(sometimes after the merge with the ORTE branch), the opal_pointer_array
will became the only pointer_array implementation (the orte_pointer_array
will be removed).
This commit was SVN r17007.
The commit has been tested for C/R and Cray operations, and on Odin (SLURM, rsh) and RoadRunner (TM). I tried to update all environments, but obviously could not test them. I know that Windows needs some work, and have highlighted what is know to be needed in the odls process component.
This represents a lot of work by Brian, Tim P, Josh, and myself, with much advice from Jeff and others. For posterity, I have appended a copy of the email describing the work that was done:
As we have repeatedly noted, the modex operation in MPI_Init is the single greatest consumer of time during startup. To-date, we have executed that operation as an ORTE stage gate that held the process until a startup message containing all required modex (and OOB contact info - see #3 below) info could be sent to it. Each process would send its data to the HNP's registry, which assembled and sent the message when all processes had reported in.
In addition, ORTE had taken responsibility for monitoring process status as it progressed through a series of "stage gates". The process reported its status at each gate, and ORTE would then send a "release" message once all procs had reported in.
The incoming changes revamp these procedures in three ways:
1. eliminating the ORTE stage gate system and cleanly delineating responsibility between the OMPI and ORTE layers for MPI init/finalize. The modex stage gate (STG1) has been replaced by a collective operation in the modex itself that performs an allgather on the required modex info. The allgather is implemented using the orte_grpcomm framework since the BTL's are not active at that point. At the moment, the grpcomm framework only has a "basic" component analogous to OMPI's "basic" coll framework - I would recommend that the MPI team create additional, more advanced components to improve performance of this step.
The other stage gates have been replaced by orte_grpcomm barrier functions. We tried to use MPI barriers instead (since the BTL's are active at that point), but - as we discussed on the telecon - these are not currently true barriers so the job would hang when we fell through while messages were still in process. Note that the grpcomm barrier doesn't actually resolve that problem, but Brian has pointed out that we are unlikely to ever see it violated. Again, you might want to spend a little time on an advanced barrier algorithm as the one in "basic" is very simplistic.
Summarizing this change: ORTE no longer tracks process state nor has direct responsibility for synchronizing jobs. This is now done via collective operations within the MPI layer, albeit using ORTE collective communication services. I -strongly- urge the MPI team to implement advanced collective algorithms to improve the performance of this critical procedure.
2. reducing the volume of data exchanged during modex. Data in the modex consisted of the process name, the name of the node where that process is located (expressed as a string), plus a string representation of all contact info. The nodename was required in order for the modex to determine if the process was local or not - in addition, some people like to have it to print pretty error messages when a connection failed.
The size of this data has been reduced in three ways:
(a) reducing the size of the process name itself. The process name consisted of two 32-bit fields for the jobid and vpid. This is far larger than any current system, or system likely to exist in the near future, can support. Accordingly, the default size of these fields has been reduced to 16-bits, which means you can have 32k procs in each of 32k jobs. Since the daemons must have a vpid, and we require one daemon/node, this also restricts the default configuration to 32k nodes.
To support any future "mega-clusters", a configuration option --enable-jumbo-apps has been added. This option increases the jobid and vpid field sizes to 32-bits. Someday, if necessary, someone can add yet another option to increase them to 64-bits, I suppose.
(b) replacing the string nodename with an integer nodeid. Since we have one daemon/node, the nodeid corresponds to the local daemon's vpid. This replaces an often lengthy string with only 2 (or at most 4) bytes, a substantial reduction.
(c) when the mca param requesting that nodenames be sent to support pretty error messages, a second mca param is now used to request FQDN - otherwise, the domain name is stripped (by default) from the message to save space. If someone wants to combine those into a single param somehow (perhaps with an argument?), they are welcome to do so - I didn't want to alter what people are already using.
While these may seem like small savings, they actually amount to a significant impact when aggregated across the entire modex operation. Since every proc must receive the modex data regardless of the collective used to send it, just reducing the size of the process name removes nearly 400MBytes of communication from a 32k proc job (admittedly, much of this comm may occur in parallel). So it does add up pretty quickly.
3. routing RML messages to reduce connections. The default messaging system remains point-to-point - i.e., each proc opens a socket to every proc it communicates with and sends its messages directly. A new option uses the orteds as routers - i.e., each proc only opens a single socket to its local orted. All messages are sent from the proc to the orted, which forwards the message to the orted on the node where the intended recipient proc is located - that orted then forwards the message to its local proc (the recipient). This greatly reduces the connection storm we have encountered during startup.
It also has the benefit of removing the sharing of every proc's OOB contact with every other proc. The orted routing tables are populated during launch since every orted gets a map of where every proc is being placed. Each proc, therefore, only needs to know the contact info for its local daemon, which is passed in via the environment when the proc is fork/exec'd by the daemon. This alone removes ~50 bytes/process of communication that was in the current STG1 startup message - so for our 32k proc job, this saves us roughly 32k*50 = 1.6MBytes sent to 32k procs = 51GBytes of messaging.
Note that you can use the new routing method by specifying -mca routed tree - if you so desire. This mode will become the default at some point in the future.
There are a few minor additional changes in the commit that I'll just note in passing:
* propagation of command line mca params to the orteds - fixes ticket #1073. See note there for details.
* requiring of "finalize" prior to "exit" for MPI procs - fixes ticket #1144. See note there for details.
* cleanup of some stale header files
This commit was SVN r16364.
- If one wants to use this solution, remember to unload the project 'orte-restart' which is currently not working for Windows.
This commit was SVN r15680.
calls MPI_INIT and that causes problems during make distcheck. Instead
put it in check_PROGRAMS which lets it get built, but doesn't run it.
This commit was SVN r14832.
The following SVN revision numbers were found above:
r14831 --> open-mpi/ompi@9258c5200a
* Require Autoconf 2.60 or higher and remove some cruft
required for AC 2.59 or the AC 2.59 / AC 2.60 mix
* Remove a bunch of now unnecessary AC_SUBST calls
* Use the libtool-provided variables for the -I and
library to use when compiling against ltdl
Fixes trac:1000
This commit was SVN r14652.
The following Trac tickets were found above:
Ticket 1000 --> https://svn.open-mpi.org/trac/ompi/ticket/1000
Without this a compiler could assume that the counter is not updated
my the malloc call and remove the test in the assert and always trigger
the assertion.
This commit was SVN r12419.
Other changes:
1. Remove the old xcpu components as they are not functional.
2. Fix a "bug" in orterun whereby we called dump_aborted_procs even when we normally terminated. There is still some kind of bug in this procedure, however, as we appear to be calling the orterun job_state_callback function every time a process terminates (instead of only once when they have all terminated). I'll continue digging into that one.
This will require an autogen/configure, I'm afraid.
This commit was SVN r11228.
After seeing the uglyness that is removing directories in the
codebase I decided to push down this to the OPAL by extending the
opal/os_create_dirpath.(c|h) to contain some more functionality.
In this process I renamed 'os_create_dirpath' to 'os_dirpath' since it
is a bit more general now.
Added a few functions to:
- check if an directory is empty
- check to see if the access permissions are set correctly
- destroy the directory at the end of the dirpath
- By using a caller callback function (a la Perl, I believe)
for every file, the caller can have fine grained control over
whether a specific file is deleted or not.
This simplifies things a bit for orte_session_dir_(finalize|cleanup)
as it should no longer contain any of this functionality, but uses
these functions to do the work.
From the external perspective nothing has changed, from the
developer point of view we have some cleaner, more generic code.
This commit was SVN r10640.
to compute the overhead of the convertor (and all convertor related operations).
The second will check the position setting on the convertor. Not yet completed ...
This commit was SVN r10432.
done with the same values on 2 different types return the same value. The 2
types belong to 2 differents classes: contiguous and sparse. With this test
I simulate the behavior of the buffered send, where the sender send the
data from the user attached buffer (which is contiguous) and the receiver
receive it in a sparse type.
This commit was SVN r10372.
actual files, so we should not have a clean rule for them - instead,
make it maintainer-clean. Neither clean nor distclean should
remove files that were in the tarball...
This commit was SVN r9351.
- move files out of toplevel include/ and etc/, moving it into the
sub-projects
- rather than including config headers with <project>/include,
have them as <project>
- require all headers to be included with a project prefix, with
the exception of the config headers ({opal,orte,ompi}_config.h
mpi.h, and mpif.h)
This commit was SVN r8985.
both mmap and munmap), adjusting the configure script so that the
component will only be activated on systems that use ptmalloc2 in the
first place -- ie, Linux
* Remove the malloc_hooks component - it became an unworkable solution
once threads and such were considered.
* Remove malloc_interpose component - it never worked quite right and
was not going to be able to intercept malloc, so it wasn't going to
be useful for OMPI's purposes.
* Update tests a little bit to match recent memory hooks api
issues - still needs a bit of work.
This commit was SVN r8381.
When compiling C++ code that includes something that looks for the C++
header file "memory" (stupid C++ headers not having .h extensions), it
goes through the header file search path, which includes $(topsrcdir)/opal,
so it finds the directory $(topsrcdir)/opal/memory/ and tries to load
that as the memory header file and all goes downhill.
This commit was SVN r8111.
AM_INIT_AUTOMAKE, instead of the deprecated version.
* Work around dumbness in modern AC_INIT that requires the version
number to be set at autoconf time (instead of at configure time, as
it was before). Set the version number, minus the subversion r number,
at autoconf time. Override the internal variables to include the r
number (if needed) at configure time. Basically, the right thing
should always happen. The only place it might not is the version
reported as part of configure --help will not have an r number.
* Since AM_INIT_AUTOMAKE taks a list of options, no need to specify
them in all the Makefile.am files.
* Addes support for subdir-objects, meaning that object files are put
in the directory containing source files, even if the Makefile.am is
in another directory. This should start making it feasible to
reduce the number of Makefile.am files we have in the tree, which
will greatly reduce the time to run autogen and configure.
This commit was SVN r7211.
1. Valgrind is good for something - chasing down memory leaks in registry led me to re-visit the dictionary functions and discover that I wasn't keeping track of the number of dictionary entries on each segment! Resulted in wasted time searching blank entries as well as leaked memory. This has now been fixed.
2. Fixed the orte_bitmap test. The init function for that class has been eliminated and the constructor adjusted to provide that functionality.
This commit was SVN r7136.
add a -I to find the included ltdl.h (vs. a system-installed ltdl.h)
- Clean up kruft in a bunch of Makefile.am's to remove now-unnecessary
AM_CPPFLAGS settings to get static-components.h for each framework
- Move the component_repository API functions out of opal/mca/base/base.h
and into opal/mca/base/mca_base_component_repository.h in order to
decrease unnecessary dependencies (e.g., before this, almost
everything in the tree depended on ltdl.h, which is unnecessary --
only a small number of files really need ltdl.h)
This commit was SVN r7127.
orte_init_stage1(), since not all ORTE processes call orte_init().
* Expad opal_error test case to make sure ORTE error codes print
properly
* Make project error codes start at easy values (OPAL is -1 to -100,
ORTE is -101 to -200, OMPI is less than -201) to make it easier
to figure out what an error code as an integer means. Also has
the nice property of not changing the values of error codes ever
time a new error code is added.
This commit was SVN r7061.
tree.
- fix up #include's throughout the tree (yay contrib/search_replace.pl!)
- remove a few extraneous #include's
- remove orte_sys_info*() from opal_init()/opal_finalize() (it's
already in orte_init_stage1() and orte_system_finalize())
- remove dependencies in opal on orte_system_info -- util/os_path.c
and util/os_create_dirpath.c (they only used path_sep, anyway --
easily changed to #defines)
This commit was SVN r7059.
cast the return to an int in the C++ test case, just in case.
* C++ sucks. If compiling with C++ on some GNU compiler/linker
combos, the initialize hook isn't automagically fired for the
malloc code. Add a backup setting during opal_init, which is
early enough not to cause any damage.
This commit was SVN r6983.
OPAL_ERROR, same for all the other error codes. Also, make sure that there
are never conflicts between OPAL anr ORTE error codes (for example).
Finally, provide opal_perror(), opal_strerror(), and opal_strerror_r() to
give stringified error messages for the different error codes
This commit was SVN r6969.
to opal_progress() to use the timers instead of a tick count for deciding
whether to call the event loop or not. Currently supported platforms are:
- solaris (x86 / sparc)
- Linux (x86 / x86_64 / IA64)
- Mac OS X (x86 / Power PC)
This commit was SVN r6922.
* Add memory intercept routines for Darwin using the official Darwin
API (thanks to Drew Gallatin from Myricom for pointing me to some
information from Apple engineers about how to make this work)
* add debugging output to functionality test
This commit was SVN r6920.
callbacks to be triggered when memory is about to leave the current
process. The system is designed to allow a variety of interfaces,
hopefully including whole-sale replacement of the memory manager,
ld preload tricks, and hooks into the system memory manager. Since
some of these may or may not be available at runtime and we won't know
until runtime, there is a query funtion to look for availability of
such a setup.
* Added ptmalloc2 memory manager replacement code. Not turned on by
default, can be enabled with --with-memory-manager=ptmalloc2.
Only tested on Linux, not even compiled elsewhere. Do not use
on OS X, or you will never see your process again.
* Added AM_CONDITIONAL for threads test to support ptmalloc2's build
system
This commit was SVN r6790.
I'll send out a general note about this in the morning, but for now I'll just notify people through this note that the new simplified "put" commands have been debugged and work just fine. I'll add documentation to the gpr.h file later - only think to really be aware of is that the tokens array must be NULL terminated. Other than that, things work pretty much as you'd expect.
This commit was SVN r6700.
- Update svn:ignore's to match new exectuable names
- Consolidate the unit test Makefile.am flags into a testing
Makefile.options
- Remove a bunch of SUBDIRS from test/mca/Makefile so that they don't
run by default, but can be invoked manually (they're still in
DIST_SUBDIRS)
This commit was SVN r6598.
* rename ompi_malloc to opal_malloc
* rename ompi_numtostr to opal_numtostr
* start of rename of ompi_environ to opal_environ
This commit was SVN r6332.
* rename ompi_basename to opal_basename
* rename ompi bitop functions to opal
* rename ompi_cmd_line to opal_cmd_line
* rename ompi_sizet2int to opal_sizet2int
* rename orte_daemon_init to opal_daemon_init
* rename ompi_few to opal_few
This commit was SVN r6330.
unit tests without screwing up the nightly builds.
These changes fix the problem of not including the test/mca/gpr
directory in the nightly tarball, prevent the tests from being
compiled, but leave the door open for manual compilation when the time
comes to start the work to re-enable them (e.g., uncomment a few
lines in gpr/Makefile.am).
This commit was SVN r6175.
Also included is a fix to the attribute problem for singletons.
Short explanation:
The prior system placed triggers and subscriptions on the registry for each process - approximately eight/process. Each of these had to be checked every time there was a registry operation such as a "put" or "increment-value". For large numbers of processes, this repetitive checking consumed some significant time.
The new system allows processes to "attach" to existing triggers and subscriptions, without creating a new one. Thus, there are now only eight triggers and five subscriptions on a job - *regardless of how many processes are being run*. This means that the registry now takes the same amount of time (which is pretty darn short) to process an operation regardless of how many processes are in a job.
I'll provide some startup times from scalability tests shortly - need to complete the commit so I can move the system to an appropriate cluster.
This commit was SVN r6164.
the RDS selection logic, which is, unfortunately, not yet well
supported by the testing infrastructure (it causes false failures in
the nightly build).
This commit was SVN r6073.
