished
204 * New function MSG_task_dsend() to send a task and detach it. When a
205 communication is detached, you are never notified of its success or failure
206 and the memory is released automatically once it is finished. This function
207 is useful when you don't care about the end nor the success of a
209 * Change the prototypes of action replay. Sorry for inconvenience,
210 but this is really more efficient this way (and to adapt your code,
211 you just have to fix the initialization, that shouldn't be too long)
212 * Kill the braindead MSG_task_refcount_dec() function. I guess nobody
213 ever managed to do anything useful with it.
214 * New function MSG_comm_testany(). Similarly to MSG_comm_waitany(), it
215 takes a dynar of communications. It returns immediately and gives the
216 index of a finished communication (if any).
217 * New example: a basic implementation of the Chord P2P algorithm.
220 * New model for multi-core CPUs. You can now use the core attribute to
221 precise the number of cores of a host. This is a basic model. Every
222 process running on the host receives at most the power provided in
223 the DTD (throughput<=power). Total throughput of process cannot exceed
225 * New peer tag. This peer tag creates a tiny AS comprising a host and a
226 router linked by an up-link and a down-link (possibly asymmetrical).
227 This kind of pattern allows to easily build last-mile model style platforms.
228 Aggregating such patterns in a rule-based AS is thus the technique of
229 choice for modeling large peer-to-peer/volunteer computing/cloud platforms.
230 * New model for Vivaldi routing. We transformed the Vivaldi network model
231 into a Vivaldi routing model (based on the rule-based model). This allows to
232 combine Vivaldi based latencies with last-mile platforms.
235 * Added a check for NaN of IEEE754 infinite in the double entries of
237 * Introduce a new context factory "raw", highly inspirated from the
238 ucontext factory, but using manually crafted functions in assembly to
239 do the work in an efficient manner.
240 * Allow to change the used context factory at run time, not only at
241 compilation time. Use --cfg=contexts/factory:raw for maximal speed.
242 * Add an option --cfg=contexts/stacksize:N to set the stack size of the user
243 contexts at runtime (only with raw contexts or ucontexts).
244 * Completely rewrote this module to allow parallel execution of user
245 processes. Use --cfg=contexts/nthreads:N to execute user processes
246 with N parallel threads (the default is 1, meaning no parallelism).
247 * Allow to decide dynamically between sequential and parallel modes.
248 When nthreads > 1, you can use --cfg=contexts/threshold:P to run the user
249 processes in parallel only when their number is greater than or equal to P
251 * Added a check for NaN of IEEE754 infinite in the double entries of
255 * New command line option: if you pass --cfg=verbose-exit:0, SimGrid
256 won't output the state of processes when interrupted with Ctrl-C
257 * Add a new function xbt_dynar_to_array that transforms a dynar into a
258 NULL-terminated array. This may solve backward compatibility issues
259 due to the change to return type of SD_simulate. See also:
260 http://lists.gforge.inria.fr/pipermail/simgrid-user/2010-December/002206.html
261 * Add new macros with variable number of arguments.
262 - in xbt/log.h: XBT_DEBUG, XBT_VERB, XBT_INFO, etc.
263 - in xbt/asserts.h: xbt_assert
264 - in xbt/cunit.h: xbt_test_{add,fail,assert,log}
265 - in xbt/ex.h: THROWF and RETHROWF.
266 Define XBT_USE_DEPRECATED if you want to use the old numbered macros like
268 * Change xbt_die() to accept a format string with arguments, just like printf.
269 * New data structure: xbt_lib_t, like a dict but more general and with better
273 * New configuration options
274 Options triva/categorized and triva/uncategorized can be used to generate
275 graph configuration files for Triva visualization tool.
276 * Configuration option tracing/platform is renamed to tracing/categorized
277 * XBT logging makes tracing error checks easier, new root log hierarchy: instr
278 * New TRACE_user_link_variable interface:
279 User provides the name of the link and the tracing variable to attach to it
280 * the declaration of tracing categories must be done after the environment
282 * simpler tracing interface, just one way to declare categories
283 TRACE_category or TRACE_category_with_color, it is up to you
284 * links in the trace file are again identified by their names
285 * trace contains the full platform hierarchy exactly as declared using the ASes
286 * Options tracing/msg/[task|process]:1 groups the process by hosts
287 for both cases, tasks and processes must have names that are unique during the simulation
288 these options generate traces that are suited to gantt-charts, such as the space-time view of Paje
289 * The experimental option tracing/msg/volume is deprecated
290 its functionality may be reincorporated if needed
292 The tracing generates a trace file with unordered timestamped events,
293 because of the way the core simulator (surf) works. A script available
294 at the tools directory (fix-paje-trace.sh) can be used to put the events
295 in order. We have changed the tracing so it can generate ordered timestamped
296 events in the final trace, but depending on the simulator (and how much time
297 is simulated) that can lead to a huge memory utilization. It is deactivated
298 by default, but it can be activated using the --cfg=tracing/buffer:1 switch.
301 * Define a SIMGRID_VERSION macro in simgrid_config.h.
302 - We are trying hard to keep the API stable, but it may happen that
303 some things change (we're a research project after all, not a
304 nuclear plant operating system). If such things should happen, you
305 could rely on that macro to adapt.
306 - current value: 30600 for 3.06.00, aka 3.6
307 * Define macro MAKE_SIMGRID_VERSION(major, minor, patch) to help building
308 a number that can be compared with SIMGRID_VERSION.
309 * Add a build option -Denable_debug (set to ON by default): when set to OFF,
310 assertions and verbose/debug logging events are disabled at compile time.
312 -- Tue Jun 21 08:57:43 CEST 2011 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
314 SimGrid (3.5) stable; urgency=medium
317 * New feature to any SimGrid-based simulator: Model-Checking
318 Check SIN#1 for more details.
321 * New Model SMPI (three-interval linear regression for correction factors)
322 See RR-7426, available at http://hal.inria.fr/inria-00527150
323 * Ability to use FORTRAN MPI code (through f2c, automatically privatized)
324 * New MPI functions supported: MPI_Get_count(), MPI_Comm_split()
325 * New: RAM folding (see RR-7426 and examples/smpi/NAS/DT-folding)
326 * New: execution sampling (see RR-7426 and examples/smpi/NAS/EP-sampling)
327 * See also src/smpi/README
332 - Tracing API changes: TRACE_start and TRACE_end should not be called
333 by user-code. They are automatically called by simulators created
334 with SimDAG, MSG and SMPI if the toolkit is compiled with
335 tracing_enabled=ON. Categories declaration and utilization remain the
336 same for MSG and SimDag.
337 - A function was added to the tracing API to declare categories with
339 - TRACE_category_with_color (char *category, char *color)
340 where color must be in the following format
341 "%f %f %f", red, green, blue
342 and red, green, blue are float values in the interval [0, 1]
343 - User can specify NULL as color parameter, or continue calling
345 On that case, the tracing system will define random colors
346 - The following command-line options are supported:
347 --cfg=tracing/filename:msg.trace
348 --cfg=tracing:1 (activate tracing, needed to use others)
349 --cfg=tracing/platform:1 (categorized resource use)
350 --cfg=tracing/uncategorized:1 (uncategorized resource use)
351 --cfg=tracing/msg/task:1 (task creation)
352 --cfg=tracing/msg/process:1 (process creation, migration)
353 --cfg=tracing/msg/volume:1 (volume of MSG send/recv)
354 --cfg=tracing/smpi:1 (SMPI interface tracing)
355 --cfg=tracing/simdag:1 (allow SimDAG tasks receive categories)
356 - examples of examples/msg/tracing updated
358 - DAXLoader and DOTLoader functions can generate tasks with categories
359 - A new function to attribute a category to SD tasks:
360 TRACE_sd_set_task_category (SD_task_t task, char *category)
361 * Tracing the MPI interface implemented by SMPI
362 - Collective operations are traced with states
363 - Point-to-Point operations are traced with states/links
364 - Tracing activated by a parameter "-trace filename" passed
365 to smpirun during execution (considering that simgrid
366 is compiled with tracing enabled)
367 - To run the simulation with gdb, the simulator
368 accepts --cfg=tracing/smpi:1 to trace SMPI
369 - tesh files to check if smpi tracing is ok
370 - See examples/smpi/NAS/DT-trace
371 * GTNetS tracing re-worked
372 - adaptation to the tracing system of GTNets to cope
373 with modifications regarding the fullduplex mode
374 - new tesh files to check if gtnets tracing is ok
377 * Asynchronous communications through the functions:
378 MSG_task_isend/irecv and MSG_comm_test/wait/waitall
379 * New function: MSG_load_platform_script()
380 to make possible using a lua script instead of XML files to set up platforms
381 * New function: MSG_set_function
382 to associate functions to processes, used when bypassing the parser
383 * New functions: MSG_task_set_name(), MSG_task_set_compute_duration()
385 Platforms: Add some more examples in examples/platforms
386 * Grid'5000: see www.grid5000.fr
387 * *_30000_hosts.xml: various huge files [mainly scalability testing]
390 * Change the XML format. This is a very important modification. SimGrid
391 3.5 introduces a new hierarchical format based on the notion of
392 Autonomous Systems. Compatibility with old format is ensured
393 through the perl script provided in the install bin directory
394 bin/simgrid_update_xml.
395 It is now possible to build platforms with specific routing
396 mechanism (Full/Dijkstra/DijkstraCache/Floyd) and to easily
397 connect several platforms together. We will try to provide soon
398 set of realistic platforms exploiting these properties (have a
399 look at examples/platforms/ for the moment).
400 * Take the opportunity of the XML format change to be a good XML citizen:
401 rename link:ctn to link_ctn and similar changes (also dealt with by
403 * Add a new routing scheme (rule-based) using regular expressions. It
404 enables to have an extremely low memory footprint when the
405 underlying routing is simple and can be compactly described. You
406 need to have libpcre4-dev (perl regular expressions) installed if
407 you want to use this routing scheme.
408 * Revive the cluster TAG and allow to easily and efficiently (both in
409 term of memory and speed) connect clusters together. Have a look
410 at teshsuite/simdag/platforms/ to see how this can be done. With
411 this tag, you can create clusters with thousands of tasks at no
412 cost (have a look at examples/platforms/).
413 Note: clusters are implemented as ASes, so there is no need for an
414 enclosing AS tag if you have only one cluster in your platform.
415 * Add new generic functions in the public interface that allows the user
416 to call SURF 'create_resource' methods from your code (same
417 functionality as the XML bypass mechanism but with a much lighter
419 * Add a new model (enabled through command line --cfg=network/model:SMPI)
420 that uses a piecewise linear approximation to produce better
421 results when exchanging small messages.
422 * Add a new parameter to handle correctly full duplex link and account
423 for interferences between uplink and downlink communications
424 (activate with --cfg=fullduplex:1).
427 * Rename the SD_READY (all dependencies are satisfied and task is
428 scheduled) state in SD_RUNNABLE and define a new SD_SCHEDULABLE (all
429 dependencies are satisfied) state.
430 This prevents a confusion between the notion of "ready to schedule"
431 (SD_SCHEDULABLE) used in DAG scheduling and that of "ready to be
432 simulated" (SD_RUNNABLE) used by the simulation kernel.
433 * Change the way a task is considered as ready. Instead of removing
434 dependencies when a task is done, a counter is decreased. This way,
435 it is always possible to reach ancestors thanks to the
436 SD_taks_get_parents function (even after the end of the simulation.)
437 * Change the return type of SD_Simulate from (SD_task_t*) into
438 xbt_dynar_t. This function was in handling a dynar internally and
439 converted it into a NULL terminated array for historical reasons.
440 * New function SD_dotload(char*) to load a DAG described in dot
441 format. This loader and the corresponding examples require the
442 installation of the graphviz library.
443 * Fix a bug in the management of tasks of size 0 in the surf network
444 models. This problem was only visible with SIMDAG and you should
445 thus disregard results produced with earlier versions if you
446 relied on this feature (some tasks were blocked because of this).
447 * Fix a bunch of stuff that prevented to use classical models with SIMDAG
448 even though your applications were doing only point-to-point
449 communications and sequential computations. Now you can really use any
450 model you want (of course, if you create real parallel tasks, which are
451 not implemented in most models beside ptaskL07, this will abort).
452 * Add an example that schedules a DAX on an heterogeneous platform
453 using a Min-Min strategy.
454 * New function SD_workstation_get_current_task() that returns the kind
455 of task currently running on a workstation in the sequential access
457 * Raise some warnings when unexecuted tasks remains at the end of the
458 simulation. This is usually caused by cycles in the DAG.
461 * New function: SIMIX_process_set_function() called by MSG_set_function
462 * Change the underlying waiting queue in semaphores so that a process
463 can wait on several of them simultaneously (as in waitany).
464 * Fix the way to handle tokens in semaphores so that all access patterns
465 work: {acquire, acquire_timeout, waitany} / {release, release_forever}.
466 * kill the dirty pimple SIMIX_message_sizes_output()
467 Please use (proper) visualization instead
470 * New data container: setset (set of sets of elements)
471 * New module: mmalloc (mapped malloc, allowing to have several
472 independent segments of malloc)
473 * New function: xbt_dict_cursor_set_data()
474 * New functions: xbt_dynar_sort(), xbt_dynar_compare()
475 * New function: xbt_dynar_is_empty()
476 * New function: xbt_fifo_get_last_item()
477 * Fix xbt_dynar_shrink(): use the right element size.
478 * Fix xbt_dynar_set*(): allow index larger than current size and memset 0
479 uninitialized areas during expand.
480 * Fix semaphores: previous implementation was severely broken.
481 * Use library init/fini functions for our initialization.
482 - you can use logs and other feature as soon as you want in your
483 code (even before the xbt_init / MSG_init)
484 - xbt_exit is now a no-op and produce a warning when used.
487 * Port GRAS to new SIMIX mechanisms. This allows gras users to
488 benefit from the latest improvement to the simulation kernel.
489 * Kill measurement sockets for now. If you rely on them, sorry. This
490 release is not for you. This feature will be reintroduced in the
491 future, but we cannot delay the release any further.
492 * New function: gras_msgtype_get_name().
493 * Implement gras_agent_spawn in RL too (the prototype changed a bit)
494 * Fix (at last) the pmm example: it should not randomly fail anymore.
496 Build chain: bug fixes and overall polishing
497 * Cmake is now stable enough. Hence, we killed the autotools.
498 * Port to windows ( TM :)
499 * Fix the 'make install' target.
500 No need to use 'make install-simgrid' anymore
501 * Introduce a 'make dist' target compiling a *source* archive
502 'make package' compiles a binary archive
503 * Compile java files only on need
504 * Add --cd and --setenv command line options to tesh
505 * Out of source builds are not fully supported yet, but we are close
506 * Enable supernovae and optimization flags by default for our users
509 * Add layer to set up environment directly from lua, without XML.
510 * The effect of gras_stub_generator can be achieved through
511 lua too (check examples/gras/console/ping_generator.lua)
513 -- Wed, 01 Dec 2010 22:09:23 +0100 Da SimGrid team <simgrid-devel@lists.gforge.inria.fr>
515 SimGrid (3.4.1) stable; urgency=low
517 The "Polishing easter eggs is probably a good idea" release.
518 This is a bug fixes release only.
522 * Fix a bug preventing the tasks from begin garbage collected.
525 * Fix a bug occuring when a host involved in a communication fails.
526 This was not detected properly by the other peer involved in the
527 communication. Now, it's reported as a network error.
530 * Warn the user about loop dependencies in data flow of DAX files
531 * Obey the control-flow dependencies of DAX files
534 * Add option "enable_smpi" allowing to not compile SMPI.
535 Probably useful for the (Mac) users experiencing a build error here
536 * Improve the detection of lua5.1 and ruby1.8
538 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Tus, 04 May 2010 28 16:11:16 +0100
540 SimGrid (3.4) stable; urgency=low
542 The "Easter in Cargese" release. Also known as (major changes):
544 * the "se habla Java, Ruby 話せます, fala-se Lua (and deaf-friendly)"
545 ~> bindings were greatly improved
546 ~> new tracing infrastructure for better visualization introduced
548 * the "Welcome to configury modernity" release.
549 ~> we switched from autotools to cmake, and improved our cdash
552 A more detailled list of changes follow (full detail in svn log).
554 Java Bindings: Various Cleanups
555 * (install java-gcj-compat-dev on debian-like to use them)
556 * Remove put/get: no need to export deprecated interface in Java
557 Use send/receive instead.
558 * Cleanup the examples and add a README per directory
559 * Remove example autoDestination (that's the only way to go now)
560 * Remove example explicitDestination (was a plain copy of basic)
561 * Make JniException a runtime exception, so that there is no need to
562 declare the fact that you may encounter such a beast. I guess that
563 nobody will ever want to survive such error.
564 * Create specific errors for each MSG case of failure:
565 host failure, transfer failure, timeout, task cancelled
566 * Cleanup the exceptions that may get thrown by each function
567 * Other internal cleanups in Java bindings. Performance still bad :/
568 Ruby and Lua Bindings: create them
569 * (install ruby1.8-dev/liblua5.1-0-dev on debian-like to use them)
570 * That's new and great, you should try them out.
571 Same functionalities than Java bindings, only even less polished
573 * Kill the useless "rate" argument of SD_task_get_execution_time()
574 Everyone used to provide -1 as a value, it was not used, and the
575 semantic of a possible use wasn't even clear.
576 * SD_SCHED_NO_COST: Constant to use as cost in SD_task_schedule()
577 either as comm costs or compute costs to mean that there is no
578 such thing for that specific task.
579 * Add a SD_task_set_name() function
580 * Fix SD_task_unschedule() on typed tasks
581 * Fix SD_task_get_execution_time() to return seconds, not flop*sec
582 * In DAX loader, accept useless 'level' attributes to <job> since
583 LIGO DAGs have them (seem to be to ease graphical representation).
585 * Add an example masterslave_mailbox.c using send/receive and not
586 the deprecated put/get interface.
587 * Kill the MSG_paje_output() function. It's a noop since 2 years.
588 * Kill MSG_WARNING and MSG_FATAL return codes: they were not used
590 * Rename MSG_TIMEOUT_FAILURE into MSG_TIMEOUT for sake of logic
591 (declare MSG_USE_DEPRECATED to still have the old name)
592 * Add a MSG_task_set_data() function
593 * About trace replay (see examples/msg/actions):
595 - Allow to work with splitted trace files for each process
596 Give the specific trace file as argument of each process,
597 and call MSG_action_trace_run(NULL)
598 You can still have one merged file for all processes.
599 - Fix implementation of collective operations
600 * Allow task_execute() on 0-sized tasks (closes #10063)
602 * This is the first release of SimGrid where SMPI is not considered
603 beta anymore (even if some corners should still be improved)
604 * Port over the new SIMIX_network submodule (internal refactoring)
605 * Basic support to log events as with SMPE (use --cfg=SMPE:1)
606 * Implement more missing elements of the standard:
608 - MPI_MAXLOC MPI_MINLOC + all associated datatype MPI_DOUBLE_INT,
610 - MPI_Address() MPI_Get_count() MPI_Type_free() MPI_Type_extent()
611 MPI_Scan() MPI_Get_processor_name()
612 - Added implementation of missing case for Alltoall (warning: it's
613 *not* the bruck variant from OpenMPI; based on Alltoallv instead)
614 - SMPI_MPI_Gather() SMPI_MPI_Gatherv() SMPI_MPI_Scatterv()
615 SMPI_MPI_Reduce_scatter() SMPI_MPI_Allgather()
616 SMPI_MPI_Allgatherv()
618 - MPI_Waitsome() was broken
619 - Allow relative includes in smpicc
620 - Command line cfg argument 'reference_speed' was ignored...
621 - Some functions did not properly lead to auto-benching of user code
622 - smpicc passes -O2 by default (just like openmpi one)
624 * add SIMIX_action_suspend() and SIMIX_action_resume() functions
625 * Bug fixes about timeouts during communications
626 * add SIMIX_message_sizes_output() as a pimple to write to file the
627 amount of messages per size. Use gnuplot to get histogram.
628 Pimple because that's the only user-visible function of simix,
629 defined directly in xbt.h (irk, sorry)
631 - Add a SIMIX_sem_get_capacity() function
632 - Fix interactions with processe resume/suspende
633 - release_forever() was stupidly broken
634 - Fix SIMIX_display_process_status() for processes in a semaphore
635 - Make SIMIX_sem_block_onto() user-visible
636 * Refactoring context stuff:
637 - Use pseudo-OOP for better modularity
638 - reimplement SIMIX_process_kill() without process_schedule() so
639 that the latter can take as invariant that it is called from
641 - Merge context_start into context_new for sake of simplicity
643 * Add a Vivaldi network model, coded live during SUD'10 ;)
644 * Rename configuration variables to start a hierarchy:
645 o cpu_model -> cpu/model
646 o network_model -> network/model
647 o workstation_model -> workstation/model
648 * New configuration variables:
649 o network/bandwidth_factor: correction to bandwith
650 o network/latency_factor: correction to latency
651 o netwotk/weight_S: correction to the weight of competing streams
652 * Add a long description to the models, that users can see with such
653 argument on the command line: --cfg=cpu/model:help
654 * --help-models display the long description of all known models
656 * config: add the ability to set a default value after registration
657 Does not override any previously set value (e.g. from cmd line)
658 * dict: allow to have integer key and data.
659 When so, you need to use the following functions
660 void xbt_dicti_set(xbt_dict_t dict, uintptr_t key, uintptr_t data);
661 uintptr_t xbt_dicti_get(xbt_dict_t dict, uintptr_t key);
662 void xbt_dicti_remove(xbt_dict_t dict, uintptr_t key);
663 In contrary to regular dicts, the key is not malloced before copy.
664 Mixing scalar and regular elements in the same dict is not tested
666 * Allow to use xbt_dynar_shrink() to expend the dynar instead
667 Tracing for Visualization:
668 * SimGrid is now instrumented in order to generate a trace file for
669 visualization analysis: to use it, need to compile SimGrid with the
670 "tracing" option enabled, and instrument the program using SimGrid with
671 TRACE_start, TRACE_category, TRACE_msg_set_task_category and TRACE_end
672 (among other functions).
673 * The instrumentation only traces the platform utilization for now
674 * Documentation to use the tracing functions and how to analyze the
675 traces with the Triva tool is written.
676 * More information about: SimGrid FAQ (in the section Tracing Simulations
679 * We moved to cmake as default build system. Autotools support will
680 be dropped soon. Check the FAQ for more info about how to use it.
681 * Greatly improved our cdash/ctest interactions
682 Check http://cdash.inria.fr/CDash/index.php?project=Simgrid
683 * Added memory checking tests with valgrind; lot of memleak fixing.
684 This may be the first release of simgrid with so few memory issues
685 * Added code coverage tests.
686 Our coverage is still improvable, but at least we see it on cdash.
688 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 28 Apr 2010 28 17:11:16 +0100
690 SimGrid (3.3.4) stable; urgency=low
692 The "Desktop Grid needs love too" release (also called Xmas release).
695 * Major speedup in the maxmin system solving by using lazy evaluation
696 Instead of solving completely the maxmin system at each iteration,
697 only invalidate (and recompute) the modified parts.
698 This new feature is enabled in default models but you can try to
699 turn it on with "--cfg:maxmin-selective-update=1" for other models.
700 * Cas01 IMproved as default CPU model
701 This CPU model is the same Cas01 model, but it uses the
702 maxmin-selective-update flag and a heap structure to manage
703 actions on SURF kernel.
704 It reduces the complexity to find the next action to finish and,
705 consequently, it's faster than the old Cas01.
706 This is the new default CPU model (Cas01).
707 * Rename the old Cas01 model to Cas01_fullupdate
708 Keep the old cpu model Cas01 with the new name of Cas01_fullupdate.
709 Use "--cfg=cpu_model:Cas01_fullupdate" to use the old default CPU model.
710 * CpuTI (CPU Trace Integration)
711 A new CPU model whose objective is simulate faster when using
712 availability trace files.
713 Instead of using a full featured, over engineered maxmin system for
714 CPU modeling, this model does the pre-integration of traces files
715 to calculate the amount of CPU power available, and so, executes
716 faster than the old CPU models.
717 Use "--cfg=cpu_model:CpuTI" to change to this CPU model.
718 * Use LV08 as default network model since it gives better accuracy
719 for small messages and shouldn't change things for big ones.
720 Use --cfg=network_model:CM02 to get the previous behavior.
723 ******************************************
724 *DO NOT MIX 3.3.4 RESULTS WITH OLDER ONES*
725 ******************************************
726 * The new CPU model may changes simulations!
727 The point is that events occurring at the exact same timestamp
728 are not scheduled in the same order with the old and new
729 version. This may be enough to completely change the execution
730 of simulations in some cases.
731 * The new network model will change simulations!
732 This new model is more realistic than the previous one, so you
733 should consider redoing your old experiments with this model.
734 Sorry for the inconvenience.
737 * Introduce the supernovae compilation mode
738 When compiled that way, the whole SimGrid (or almost) is put in a
739 single compilation unit and compiled in one shoot.
740 This is to help gcc which has difficulties to inline stuff from one
742 The speedup seem to be above 15%, althrough more tests are needed on
743 amd64 to confirm that gain.
746 * Port of MSG's mailbox on top of SIMIX network
747 The put/get mechanism was greatly simplified on the way.
750 * New SIMIX network module. Provides:
751 - Mailbox: rendez-vous mecanism to find with who you want to speak
752 - Synchronous send/recv: easier and hopefully faster since the
753 logic is handled in the maestro process directly now
754 - Asynchronous send/recv: you dreamt of it? It's here now
755 Too bad that nobody cared enough to propagate the change to MSG.
756 * Add semaphores as SIMIX synchronization mechanism.
759 * new function SD_daxload(char*) to load a DAX file
760 (see http://vtcpc.isi.edu/pegasus/index.php/WorkflowGenerator)
761 * Introduce typed tasks. Specify its kind and cost at creation.
762 At scheduling, just give where it should be placed, and the cost
763 for each involved resource is automatically computed.
764 Existing constructors so far (more to come of course):
765 - SD_task_create_comm_e2e() for end-to-end communication
766 - SD_task_create_comp_seq() for sequential computation
767 Use SD_task_schedulev() / SD_task_schedulel() to schedule them.
768 * new function SD_task_dump() for debuging display
769 * new function SD_task_dotty(task,FILE*) writing to file the info
770 about the task in dotty format
771 * SD_task_dependency_exists() can now cope with having one of its
772 arguments NULL. If so, it tests whether the other argument has any
774 * Add getters on list of preceding/following tasks:
775 SD_task_get_parents(task) and SD_task_get_children(task)
776 * Add getters on amount of workstations and list:
777 SD_task_get_workstation_count(t) and SD_task_get_workstation_list(t)
778 * Add getter on task kind: SD_task_get_kind(task)
779 * Update the start_time and finish_time of tasks on completion/failure
780 * Bugfix: Remove task from state swags when destroyed
783 * New function: void gras_cpu_burn(double flops) -- a simple CPU burner
786 * New function: xbt_dynar_dopar(dynar,fun) to map a function over the
787 dynar with one separate thread per value of the dynar.
788 * Change the prototype of xbt_thread_create(), sorry.
789 Added a boolean parameter indicating whether we want to join this
790 thread (used in SG only for now)
791 * Implement xbt_thread_join and xbt_thread_yield in SG also.
794 * GTNetS wrappers should now be usable again (and betterly tested too)
795 * Fix a major regression from 3.2 where the timeout provided to
796 MSG_task_put_with_timeout() was used as absolute time before which
797 the comm should be done.
798 * Start to fix the <cluster> tag.
799 - Internal links should be good now (beside of the loopback, which
800 use the private link instead)
801 - paths to the external world is still rather broken
802 - the <route:multi> tag is just broken. Actually that's brain-dead.
803 We need sth like <route:multi src="myCluster" dst="$*-${myCluster}">
804 to make it less stupid
805 ** Check your platform with teshsuite/simdag/platforms/flatifier **
806 * Fix a source-level compatibility glitch from 3.2: after defining
807 MSG_USE_DEPRECATED, you can use the old name
808 MSG_task_put_with_time_out() for MSG_task_put_with_timeout()
809 * Allow to compile from the SVN with automake 1.11
810 * Fix some problems when using the "start_time" tag in deployment XMLs.
811 * Fix #8569: XBT/synchro.h has redundant declarations
812 * Fix #8563: MSG return values and exceptions
813 Introduce a MSG_TIMEOUT_FAILURE return code and use it consistently.
814 * Integrate patch #8636: Obey DESTDIR when installing documentation.
815 Thanks to Robson Peixoto.
816 * Fix a vicious bug in dictionaries inducing that some elements were
817 not freed on xbt_dict_free()
819 Portability report of this version:
820 * Main portability targets:
821 - linux (ubuntu (804/810/910) /debian (4/5/testing) /fedora (core11))
823 - mac leopard on i386
824 Known problems: http://cdash.inria.fr/CDash/index.php?project=Simgrid
825 but nothing critical.
826 * Other platforms: windows, AIX and others were not tested for this release
828 Timing report of this version:
829 * Lazy evaluation brings arbitrary speedup (ie, speedup depending on
830 scenario parameters). From 8h to a few seconds in desktop grid settings.
831 * Supernovae brings about 25% speedup on i386.
833 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 24 Dec 2009 19:07:39 +0100
835 SimGrid (3.3.3) stable; urgency=low
837 The "Need for Speed" release.
839 The timings done to validate the 3.3.2 were faulty.
840 Instead of being 5% faster, it was 15% slower (compared to 3.3.1).
842 The problem was a conversion from a manually handled vector to
843 xbt_dynar_t on the critical path.
844 xbt_dynar_foreach calls functions, inducing stack management crap.
846 We inlined these functions and xbt_dynar_foreach is now breath taking.
847 We also inlined xbt_swag_belong on the way.
849 Here are some approximate speedup measurements (on master/slaves
850 simulations lasting between 10s and 20s each):
851 3.3.1 -> 3.3.2: about same performance
852 3.3.2 -> 3.3.3: 40% speedup
853 3.3.1 -> 3.3.3: 40% speedup
854 3.3.1 with inline patch -> 3.3.3: 30% speedup
856 Our reading is that the refactoring which occurred in 3.3.2 made us
857 suffer much more from the xbt_dynar_foreach low performance, but
858 once we solved this, this refactoring proved to be very performance
859 effective. From the 40% speedup, somehow, 10% are due to the
860 inlining and 30% to the refactoring.
862 That's a pitty that gcc cannot inline functions placed in other files
863 alone. We have to choose between:
864 - break the encapsulation (by putting private data structures and
865 accessors in headers files to help gcc)
866 - live with low performance
867 - switch to a decent compiler such as icc (not quite possible).
869 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 20 Aug 2009 21:21:33 +0200
871 SimGrid (3.3.2) stable; urgency=low
873 The "Simplicity does not preceed complexity, but follows it" release.
875 The main contributors of this release were (lexical order):
876 Silas De Munck, Stéphane Genaud, Martin Quinson, Cristian Rosa.
879 * Extract the routing logic into its own object.
880 (was dupplicated in network.c and workstation_LV07.c;
881 Allows to implement other ways of storing that info)
882 => kill now useless network_card concept
883 - Use dynar to represent routes (instead of void** + int*)
884 - kill link_set (use surf_network_model->resource_set instead)
885 - Add a command-line option to choose the routing schema to use
886 - Add three new models:
887 * Floyd (shortest path computed at initialization)
888 * Dijikstra (shortest path recomputed all the time)
889 * Cached Dijikstra (shortest path computed on need)
890 All these models where contributed by Silas De Munck, and are
891 described in his ICCS09 paper.
893 * Simplify model declaration
894 (less redirections, less function to write when defining a model)
895 - Factorize stuff between models:
898 surf_model_resource_set(model)
899 surf_model_resource_by_name(model, name)
900 - Unify the types of models in s_surf_model_t (using an union)
901 - Embeed fields of common_public directly into s_surf_model_t
902 - Rename model methods:
903 action_free ~> action_unref
904 action_change_state ~> action_state_set
905 action_get_state ~> action_state_get
906 - Change model methods into functions :
907 (model)->common_public->action_use ~> surf_action_ref
909 * Implement a generic resource; use it as ancestor to specific ones
910 (allows to kill duplicated code in models)
911 Drawback: timer command don't need no name nor properties;
912 workstation_CLM03 don't need no properties
913 (but I guess we can live with those few bytes wasted)
915 * Improve the action object model
916 - implement a constructor avoiding dupplicated code about field
917 initialization in generic_action part.
919 * Kill the SDP model: it has an external dependency, is deprecated
920 in flavor of modern lmm models, and didn't compile since a while
923 * Relocation of the context module from XBT to SIMIX.
924 (the context were decoupled from the simix processes, duplicating a lot of code)
925 => a lot of code was factorized
926 - less overhead is introduced during scheduling
927 - simpler API for the context factory
928 - the logic for process creation,destruction and manipulation was simplified
929 * Simplification of the s_smx_process_t data structure.
930 => accesing the simix level data associated to a process is faster now,
931 and the code is a lot more readable.
934 * Implement some more MPI primitives:
935 MPI_Bcast, MPI_Waitany, MPI_Waitall, MPI_Reduce, MPI_Allreduce, MPI_Scatter, MPI_Sendrecv, MPI_Alltoall
936 -implementation: Bcast: flat or 2-ary tree (default),
939 Allreduce: Reduce then Bcast
940 Alltoall: "basic_linear" if data per proc < 3Kb, "otherwise pairwise".
941 Not yet implemented: "Bruck" for data per proc < 200b and comm size > 12
942 Alltoallv: flat tree, like ompi
944 * Add support for optimized collectives (Bcast is now binomial by default)
945 * Port smpirun and smpicc to OS X
948 * Kill SD_link_get_properties: hard to maintain and makes very little sense
949 Shout out if you used it.
952 * Display the list of still queued messages in SG mode when existing
956 * Add xbt_set_get_by_name_or_null() [Silas De Munck]
957 * Add xbt_graph_node_get_outedges() [Silas De Munck]
958 * Add xbt_str_from_file(FILE*)
959 * Add xbt_dict_get_key achieving a linear reverse search
960 * Remove the context module
962 Portability report of this version:
963 * Main portability targets:
964 - Linux(debian)/x86/context
965 - Linux(debian)/x86/pthreads
966 - Linux(debian)/amd64/context
967 - Linux(debian)/amd64/pthreads
968 On these, we still have the eratic breakages of gras/pmm and
969 amok/saturate_sg reported in previous version. We still think
970 that the tests are the cause of the fault, not the tested code.
972 - Mac OSX Leopard/x86/context
973 Still false negative in tesh autotesting.
974 Smpi still fails, but this time because readlink does not accept -f
975 Everything seems to work properly beside of that.
978 - AIX version 5.3 (only tested contexts this time)
979 Smpi still fails there because mktemp is not installed.
980 Everything seems to work properly beside of that.
982 I managed to compile it for the first time, but several breakages.
983 Won't delay the release for this exotic platform.
985 * Windows: it's still lagging behind. If you want to help, please
988 Timing report of this version:
989 This version seem to be more than 5% faster than 3.3.1 (on linux
990 64bits with contextes). The gain is less than expected, we are
991 investigating this for next release.
993 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Wed, 19 Aug 2009 17:07:12 +0200
995 SimGrid (3.3.1) stable; urgency=low
998 * Implement a --cfg-help to show existing configuration variables
999 * Build chain do not require doxygen in maintainer mode
1002 * fix a bug on struct sizeof computation, which prevented the
1003 exchange of arrays of structs in some conditions
1004 - added a regression test about this in datadesc_usage
1005 * Allow the exchange of 0-long dynamic vectors.
1006 - for that, use -1 as indicator of dynamic size instead of 0
1007 - This implied to change any size from unsigned long to long,
1008 reducing a bit communication abilities, but I guess that with
1009 64bits being quite common, this is more than enough.
1010 - This also induce a protocol change, thus bumping network protocol
1011 version from 0 to 1 (if we have external users, we have to get
1012 clean on that point too ;)
1013 - added two regression tests about this in datadesc_usage
1014 * Be more verbose when propagating local exceptions
1015 This helps debugging.
1016 * Display the status of simulated processes when receiving SIGINT in
1020 * Allow to control the simulation from a trace file.
1021 New functions MSG_action_register() and MSG_action_trace_run()
1022 The first one allows to associate a function execution to each
1023 kind of action while the second one parses a trace file and
1024 triggers the corresponding actions within the system.
1025 For now, only a toy example is provided in examples/msg/actions
1026 * Add an exemple of process migration in examples/msg/migration
1027 * Fix a bug in task exchange which broke MSG_task_get_sender()
1028 Add a teshsuite regression test for that.
1029 [Bug: if MSG_task_get_sender() is called after sender exit,
1031 * Fix a bug which prevented suspend/resume to work properly
1032 * Display the status of simulated processes when receiving SIGINT
1033 This fixes a regression of v3.3. due to the introduction of SIMIX
1034 * Bug fixing in failure management:
1035 - trace could not start by a failure at time 0
1036 - failure during communications were not working
1039 * Add SIMIX_process_set_name() to change the name of the current
1040 process in the log messages.
1041 * Store smx_hosts in a dict since we only retrieve them by name
1042 * Move the configuration infrastructure to surf
1045 * Move the configuration infrastructure to surf
1048 * Massive internal cleanups:
1049 - Store internal structures on processes instead of hosts (allows
1050 to have more than one process per host, in addition of being more
1052 - Cleanup the initialization/finalization process
1053 - Kill a whole bunch of unneeded synchronization:
1054 processes run in exclusive manner within the simulator
1055 - Move queues from global tables to process data fields
1057 - now accept -platform and -hostfile arguments
1058 - Pass the right rank value to processes according to the hostfile
1059 * Compile the examples by default, and use them as regression tests
1060 * Implement MPI_Wtime()
1061 * Change the reference speed to a command line option
1064 * TCP_gamma can now be specified as command line option using
1065 --cfg=TCP_gamma:10000000.0
1066 * Change the --surf-path cmd line option into --cfg=path:
1069 * Also include strbuff from xbt.h public header
1070 * xbt_ex_display(): do not free the exception after displaying
1071 This allows to do more with the given exception afterward.
1072 Users should call xbt_ex_free() themselves.
1076 Portability report of this version:
1077 * Main portability targets:
1078 - Linux(debian)/x86/context
1079 - Linux(debian)/x86/pthreads
1080 - Linux(debian)/amd64/context
1081 - Linux(debian)/amd64/pthreads
1082 These targets fail about 1/10 of times on gras/pmm, but we believe
1083 that this is because of the test, not because of simgrid.
1084 amok/saturate_sg fails even more rarely, and the test may not be
1087 - Mac OSX Leopard/x86/context
1088 The test suite still spits tons of errors because some obscure
1089 force prevents us from removing the temporary directories
1090 arguing that they still contain some metadata I've never heard of.
1091 Smpi fails because seq is not installed.
1092 Everything seems to work properly beside of that.
1095 - AIX version 5.3 (both contexts and pthread)
1096 Smpi still fails there because mktemp is not installed.
1097 XML inclusions seems rosty on AIX.
1099 * Windows: it's still lagging behind. If you want to help, please
1102 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sat, 27 Jun 2009 00:14:30 +0200
1104 SimGrid (3.3) stable; urgency=high
1108 * JAVA BINDINGS for MSG (you dreamt of them? We made them)
1109 [Malek Cherier & Mt]
1111 * Introduce the SIMIX module: factorize code between MSG and GRAS.
1114 Until now, GRAS were using MSG as an interface to SURF. It was
1115 quite difficult because both interface have several differences
1116 (MSG channels vs GRAS sockets were the most notable point).
1118 This also opens the gate to SMPI (which should occur soon) and speed
1119 up simulations by to 40% (even if it were not the main goal).
1121 **************************************
1122 *DO NOT MIX 3.2 RESULTS WITH 3.3 ONES* Simix may changes simulations!
1123 **************************************
1124 The point is that events occuring at the exact same timestamp are
1125 not scheduled in the same order with the old and new version. This
1126 may be enough to completely change the execution of simulations in
1127 some cases. Sorry for the inconvenience.
1129 * Cleanup and upgrade the XML format to push further scalability
1130 issues (check http://hal.inria.fr/inria-00256883/ for more info)
1132 * Improve the testing infrastructure with tesh. Now a very large part of
1133 the code is tested not only by being run but also by checking that the
1134 output match an expected output [Mt].
1136 * Move on to FleXML v1.7 for the embeeded XML parsers. This version
1137 is really less memory-demanding, which should allow you to use
1138 larger files in SimGrid [AL].
1140 * Inform valgrind about our contextes, so that it becomes usable
1141 with the default (and more effecient) version of SimGrid
1142 [contributed by Sékou Diakite, many thanks]
1145 * Introduce a listener thread in charge of receiving incomming
1146 messages from the network. It allows to overlap communication and
1147 computation but most notably, it removes some stupid deadlocks due
1148 to the fact that so far, a process could not send and receive at
1149 the same time. This made most non trivial communication schema
1151 * Convert the PIDs from long int to int to match the MSG ones (and
1152 linux ones too) [Mt]
1153 * New function: gras_agent_spawn() to launch a new process on
1154 current host. Only working in simulation for now. [Mt]
1155 * New function: gras_os_hostport() returning a constant form (ie,
1156 not needing to be freed) of "gras_os_hostname():gras_os_myport()"
1159 * Make the backtrace of exceptions more human readable [Mt]
1160 * New module: xbt/str [Mt]
1161 a ton of string utility functions (split, join, printf to a newly
1162 allocated buffer, trim, etc)
1163 * New module: xbt/hash [Mt]
1164 SHA1 hashing algorithm (more to come if needed)
1165 * New module: xbt/synchro [Mt]
1166 synchronization tools (mutex and conditions) working the same way
1167 in simulation and in real life (mainly useful for GRAS, but not
1169 * New module: xbt/queue [Mt]
1170 classical producer/consumer synchronization scheme
1171 * xbt_dynar_new_sync() creates a synchronized dynar. All access
1172 (using the classical functions will get serialized) [Mt]
1173 * Make dictionary internal table dynamic. No need to specify its size
1174 anymore; functions xbt_dict_new_ext() and xbt_dict_hashsize_set()
1176 * Make sure the log channels are organized as a tree under windows
1177 (because of ANSI C compatibility issue, any channel were child of
1178 root directly) [Mt].
1181 * Cleaned many thing in surf and fixed a few bugs [AL].
1182 * Add a nice command line configuration mechanism to compose models [AL].
1183 * Add a new model for parallel tasks (ptask_L07) that is less buggy than
1184 the previous one (KCCFLN05). It relies on something that looks like
1185 a max-min sharing mechanism but cannot be written as such. A new solver
1186 was thus designed [AL].
1187 * Add a new solver to lmm. Based on Lagrange optimization and
1188 gradient-based descent, it enables to efficiently maximise systems s.a
1190 sum f_i(x_i) s.t Ax<= b with A_{i,j}>=0 and f_i a concave function.
1192 This solver enables to propose two new network models for TCP Reno and
1193 TCP Vegas based on Low's work. These models still need to be fully
1194 tested though [Pedro Velho].
1197 * Bug fix in SD_simulate. Now the time bound given as argument is
1199 * Use the new parallel task model (ptask_L07) as default.
1200 * Use the SURF command line configuration mechanism.
1201 * 0-size tasks (for synchronization) should now work.
1203 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Sun Apr 12 05:20:36 CEST 2009
1205 SimGrid (3.2) stable; urgency=high
1209 We still experience issues on this platform, but we believe that at
1210 least MSG is usable.
1212 GRAS API BREAKAGE (for simplification purpose, sorry):
1213 * the gras_msgtype_by_name is not used anymore. Instead of
1214 gras_msg_send(toserver, gras_msgtype_by_name("request"), &request);
1215 you can write (and must)
1216 gras_msg_send(toserver, "request", &request);
1217 - If you still want to pass a gras_msgtype_t to the function (to cache
1218 the type and avoid the lookup time), use the gras_msg_send_() variant.
1219 - Impacted functions:
1220 gras_cb_register, gras_cb_unregister, gras_msg_send, gras_msg_wait,
1221 gras_msg_rpccall, gras_msg_rpc_async_call, gras_msg_wait_ext
1222 * The callbacks are now expected to return 0 when everything went well
1223 (just like the main() function)
1225 GRAS new features and improvements:
1226 * New module mecanism where user code can use per process globals [Mt]
1227 This is similar to gras_userdata_*() functions, but for libraries. It
1228 factorize some code developped over and over in the examples and AMOK.
1229 It has still to be documented and used (only amok/peermanagement is
1231 * Fix a vicious bug in the TCP buffering mecanism which leaded to message
1232 loss when they were small enough to fit into the buffer and sent quickly
1233 enough so that they can all get received in one shoot.
1234 * gras_datadesc_by_name and gras_msgtype_by_name: now raise an exception
1235 if not found. Use the *_or_null() variant for the old semantic.
1236 * In gras_msg_handle, do not discard messages without callback.
1237 They are probably messages to be explicitly awaited later (ie, proofs of
1238 mis-synchronization in userland since they are sent before being awaited)
1239 No big deal usually.
1240 * gras_socket_meas_send/recv: semantic changed!
1241 The numerical arguments used to be (1) the total amount of data to send
1242 and (2) msg_size. This was changed to (1) msg_size and (2) amount of
1243 messages. This was need for the fool willing to send more than MAXINT
1244 bytes on quite fat pipes.
1247 * Do really rename the hostmanagement module to peermanagement. [Mt]
1248 Ie, rename functions from amok_hm_* to amok_pm_*. This breaks the API,
1249 but this is rather new and this was documented in the module
1250 documentation (poor excuses, I admit)
1251 * Bandwidth measurement semantic changed! This follows the changes to
1252 gras_socket_meas_send/recv explained above.
1255 * A sequential mode has been added to the workstations. When a workstation
1256 is in sequential mode, it can execute only one task, and the other tasks
1257 are waiting in a FIFO. [Christophe Thiery]
1260 * The KCCFLN05 workstation model now handles parallel tasks. It is the
1261 model for SIMDAG. [Christophe Thiery]
1262 * Bug fix in the maxmin solver: Some values were close to 0 instead of
1263 equal to 0, which caused some bad behaviors in
1264 saturated_constraint_set_update. I now use a threshold mechanism like in
1268 * When running manually src/testall, you select specific units [Mt]
1269 testall is the result of our cunit mecanism, and should replace all
1270 the scripty thingy around since bash don't run easily on billware.
1272 * A mallocator system has been added. [Christophe Thiery]
1273 Mallocators allow you to recycle your unused objects instead of freeing them
1274 and allocating new ones.
1276 Documentation update:
1277 * FAQ reworking + New FAQs:
1278 - "Valgrind spits tons of errors!" [Mt]
1279 - "How to repport bugs" [Mt]
1280 - "Cross-compiling a Windows DLL of SimGrid from Linux" [Mt]
1281 - "What is the difference between MSG, SimDag, and GRAS?" [Mt]
1282 - Communication time measurement within MSG [AL]
1283 - I experience weird communication times when I change the latency [AL]
1284 * GRAS tutorial [Mt]
1286 - an introduction to the framework and to the used communication model
1287 - an initiatic tour introducing the most proheminent features:
1289 . Lesson 0: Installing GRAS
1290 . Lesson 1: Setting up your own project
1291 o Part 2: Message passing
1292 . Lesson 2: Exchanging simple messages
1293 . Lesson 3: Passing arguments to the processes (in SG)
1294 . Lesson 4: Attaching callbacks to messages
1295 . Lesson 5: Using globals in processes
1296 . Lesson 6: Logging informations properly
1297 . Lesson 7: Using internal timers
1298 . Lesson 8: Handling errors through exceptions
1299 . Lesson 9: Exchanging simple data
1300 . Lesson 10: Remote Procedure Calling (RPC)
1301 . Lesson 11: Explicitely waiting for messages
1302 . Recapping of message passing features in GRAS
1303 - A HOWTO section containing:
1304 o HOWTO design a GRAS application
1305 More are due, of course. They will come latter. In the meanwhile, you can
1306 check the examples which are still here.
1308 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri Mar 16 21:11:46 CET 2007
1310 SimGrid (3.1) stable; urgency=high
1314 There was a stack corruption somewhere, visible only when optimizing
1315 with these versions. [Vince]
1318 * This is a NEW module! SimDAG (SD for short) is a revival of the old SG
1319 module that enabled to play with Directed Acyclic Graphs. It is built
1320 directly on top of SURF and provides an API rather close to the old
1321 SG. Some old codes using SG are currently under rewrite to check that
1322 all needful functions are provided. [Christophe Thiery]
1325 * Complete rewrite of the KCCFLN05 workstation model. It is now an
1326 extension of the classical CLM03 model that gracefully handles
1327 failures. This is now the default model for MSG and GRAS. It doesn't
1328 handle parallel tasks yet though. [AL]
1329 * Bug fix: Weights were not correctly set in the network part.
1330 WARNING: This may have resulted in incorrect results with simulations
1331 where there are more than one flow on a given link. [AL]
1334 * After a (long ?) discussion on simgrid-devel, we have decided that the
1335 convention we had on units was stupid. That is why it has been decided
1336 to move from (MBits, MFlops, seconds) to (Bits, Flops, seconds).
1337 WARNING : This means that all previous platform files will not work as
1338 such with this version! A warning is issued to ask users to update
1340 A conversion script can be found in the contrib module of the CVS, under
1341 the name contrib/platform_generation/surfxml_update.pl [MQ]
1344 * Bug fix: Processes were started in reverse order, wrt deployment file.
1345 WARNING: if your code relies on this bug, please fix it. [AL]
1346 * Bug fix: Add a test in MSG_task_execute to stop whenever a task is
1347 being executed on two different locations. [AL]
1348 * Bug fix: Failures are now better supported thanks to Derrick's tests
1349 (there was many failure situations I hadn't thought of and that weren't
1350 correctly handled). [AL]
1351 * New function: MSG_host_is_avail indicates you whether a given m_host_t
1355 * New! a real RPC mecanism, as it ought to be since too long. [MQ]
1356 Exception occurring on server-side are propagated back to client (!).
1358 API CHANGE: the callback changed their prototype. Change:
1359 int my_handler(gras_socket_t expeditor, void *payload_data) {
1361 int my_handler(gras_msg_cb_ctx_t ctx , void *payload_data) {
1362 gras_socket_t expeditor=gras_msg_cb_ctx_from(ctx);
1364 * New! function: gras_msg_handleall to deal with all messages arriving
1365 within a given period.
1366 * New! function: gras_socket_server_range to get a server socket in a
1367 range of port numbers (ease to avoid port number conflicts) [MQ]
1368 * New! gras processes display their backtrace when they get a SIGUSR1
1369 or when Ctrl-C is pressed. Use Ctrl-C Ctrl-C to exit.
1370 Sweet to debug RL processes [MQ]
1374 - Do not force experiment sizes to be expressed in kb, or it becomes
1375 impossible to measure the latency this way (needs one byte-long tests)
1376 WARNING: this changes the amok_bw_* function semantic. [MQ]
1377 - Implements the link saturation stuff. [MQ]
1378 * Peer management module:
1379 New! module factorizing code that we wrote over and over [MQ].
1382 * New module: cunit (my jUnit implementation in ansi C) [MQ]
1383 - Test units are placed directly into the library code, they get extracted
1384 automatically and placed into the src/testall binary.
1385 - Convert most of the XBT tests to this system.
1386 * New functions: xbt_dynar_getfirst_as() and xbt_dynar_getlast_as() [MQ]
1387 * XML parsing: rewrote parts of flexml to enable multiple xml parsers to
1388 live in the same C code. This required to change a little bit the API
1389 of surfxml parsing but shouldn't be an issue for end-users. [AL]
1390 * New module: sparse graph structure with basic algorithms (this is work
1391 in progress and the API is not considered to be frozen yet). [AL]
1392 * Display more information on backtraces: source line & function names are
1393 now displayed just like valgrind does (rely on addr2line tool) [MQ]
1394 * New function: xbt_backtrace_display(). Sweet while debuging [MQ]
1395 * Reworked a little bit some #include statements to load only required
1396 headers. Some user code that relied on SimGrid to include stdlib or
1397 stdio may need to include it by themselves. [AL]
1398 * Fixed xbt/log.h. A missing SG_BEGIN_DECL prevented compilation with
1400 * Renamed xbt_host_t into xbt_peer_t since it betterly describes what I
1401 meant. This breaks the API of AMOK and of xbt/config. Sorry about this,
1402 but I guess that almost nobody used those parts. [MQ]
1404 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 14 Jul 2006 01:32:27 +0200
1406 SimGrid (3.0.1) stable; urgency=low
1409 * Unfortunately, I had missed 5 misnamed functions:
1410 xbt_fifo_item_t xbt_fifo_newitem(void);
1411 void xbt_fifo_freeitem(xbt_fifo_item_t);
1412 xbt_fifo_item_t xbt_fifo_getFirstItem(xbt_fifo_t l);
1413 xbt_fifo_item_t xbt_fifo_getNextItem(xbt_fifo_item_t i);
1414 xbt_fifo_item_t xbt_fifo_getPrevItem(xbt_fifo_item_t i);
1415 They're now deprecated. Please use their new versions:
1416 xbt_fifo_item_t xbt_fifo_new_item(void);
1417 void xbt_fifo_free_item(xbt_fifo_item_t);
1418 xbt_fifo_item_t xbt_fifo_get_first_item(xbt_fifo_t l);
1419 xbt_fifo_item_t xbt_fifo_get_next_item(xbt_fifo_item_t i);
1420 xbt_fifo_item_t xbt_fifo_get_prev_item(xbt_fifo_item_t i);
1422 * Bugfix: really disconnect fifo items which are remove_item()ed [AL]
1423 * Documentation: xbt_log module unmercifully reworked [MQ]
1424 * Bugfix: there was a problem with the ending of contexts with
1425 the pthread backend. It caused some weird deadlock or behavior
1426 depending on the pthread implementation. [AL]
1427 * Bugfix: get the exceptions raised in the simulator repport where
1428 and why they come from when they are not catched in time [AL, MQ]
1431 * Bugfix: Do repport the error when two non-connected hosts try to
1432 exchange data (Thanks to Flavien for stumbling into this one) [AL]
1435 * Add additionnal checkings on communications. Assert that two
1436 communicating hosts are connected by a set of links... [AL]
1439 * Add additionnal checkings on channel values in communication [AL]
1440 * New: MSG_task_get_source to see on which host a task was generated [HC]
1441 * New: int MSG_task_probe_from_host(int channel, m_host_t host): returns
1442 the number of tasks waiting to be received on channel and sent
1444 * New: MSG_error_t MSG_task_get_from_host(m_task_t * task, int channel, m_host_t host);
1445 waits for the first task coming from a given host.. [AL]
1447 GRAS new functionnalities: [MQ]
1448 * Enhance the parsing macro to allow the size of multidimentional objects
1449 to be given thru annotations.
1450 * New example (and documentation): Matrix Multiplication a la RPC
1451 (as when I was young!) and fix a bunch of bugs found on the way.
1453 GRAS performance improvements: [MQ]
1455 * Reduce the amount of cbps creation/destruction by making it static to
1456 datadesc_send/recv() and using a (newly created) cbps_reset (based on
1459 * Change libdata to a set so that we can search for stuff by ID (and thus
1460 reduce the insane amount of dict lookups)
1463 * Actually implement gras_datadesc_copy() so that we don't have to mimick
1464 RL communication on top of SG since it's so uneffective.
1465 It may also be used for inter-thread communication in RL, one day.
1466 * Use gras_datadesc_copy() to exchange messages on top of SG
1468 - improve message exchange performance on top of SG
1469 - deprecate transport_plugin_sg.c:gras_trp_sg_chunk_send() & recv()
1470 * Don't exchange on the network the size of the used part of buffer,
1471 instead, specify the possible buffer size to read().
1473 - reduces the amount of read/write calls (one pair per exchange)
1474 - reduces the amount of exchanged data (the size)
1475 - allows to retrieve all arrived data on receiver side, if we don't need
1476 it right now (subsequent read will peek the buffer)
1477 - allows the receiver to proceed with the begining of the stream before
1478 everything is arrived
1479 - make it possible to build an iov transport (using readv/writev)
1481 - take care of the data with non-stable storage (like stacked data),
1483 * If possible, TCP send uses vector I/O (when writev() is here)
1484 - Don't use it for receive since we send data sizes and data on the
1485 same stream, so we wouldn't be able to chain large amount of chunks
1486 before having to flush the stuff to read the size.
1487 * Rework the transport plugin mecanism to simplify it and reduce the
1488 amount of pointer dereferencement when searching for the right function
1491 * I guess that now, we do almost as few system calls as possible while
1492 doing as few data copy as possible.
1494 To improve it further, we could try to send all the sizes first and then
1495 all the data (to use iov on receiving size), but it's only a partial
1496 solution: when you have 2 dimensional data, the sizes of the second
1497 dimension is data of the first dimension, so you need 3 streams.
1499 I'm not sure the potential performance gains justify the coding burden.
1501 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 21 Oct 2005 14:42:20 +0200
1503 SimGrid (3.00) stable; urgency=high
1506 * New! Give the possibility to hijack the surf parser and thus bypass
1507 MSG_create_environment and MSG_launch_application. Have a look at
1508 examples/msg/msg_test_surfxml_bypassed.c to see how it can be done.
1510 -- Arnaud Legrand <simgrid-devel@lists.gforge.inria.fr> Sat, 20 Aug 2005 23:25:25 -0700
1512 SimGrid (2.96) unstable; urgency=low
1517 * New! Exception handling with setjmp or such (code from OSSP ex) [MQ]
1518 This deprecates the xbt_error_t mecanisms.
1519 It modifies (simplifies) all XBT and GRAS API.
1520 MSG API keeps unchanged (exceptions raised by XBT are catched from
1521 within MSG and masked with existing error handling facilities)
1524 * New! Add a FATPIPE model. [AL]
1525 * New! Add a parallel task model. [AL]
1526 * New! Add automatically a loopback interface (in the default
1527 network model) if none was precised.
1530 * Bugfix: MSG_process_resume now works with the current running process.
1532 * New! Add MSG_parallel_task_create and MSG_parallel_task_execute. [AL]
1533 * Modification of MSG_task_get_compute_duration. Once a task has been
1534 processed, the value returned by this function is now equal to 0. [AL]
1535 * New! Add double MSG_task_get_remaining_computation(m_task_t task) and
1536 MSG_error_t MSG_task_cancel(m_task_t task). Add a state
1537 (MSG_TASK_CANCELLED) to MSG_error_t corresponding to the cancelation
1538 of a m_task. For now, MSG_task_cancel only works with computation
1540 * New! Add double MSG_get_host_speed(m_host_t h) that returns the speed
1541 of the processor (in Mflop/s) regardless of the current load on the
1543 * API Change: use proper naming convention for MSG_getClock and
1544 MSG_process_isSuspended: MSG_get_clock and MSG_process_is_suspended.
1546 * New! Add void MSG_task_set_priority(m_task_t task, double priority).
1547 This function changes the priority of a computation task. This priority
1548 doesn't affect the transfer rate. A priority of 2 will make a task
1549 receive two times more cpu power than the other ones. This function
1550 has been added to suit the needs of Nguyen The Loc and hasn't been that
1551 much tested yet. So if it fails, please report it and send me your code.
1553 * API Change: removed all functions and types that were marked "deprecated"
1554 since many months. Renamed MSG_global_init_args to MSG_global_init.
1556 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Mon, 8 Aug 2005 17:58:47 -0700
1558 SimGrid (2.95) unstable; urgency=low
1561 * Steal some nice code to GNU pth to fix context detection and usage [AL]
1562 * Cleanup in the xbt_config API; add configuration callbacks. [MQ]
1563 * Cleanup in the initialization API: the unused "defaultlog" is dead. [MQ]
1566 * Bugfix: Allow absolute paths for platform description files [MQ]
1567 * Bugfix: do free the variables after use. Leads to drastic performance
1569 * Implement max_duration (ie, timeouts) on resources [AL]
1572 * Implement MSG_config to configure MSG at runtime. xbt_cfg test on a real
1574 * Implement MSG_channel_select_from() to help GRAS now that SURF provide
1575 the needed support (timeouts) [AL]
1578 * Implement measurement sockets. You can now get the bandwidth between two
1579 hosts thanks to AMOK (see below). [MQ]
1580 * gras_datadesc_dynar() builds a dynar type descriptor, allowing to send
1581 dynar over the network (yeah) [MQ]
1582 * Real (even if simplistic) implementation of gras_os_myname() on RL [MQ]
1583 * simple/static token-ring example. [Alexandre Colucci and MQ]
1584 * Use MSG_channel_select_from() and remove the *slow* hack we had to put
1585 in place before [MQ]
1588 * Differentiate the types "char[22]" and "unsigned char[22]" in automatic
1589 type parsing. "short" and "long" modifiers were also ignored; other
1590 modifier (such as reference level) are still ignored. [MQ]
1591 * Embeed the buffer size within the buffer itself on SG. [MQ]
1592 That way, send() are atomic and cannot get intermixed anymore (at least
1593 the ones which are less than 100k; bigger messages still have the issue)
1594 * Array size pushed by the field, not by the field type (or each
1595 and every long int will push stuff to the cbps) [MQ]
1596 * use select() to sleep since it allows to portably sleep less than one
1599 GRAS (minor cleanups)
1600 * <project>.Makefile.local (generated from gras_stub_generator) |MQ]:
1603 * Type Callbacks now receive the gras_datadesc_type_t they work on as argument.
1604 * type category 'ignored' killed as it was never used and were difficult
1606 * whether a type can cycle or not is now a flag, leaving room for more
1607 flags (as "ignored", if we feel the need one day ;)
1608 * Rename raw sockets to measurement sockets since "raw" has another
1609 meaning in networking community.
1612 * Advanced Metacomputing Overlay Kit introduction. It is based over GRAS
1613 and offers features not belonging to GRAS but that most applications
1614 need. One day, it may be a set of plugins loadable at runtime.
1615 * New module: bandwidth
1616 bandwidth measurement between arbitrary nodes running this module. [MQ]
1618 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Thu, 30 Jun 2005 16:29:20 -0700
1620 SimGrid (2.94) unstable; urgency=low
1622 The first beta release of SimGrid 3 !
1626 * Update the main page and the FAQ. Adding references to gforge.
1629 * Add a gras_os_getpid function.
1632 * Add MSG_task_get_compute_duration() and MSG_task_get_data_size()
1633 * Extend the logs so that they also print PID, hostname, date, ... if
1635 * Convert the MSG example to the use of xbt_logs instead of PRINT_MESSAGE,
1636 and kill the old version which were in testsuite/
1637 * Rewrite tools/MSG_visualization/colorize.pl for using with logs instead
1641 * Add xbt_os_time(). As the rest of xbt/portability, this is not public
1642 for users. Instead, each programming environment (GRAS, MSG,...) use it
1643 when needed to provide such a feature to users.
1644 Don't shortcut the mecanism or you will also shortcut the virtualization
1645 you need on the simulator.
1649 * Cleanups in configury with regard to compile optimization/warning flags.
1650 Also add -fno-loop-optimize to any powerpc since it's the optimization
1651 killing gcc (< 3.4.0).
1652 * Doxygen cleanups: move MSG examples, kill the second Doxygen phase
1653 needed by MSG examples complications
1654 * Borrow configury beautifications from PHP
1657 * Bugfix: XBT_LOG_NEW_DEFAULT_CATEGORY now compiles without compiler
1658 warning (thanks loris for stumbling into this one).
1659 * Bugfix: stop loading private headers (gras_config.h) from the public
1663 * Change SIMGRID_INSTALL_PATH to GRAS_ROOT in Makefiles generated for user.
1664 * Rename gras_get_my_fqdn to gras_os_myname and implement it in the simulator
1665 RL would imply a DNS resolver, which is *hard* to do in a portable way
1666 (and therefore delayed).
1667 * Implement a real timer mecanism and use it in timing macros. This allows
1668 to avoid rounding errors and get a 0.000005 sec precision in timing
1669 macros. While I was at it, various cleanups:
1670 - allow to declare more than one timed section per file (fix a stupid bug)
1671 - move some private declaration to the right place
1672 - merge conditional execution and timing macros into emulation module
1673 - document the module
1674 - make sure the module cleanups its mess on gras_exit
1675 * Documentation improvements:
1676 - (new) how to compile applications using GRAS
1677 - (new) emulation support (timing macros)
1679 -- Da SimGrid team <simgrid-devel@lists.gforge.inria.fr> Fri, 13 May 2005 10:49:31 +0200
1681 SimGrid (2.93) unstable; urgency=low
1683 Alpha 4 on the path to SimGrid 3 (aka the "neuf-trois" version)
1686 - Use Paje properly where used. Still to be sanitized properly.
1687 - Portability fix: Add an implementation of the contexts using pthread
1691 - Add xbt_procname(): returns the name of the current process.
1692 Use it to show the current process's name in all logging.
1694 - fix detection of older flex version and the reaction, since we do
1695 depend on modern ones (we use lex_destroy)
1696 - Better separation of SG and RL in the libs: remove all simulation code
1697 from libgras. As a result, this lib is now only 200k when stripped.
1698 Some of the xbt modules may also be duplicated (two sets and such) and
1699 should be cleaned/killed before SG3.
1700 - Insist on using xlC on AIX because of weird problems involving gcc there.
1701 - Cleanup the make remote stuff. This is now done by scripts
1702 tools/graspe-{master,slave} (GRAS Platform Expender). This is still
1703 mainly for our private use, but we're working on changing them to user
1706 - Bugfix: flush the socket on close only if there is some *output*.
1707 - Bugfix: flush idempotent when there's nothing to send (don't send size=0)
1709 - Add MSG_task_get_name. The task names are mainly for debugging purpose,
1712 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Fri, 4 Mar 2005 14:32:37 -0800
1714 SimGrid (2.92) unstable; urgency=low
1716 Alpha 3 on the path to SimGrid 3
1720 - New! First try of benchmarking macros.
1721 - New! First try so that gras_stub_generator generate deployment and
1722 remote compilation helpers.
1724 - Bugfix: Initialization fix in msg_test.
1728 - Bugfix: applied patch to lexer so that it doesn't need a huge heap.
1730 - Bugfix: let dicts work with NULL content (_foreach didn't) and cleanups
1732 - API Change: gras_os_sleep to take the amount of seconds as a double.
1733 Accepting an int was error prone since it was the only location where
1734 seconds were coded as such. It leaded to damn rounding errors.
1735 - Bugfix: Hard to belive that timers ever worked before this.
1737 -- SimGrid team <simgrid2-users@listes.ens-lyon.fr> Wed, 23 Feb 2005 22:09:21 +0100
1739 SimGrid (2.91) unstable; urgency=low
1741 Alpha 2 on the path to SimGrid 3
1745 - Bug fix in the lmm_solver.
1747 - New! Interface to Paje (see http://www-id.imag.fr/Logiciels/paje/)
1748 through the function MSG_paje_output.
1749 - New! Introducing two new functions MSG_process_kill() and MSG_process_killall().
1750 - It is possible to bound the rate of a communication in MSG with
1751 MSG_task_put_bounded() (was already in the previous version but I had forgotten
1752 to write it in the changelog).
1753 - Bug fix to let GRAS run on top of MSG until we move it directly on top
1758 - Various cleanups to the autotools stuff
1759 - Begin to move Gras examples to examples/gras/
1760 - Let make distcheck work again (yeah!)
1762 - documentation overhauled using doxygen.
1763 gtk-doc-tools is dead in SimGrid now.
1764 - Automatically extract all existing logging categories, and add the list
1765 to the documentation (long standing one, to say the less)
1767 - Cleanup the known architecture table. Reorder the entries to group what
1768 should be, and use a more consistent naming scheme.
1769 (some of the test dataset are still to be regenerated)
1770 - New! Allow library to register globals on each process just as userdata
1772 This is implemented using a xbt_dict and not a xbt_set, so we loose the
1773 lookup time (for now).
1774 Use it in msg and trp.
1775 This cleans a lot the internals and helps enforcing privacy of the
1776 headers between the gras components.
1777 - New! Add a timer mechanism, not unlike cron(8) and at(1).
1778 - Bugfix: gras_os_time was delirious in RL.
1779 - Bugfix: gras_trp_select/RL don't run into the wall when asked to select
1781 - Reenable GRAS now that it works.
1783 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 14 Feb 2005 14:02:13 -0800
1785 SimGrid (2.90) unstable; urgency=low
1787 Alpha 1 on the path to SimGrid 3
1789 * It is a long time since the last release of SimGrid. I'm sorry about
1790 that but as I had told you, I was rewriting a lot of things. I apologize
1791 to those who had been reporting bugs to me and that I had not answered.
1792 If your bug is still in the new version, please tell me. Here is a
1793 summary of the main changes.
1795 * REVOLUTION 1: The SimGrid project has merged with the GRAS project
1796 lead by Martin Quinson. As a consequence SimGrid gains a lot in
1797 portability, speed, and a lot more but you'll figure it out later.
1798 SimGrid now comprises 3 different projects : MSG, GRAS and SMPI.
1799 I wanted to release the new MSG as soon as possible and I have
1800 broken GRAS, which is the reason why, for now, only MSG is fully
1801 functional. A laconic description of these projects is available
1802 in the documentation.
1804 * REVOLUTION 2: I have removed SG and I am now using a new simulation
1805 kernel optimized for our needs (called SURF but only the developers
1806 should use it). Hence, MSG is now roughly 30 times faster and I think
1807 that by rewriting a little bit MSG, I could event speed it up a little
1808 bit more. Beside the gain in speed, it is also much easier to encode a
1809 new platform model with SURF than it was with SG. More to come...
1811 * REVOLUTION 3: I have tried to change a little as possible the API of
1812 MSG but a few things really had to disappear. The main differences
1813 with the previous version are :
1814 1) no more m_links_t and the corresponding functions. Platforms are
1815 directly read from a XML description and cannot be hard-coded
1816 anymore. The same format is used for application deployment
1817 description. The new format is described in the documentation.
1818 Have a look in tools/platform_generation. There is a tiny script
1819 that converts from the old platform format to the new one. Concerning
1820 the application deployment format, parsing the old one is tricky.
1821 I think most of you should however be able to convert your files. If
1822 it is really an issue, I can write a C code that does the conversion.
1824 2) the toolbox tbx does not exist anymore. We now have a library
1825 with much more data-structures but without the hash-tables (we have
1826 dictionaries that are much faster).
1828 -- Arnaud Legrand <Arnaud.Legrand@imag.fr> Mon, 31 Jan 2005 10:45:53 -0800
1830 *****************************************************************************
1831 * Follows the old GRAS changelog. It does not follow the same syntax, but I *
1832 * don't feel like converting the oldies. (Mt) *
1833 *****************************************************************************
1836 Version 2.90: "the long awaited one"
1837 - Finished rewriting and debugging MSG. Rewrote the documentation.
1838 - disable GRAS for now since it needs to be ported to the newest SG
1841 - Finish the port to windows (using mingw32 for cross-compile)
1844 - Main loop and datastructures of SURF. A cpu resource object is
1845 functional. Surf can thus be used to create cpu's with variable
1846 performance on which you can execute some actions.
1848 2004-11-15 Martin Quinson
1849 - Port to ARM. Simply added the alignment and size descriptions. Should
1850 work, but the ARM machines are so slow that I didn't had the opportunity
1851 to 'make check' over there yet.
1853 2004-11-15 Arnaud Legrand
1854 - Trace manager now written. It uses a heap structure and is therefore
1855 expected to be efficient. It may however be speeded up (particularly
1856 when many events occur at the same date) by using red and black
1857 trees. One day maybe...
1858 - Max-min linear system solver written. It uses a sparse matrix
1859 structure taking advantage of its expected use. Most operations are
1860 O(1) and free/calloc are called as few as possible. The computation of
1861 the minimum could however be improved by using a red and black tree
1864 2004-11-03 Arnaud Legrand
1865 - Rename every gras_* function that was in xbt/ to its xbt_
1867 - Add a heap and a doubly-linked list to xbt
1868 - Added a dichotomy to the dictionaries. make check works as well before
1869 so I assume that the patch is correct. I do not know however if things
1870 run effectively faster than before now. :)
1872 Inclusion of the SimGrid tree in the GRAS one. The archive is renamed to
1873 SimGrid, and the version number is bumped to 2.x
1875 2004-10-29 Martin Quinson
1876 - Introduction of the remote errors.
1877 They are the result of a RMI/RPC on the remote machine.
1878 ErrCodes being scalar values, you can't get the host on which those
1879 errors did happen. Extending the error mechanism as in Gnome is possible.
1880 No idea yet whether it is a good idea.
1882 2004-10-28 Martin Quinson
1883 - Interface revolution: the Starred Structure Eradication.
1884 I used to do typedef struct {} toto_t; and then handle *toto_t.
1885 Arnaud (and Oli) didn't like it, and I surrendered. Now, you have:
1886 - ???_t is a valid type (builded with typedef)
1887 - s_toto_t is a structure (access to fields with .)
1888 - s_toto is a structure needing 'struct' keyword to be used
1889 - e_toto_t is an enum
1890 - toto_t is an 'object' (struct*)
1892 typedef struct s_toto {} s_toto_t, *toto_t;
1893 typedef enum {} e_toto_t;
1894 Moreover, only toto_t (and e_toto_t) are public. The rest (mainly
1895 s_toto_t) is private.
1897 - While I was at it, all gras_<obj>_free() functions want a gras_<obj>_t*
1898 so that it can set the variable to NULL. It was so for dicts and sets,
1899 it changed for dynars.
1901 - Fix a bunch of memleaks in dict_remove
1902 - Fix a bug in sg/server_socket opening: it failed all the time.
1904 2004-10-07 Martin Quinson
1905 - Speed up dynar lookup operation a bit.
1907 gras_dynar_get is dead.
1909 Now, you can choose between gras_dynar_get_cpy (the old gras_dynar_get
1910 but should be avoided for efficiency reasons) and gras_dynar_get_ptr
1911 (which gives you the address of the stored data).
1913 gras_dynar_get_as is an helpful macro which allows you to retrieve a
1914 copy of the data using an affectation to do the job and not a memcpy.
1916 int toto = gras_dynar_get_as(dyn,0,int); rewrites itself to
1917 int toto = *(int*)gras_dynar_get_ptr(dyn,0);
1919 It does not really speedup the dynar test because they are
1920 setting elements all the time (and look them seldom). But the dict does
1921 far more lookup than setting.
1923 So, this brings the dict_crash test from ~33s to ~25s (200000 elms).
1925 2004-10-05 Martin Quinson
1926 - Allow to (en/dis)able the cycle detection at run time.
1928 Whether we should check for cycle or not is now a property of each
1929 datatype. When you think there may be some cycle, use datadesc_cycle_set.
1930 datadesc_cycle_unset allow to remove this property when previously set.
1932 Note that the cycle detection is off by default since it impacts the
1933 performance. Watch the data you feed GRAS with ;)
1935 This property is hereditary. Any element embedded in a structure having it
1936 set have it set for the time of this data exchange.
1938 You should set it both on sender and receiver side. If you don't set it on
1939 sender side, it will enter an endless loop. If you forget on receiver
1940 side, the cycles won't be recreated after communication.
1942 - Header reorganization.
1943 Kill gras_private.h, each submodule must load the headers it needs.
1945 2004-10-04 Martin Quinson
1946 - Interface revolution: do not try to survive to malloc failure.
1948 Now, gras_malloc and friends call gras_abort() on failure.
1949 As a conclusion, malloc_error is not a valid error anymore, and all
1950 functions for which it was the only gras_error_t return value are
1951 changed. They now return void, or there result directly.
1952 This simplify the API a lot.
1954 2004-09-29 Martin Quinson
1955 - Re-enable raw sockets.
1956 Created by gras_socket_{client,server}_ext;
1957 Used with gras_raw_{send,recv}
1960 It should allow to kill the last bits of gras first version soon.
1962 This is not completely satisfactory yet (duplicate code with
1963 chunk_{send,recv}; a bit out of the plugin mechanism), but it should
1966 - Simplify transport plugin (internal) interface by not passing any
1967 argument to _server and _client, but embedding them in the socket
1970 2004-09-28 Martin Quinson
1971 - Finish the port to AIX.
1972 autoconf was my problem (segfault within the malloc replacement
1973 function. No idea why)
1975 2004-09-16 Martin Quinson
1976 - Fix some size_t madness on 64bit architectures.
1978 2004-09-08 Martin Quinson
1979 - Reduce the number of system headers loaded, overload some more system
1980 calls (such as malloc to cast the result of the system one, and work
1982 - Fix and reintroduce the config support
1984 2004-09-07 Martin Quinson
1985 - Source code reorganization to allow Arnaud to surf all over there.
1986 - Allow to document the logging categories.
1987 - Remove all uppercase from logging categories and useless cleanup in names.
1989 2004-08-18 Martin Quinson
1990 Version 0.6.2 (protocol not changed; API changed)
1991 - Interface cleanup: gras_msgtype_by_name returns the type (instead of a
1992 gras_error_t), and NULL when not found. Functions expecting a msgtype
1993 as argument (msg_wait; msg_send) deal with NULL argument by providing a
1994 hopefully usefull message.
1995 - Portability to prehistoric sparcs again
1997 2004-08-17 Martin Quinson
1998 Version 0.6.1 (protocol not changed; ABI not changed)
1999 - prealloc some buffers to speed things up
2001 2004-08-11 Martin Quinson
2002 Version 0.6 (protocol not changed; ABI expended)
2003 - The parsing macro can deal with the references, provided that you add
2004 the relevant annotations (using GRAS_ANNOTE(size,field_name))
2006 2004-08-09 Martin Quinson
2007 Version 0.5 (protocol not changed; ABI changed)
2008 - Allow to off turn the cycle detection code in data exchange at
2009 compilation time. It should be at run time, but I'm short of time (and
2010 the config stuff is still broken). That way, we keep dict out of the
2011 critical path, which is good because the performance is poor:
2012 - search not dichotomial yet
2013 - dynar give no way to access their content and memcpy everytime
2014 - In composed data description (struct, ref and so on), stop foolness of
2015 keeping the subtype's ID, but store the type itself. This keeps sets out
2016 of the critical path, which is good since they rely on dynar and
2017 dictionnaries. The only loose of that is that we cannot detect the
2018 redeclaration of a structure/union with another content (but I'm not sure
2019 the code detected well this error before anyway). We still can detect
2020 the redefinition discrepancy for the other types.
2021 - Use a whole bunch of optimisation flags (plus -fno-strict-aliasing since
2022 it breaks the code because of type-punning used all over the place).
2023 This breaks on all non-gcc architectures (for now).
2025 All those changes (plus the buffer of last time) allow me to gain 2 order
2026 of magnitude on cruel tests consisting of 800000 array of integers on two
2027 level of a hierarchical structure (200 secondes -> 4 secondes)
2030 - the selector of reference must now return the type it points to, not
2031 the ID of this type.
2033 2004-08-06 Martin Quinson
2034 Version 0.4 (protocol changed; ABI not changed)
2035 - Allow to pass --gras-log argument to processes in simulation mode. Really.
2036 - New debugging level: trace (under debug) to see effect of GRAS_IN/OUT
2037 - Implement a buffer transport, and use it by default (it relies on tcp in
2038 real life and on sg in simulation).
2039 That's a bit hackish since I had a new field to the structure to store
2040 its data without interfering with the subtype ones. Inheritance
2041 is tricky in C. And that's a kind of reverse inheritance with one class
2042 derivating two classes. Or maybe a game with java interfaces. Anyway,
2043 that's damn hard in C (at least).
2044 Moreover, I got tired while trying to ensure plugin separation and
2045 genericity in SG mode. MSG wants me to do weird things, so let's go for
2046 cruel hacks (temporarily of course ;).
2047 See comment in transport_private.h:71
2048 - do not include all the _interface headers in private but in the files
2049 which really need them (to cut the compilation time when they are
2052 2004-07-26 Martin Quinson
2053 Version 0.3 (protocol not changed; ABI changed)
2054 - Major overhault of the datadesc interface to simplify it:
2055 - shorted the function names:
2056 s/gras_datadesc_declare_struct/gras_datadesc_struct/ and so on
2057 - add a trivial way to push/pop integers into the cbps without malloc.
2058 This allows to make really generic sub_type description, which simply
2059 pop their size of the stack.
2060 - add a function gras_datadesc_ref_pop_arr() which does what users want
2061 most of the time: Declare a dynamic array (which pops its size of the
2062 stack) and declare a reference to it. Poor name, but anyway.
2063 - kill the post-send callback, add a post-receive one
2065 2004-07-23 Martin Quinson
2066 Version 0.2 (protocol changed; ABI changed)
2067 - add some testing for cpbs in the test cases, and fix some more bugs.
2068 This invalidate again the little64 data file, since I cannot regenerate
2070 - remove an awfull optimization in the logging stuff, allowing me to:
2071 - understand it again
2072 - learn gcc how to check that the argument match the provided format
2073 - fix all errors revealed by gcc after that
2074 - internal keys of dict are not \0 terminated. Deal with it properly in
2075 loggings instead of segfaulting when the user want to see the logs :-/
2077 2004-07-22 Martin Quinson
2078 - Fix some stupid bug preventing cbps (callback postit) from working
2080 2004-07-21 Martin Quinson
2081 - Some documentation cleanups
2082 - remove the useless last argument of msgtype_declare
2083 - rename the Virtu functions to fit into the 'os' namespace
2084 - move headers src/include -> src/include/gras/ and stop fooling with
2085 gras -> . symbolic link
2086 - make distcheck is now successful
2088 2004-07-19 Martin Quinson
2090 - Build shared library also
2091 - Install html doc to the right location
2092 - stop removing maintainer files in make clean
2093 - build tests only on make check
2095 2004-07-13 Martin Quinson
2097 - No major issue in previous version => change versionning schema
2098 - Re-enable little64 convertion test now that Abdou kindly regenerated the
2099 corresponding dataset.
2101 2004-07-11 Martin Quinson
2103 - Get it working with any kind of structure (we can compute the padding
2104 bytes remotely for all the architectures I have access to)
2105 - Implement the structure parsing macro (still not quite robust/complete)
2106 - Improvement to the remote testing toysuite
2108 2004-07-10 Martin Quinson
2109 [autoconf mechanism]
2110 - get ride of a bunch of deprecated macros
2111 - actually run the test for two-compliment, not only compile it :-/
2112 - test whether the structures get packed (and bail out if yes. Damn.
2113 Alignment is a serious matter)
2114 - test whether the structures get compacted (but respecting the alignment
2115 constraints of each types)
2116 - test whether the array fields of structures can straddle alignment boundaries
2118 - Damnit, double are bigger than float (typo in creation of 'double' datadesc)
2119 (took me 2 hours to find that bug, looking at the wrong place)
2120 - Add gras_datadesc_declare_{union,struct}_close(). They must be used
2121 before sending/receiving and are used to compute the offsets of fields
2122 - Given that padding size depend even on compiler options, keep track of
2123 alignment and aligned_size only for the current architecture. Not a big
2124 deal since we send structure fields one after the other (seems
2126 - Add the datastructure used for IEEE paper by the PBIO guys to the test
2127 program, let it work on linux/gcc/little32. portability todo.
2129 2004-07-08 Martin Quinson
2130 - import and improve remote compilation support from FAST
2131 - make sure make check works on half a dozen of machines out there
2133 2004-07-07 Martin Quinson
2134 Let's say it's version 0.0.3 ;)
2135 - Implement conversions (yuhu!)
2136 - Let it work on solaris (beside conversion, of course)
2137 - Stupid me, using rand() to generate the conversion datatests in not wise.
2139 2004-07-06 Martin Quinson
2140 - Let make dist work, since I'm gonna need it to compile on remote hosts
2141 - Let Tests/datadesc_usage write the architecture on which the file was
2142 generated as first byte.
2143 - Add PowerPC (being also IRIX64), SPARC (also power4) and ALPHA
2144 architecture descriptions.
2145 - Add datadesc_usage.{i386,ppc,sparc} files being the result of execution
2146 on those architectures.
2147 - Optimization: send/recv array of scalar in one shoot
2149 2004-07-05 Martin Quinson
2150 - YEAH! GRAS/SG and GRAS/RL are both able to run the ping example !
2152 - Plug a whole bunch of memleaks
2153 - each process now have to call gras_{init,exit}. One day, their log
2154 settings will be separated
2155 - Continue the code factorisation between SG, RL and common in Transport.
2157 2004-07-04 Martin Quinson
2159 - Redistribution between SG and RL.
2160 We wanna have to accept in SG, so move accepted related parts of RL in
2161 the common part. (more precisely, the dynar of all known sockets is no
2162 more a static in transport.c, but part of the process_data)
2164 [gras_stub_generator]
2165 - Bug fix: Do call gras_process_init from gras_init (wasnt called in RL).
2167 2004-07-03 Martin Quinson
2168 - Create a new log channel tbx containing dict, set, log, dynar (to shut
2169 them all up in one shot)
2171 - Fix the ugly case of reference to dynamic array.
2172 - New (semi-public) function gras_datadesc_size to allow the messaging
2173 layer to malloc the needed space for the buffer.
2175 - gras_socket_close now expect the socket to close (and not its address to
2176 put NULL in it after it). This is because the socket passed to handlers
2177 is one of their argument (=> not writable).
2179 - propagate the interface cleanup from last week in datadesc, ie remove a
2180 superfluous level of indirection. User pass adress of variable
2181 containing data (both when sending and receiving), and not of a variable
2182 being a pointer to the data. Let's say that I like it better ;)
2183 The price for that is constructs like "int msg=*(int*)payload" in
2184 handlers, but it's a fine price, IMHO.
2186 - Let it work in RL (yuhu)
2188 2004-06-21 Martin Quinson
2190 - porting SG plugin and SG select to new standards (works almost).
2191 - plug memleaks and fix bugs around.
2194 - cleanup the prototype of data recv and force users to specify when they
2195 want to handle references to objects. Test case working even for cycles.
2196 - plug memleaks. Valgrind is perfectly ok with this.
2198 2004-06-12 Martin Quinson
2200 - cleanup the separation between plugin and main code in plugin creation
2202 2004-06-11 Martin Quinson
2204 - Reput hook for raw sockets, needed for BW experiments
2205 - kill a few lines of dead code
2206 [Data description] Interface cleanup
2207 - gras_datadesc_by_name returns the searched type or NULL.
2208 That way, no variable is needed to use a type desc once, which makes
2210 - gras_datadesc_declare_[struct|union]_append_name is removed. The last
2211 two parameters were strings (field name, type name), leading to
2213 [Dicos] Interface cleanup
2214 - gras_dico_retrieve -> gras_dico_get ; gras_dico_insert -> gras_dico_set
2215 This is consistant with the dynar API.
2217 2004-04-21 Martin Quinson
2219 - Porting to new standards.
2221 - interface cleanup.
2222 There is no bag anymore, no need to take extra provision to mask the
2223 pointers behind "ID".
2224 Better splitup of functions between files create/exchange/convert.
2225 This is still a bit artificial since convert and receive are so
2226 interleaved, but anyway.
2228 - add a queued message list to procdata (the ones not matching criteria
2230 - factorize some more code between SG and RL wrt procdata
2232 - use gras_exit in example to track memleaks
2233 - get rid of gs_example now that GS is properly integrated into gras
2234 - update run_test to integrate the lastest tests (datadesc)
2236 - rename WARNINGn macros to WARNn since it prooved error-prone
2238 2004-04-19 Martin Quinson
2240 - register init/exit functions within gras module mechanism
2241 - send/receive function.
2242 Convertion is not implemented, but short-cutted if not needed.
2243 struct/array elements are sent one by one (instead of block-wise), but
2244 nobody really cares (yet). Get a prototype before optimizing.
2245 - tests (using a file socket) for DD send/receive on:
2246 - base types: int, float
2247 - array: fixed size, string (ie ref to dynamic string)
2248 - structure: homogeneous, heterogeneous
2249 - chained list, graph with cycle
2250 Believe it or not, valgrind is not too unhappy with the results. The
2251 cycle happily segfaults, but the others are ok. And I'm sick of pointers
2255 - Bugfix when using a filename explicitely (instead of '-')
2257 2004-04-09 Martin Quinson
2259 - factorize more code between RL and SG in socket creation
2260 - Complete the implementation and tests of:
2262 o file (only in RL, and mainly for debugging)
2264 I lost 3 days to design a portable address resolver, and then decided
2265 that the prototype mainly have to run on my box.
2266 Addressing portability too early may be like optimizing too early :-/
2268 - use gras_init in the Tests instead of the crappy parse_log_opt
2269 (the latter function is removed)
2270 [Conditional execution]
2271 - New functions: gras_if_RL/gras_if_SG (basic support for this)
2272 [Code reorganisation]
2273 - Get rid of libgrasutils.a since it makes more trouble than it solves.
2274 Build examples against the RL library, since there is no way to disable
2275 its creation for now.
2277 For information, the beginning of coding on GRAS was back in june
2278 2003. I guess that every line has been rewritten at least twice since