Many of these tests were failing due to opal_init() failing in some
cases (because the opal shmem framework needs installed components, so
"make distcheck" would fail these tests because the opal shmem
components were not installed). However, all of these tests seem to
be fine with opal_init_util() -- so let's re-enable these tests.
Now that "make check" siphons off stdout/stderr to logfiles, it's ok
to have output by default from tests. This test fails often enough
that it's useful to see the diagnostic output.
This commit adds a new class: opal_fifo.h. The new class has atomic, non-atomic,
and opal_using_threads() conditoned routines. It should be used when first-in
first-out is required and should perform much better than using locks and an
opal_list_t. Like with opal_lifo_t there are two versions of the atomic
implementation: 128-bit compare-and-swap, and spin-locked. More implementations
can be added later (LL/SC comes to mind).
This commit also adds a unit test for the opal_fifo_t class. This test verifies
the fifo implementation when using multiple threads.
- Rename opal_atomic_lifo_t to opal_lifo_t to reflect both atomic and
non-atomic usage. Added new routines (opal_lifo_*_st) for non-atomic
usage as well as routines conditioned off opal_using_threads(). The
atomic versions are always thread safe and the non-atomic are always
not thread safe.
- Add a new atomic lifo implementation that makes use of 128-bit
compare-and-swap. The new implementation should scale better with
larger numbers of threads.
- Add threading unit test for opal_lifo_t.
We recognize that this means other users of OPAL will need to "wrap" the opal_process_name_t if they desire to abstract it in some fashion. This is regrettable, and we are looking at possible alternatives that might mitigate that requirement. Meantime, however, we have to put the needs of the OMPI community first, and are taking this step to restore hetero and SPARC support.
the OPAL and ORTE libraries. This is required by projects such as ORCM
that have their own ORTE and OPAL libraries in order to avoid library
confusion. By renaming their version of the libraries, the OMPI
applications can correctly dynamically load the correct one for their
build."
This reverts commit 63f619f871.
1. It's actually hashing now, whereas the old OPAL hash table was not. Thus, it is a bug fix for and, as such, should be included in the 1.8 series.
2. It is dynamic and can grow and shrink the number of buckets in accordance with job size, whereas the old OPAL hash table had a fixed number of buckets which resulted in poor retrieval performance at large scale.
This scheme has been deployed in the field on very large H.P./Mellanox systems and has been demonstrated to significantly decrease job start-up time (~ 20% improvement) when launching applications directly with srun in SLURM environments. However, neither SLURM nor direct launch are prerequisites to take advantage of this change as any entity that utilizes OPAL hash table objects can benefit (at least partially) from this contribution.
Linux statfs(2) lies about the type of fuse filesystems (it reports
fuse.encfs as an NFS filesystem). So just skip fuse filesystems in
this test until/if we ever care to add some kind of workaround.
Refs trac:4767
cmr=v1.8.2:reviewer=rhc
This commit was SVN r32152.
The following Trac tickets were found above:
Ticket 4767 --> https://svn.open-mpi.org/trac/ompi/ticket/4767
* Resolve set-but-not-used issues
* Resolve incorrect const notation (I checked with George first to see
what const notation he actually wanted)
* Comment out unused code (didn't delete it because it's useful
debugging code)
* Resolve int<-->void* casting
* Resolved signed / unsigned comparisons
This commit was SVN r30225.
On Linux, if this test is run with no command line params, it will run
"mount" and analyze the output (same as it always has).
On all platforms, if you provide one or more command line options,
each command line option is given to opal_path_nfs() and the result is
sent to stdout.
This commit was SVN r30208.
value to signal that the operation of retrieving the element from the free list
failed. However in this case the returned pointer was set to NULL as well, so the
error code was redundant. Moreover, this was a continuous source of warnings when
the picky mode is on.
The attached parch remove the rc argument from the OMPI_FREE_LIST_GET and
OMPI_FREE_LIST_WAIT macros, and change to check if the item is NULL instead of
using the return code.
This commit was SVN r28722.
When building Open MPI with CMake, a VS solution will be generated automatically, this solution can be directly used.
For the installer, it's a bit tricky, need to do more in NSIS config codes, in order to make the solution file aware the installation directory of user.
This commit was SVN r26616.
Roll in the ORTE state machine. Remove last traces of opal_sos. Remove UTK epoch code.
Please see the various emails about the state machine change for details. I'll send something out later with more info on the new arch.
This commit was SVN r26242.
zeroes);
if so, use it for bit-operations like opal_cube_dim and opal_hibit.
Implement two versions of power-of-two.
In case of opal_next_poweroftwo, this reduces the average execution
time from 83 cycles to 4 cycles (Intel Nehalem, icc, -O2, inlining,
measured rdtsc, with loop over 2^27 values).
Numbers for other functions are similar (but of course heavily depend
on the usage, e.g. opal_hibit() with a start of 4 does not save
much). The bsr instruction on AMD Opteron is also not as fast.
- Replace various places where the next power-of-two is computed.
Tested on Intel Nehalem Cluster with openib, compilers GNU-4.6.1 and
Intel-12.0.4 using mpi_testsuite -t "Collective" with 128 processes.
This commit was SVN r25270.
To enable the epochs and the resilient orte code, use the configure flag:
--enable-resilient-orte
This will define both:
ORTE_ENABLE_EPOCH
ORTE_RESIL_ORTE
This commit was SVN r25093.
No need for any CMRs to 1.5... that was already done in CMR 2728.
This commit was SVN r24545.
The following SVN revision numbers were found above:
r22841 --> open-mpi/ompi@b400b84162
There was no compelling reason to support such old kernels. Accordingly, convert the test to print a nice error message indicating we no longer support old kernels (but indicate that earlier OMPI versions do) and error out. Remove all code that was protected by "if have different pids" since it can no longer be compiled.
This commit was SVN r24531.
Setup the event API to support multiple bases in preparation for splitting the OMPI and ORTE events. Holding here pending shared memory resolution.
This commit was SVN r23943.
This is a fairly intrusive change, but outside of the moving of opal/event to opal/mca/event, the only changes involved (a) changing all calls to opal_event functions to reflect the new framework instead, and (b) ensuring that all opal_event_t objects are properly constructed since they are now true opal_objects.
Note: Shiqing has just returned from vacation and has not yet had a chance to complete the Windows integration. Thus, this commit almost certainly breaks Windows support on the trunk. However, I want this to have a chance to soak for as long as possible before I become less available a week from today (going to be at a class for 5 days, and thus will only be sparingly available) so we can find and fix any problems.
Biggest change is moving the libevent code from opal/event to a new opal/mca/event framework. This was done to make it much easier to update libevent in the future. New versions can be inserted as a new component and tested in parallel with the current version until validated, then we can remove the earlier version if we so choose. This is a statically built framework ala installdirs, so only one component will build at a time. There is no selection logic - the sole compiled component simply loads its function pointers into the opal_event struct.
I have gone thru the code base and converted all the libevent calls I could find. However, I cannot compile nor test every environment. It is therefore quite likely that errors remain in the system. Please keep an eye open for two things:
1. compile-time errors: these will be obvious as calls to the old functions (e.g., opal_evtimer_new) must be replaced by the new framework APIs (e.g., opal_event.evtimer_new)
2. run-time errors: these will likely show up as segfaults due to missing constructors on opal_event_t objects. It appears that it became a typical practice for people to "init" an opal_event_t by simply using memset to zero it out. This will no longer work - you must either OBJ_NEW or OBJ_CONSTRUCT an opal_event_t. I tried to catch these cases, but may have missed some. Believe me, you'll know when you hit it.
There is also the issue of the new libevent "no recursion" behavior. As I described on a recent email, we will have to discuss this and figure out what, if anything, we need to do.
This commit was SVN r23925.
check" (it's been deactivated for 2+ years now, anyway). It needs to
be launched via "mpirun" and needs >= 2 processes, so it wasn't a good
candidate for "make check", anyway.
The test itself has moved to OMPI's internal testing suites.
This commit was SVN r23526.
tests, not OMPI tests.
This allows us to "make distcheck" with "./autogen.sh -no-ompi"
trees (i.e., these tests will now still work even if the OMPI layer is
not present -- because they're OPAL tests and we should treat them
that way).
This commit was SVN r23524.
test, not an OMPI test.
Also fix a case where if you haven't run "make install", then
opal_show_help_string() will (rightfully) return NULL. So be sure to
handle that case and not segv.
This commit was SVN r23522.
longer necessary to link in libopen-rte if you link in libmpi (for
example) because of the fact that libmpi now completely slurps in
libopen-rte (ditto with libopen-rte and libopen-pal).
* Only build ompi_rb_tree if we have the OMPI layer.
This commit was SVN r23521.
#define CACHE_LINE_SIZE to 128. This name has a conflict on NetBSD,
and it seems kinda odd to have a header file that ''only'' defines a
single value. Also, we'll soon be raising hwloc to be a first-class
item, so having this file around seemed kinda weird.
Therefore, I replaced CACHE_LINE_SIZE with opal_cache_line_size, an
int (in opal/runtime/opal_init.c and opal/runtime/opal.h) on the
rationale that we can fill this in at runtime with hwloc info (trunk
and v1.5/beyond, only). The only place we ''needed'' a compile-time
CACHE_LINE_SIZE was in the BTL SM (for struct padding), so I made a
new BTL_SM_ preprocessor macro with the old CACHE_LINE_SIZE value
(128). That use isn't suitable for run-time hwloc information,
anyway.
This commit was SVN r23349.
(OMPI_ERR_* = OPAL_SOS_GET_ERR_CODE(ret)), since the return value could be a
SOS-encoded error. The OPAL_SOS_GET_ERR_CODE() takes in a SOS error and returns
back the native error code.
* Since OPAL_SUCCESS is preserved by SOS, also change all calls of the form
(OPAL_ERROR == ret) to (OPAL_SUCCESS != ret). We thus avoid having to
decode 'ret' to get the native error code.
This commit was SVN r23162.
The OPAL SOS framework tries to meet the following objectives:
* reduce the cascading error messages and the amount of code needed to print an error message.
* build and aggregate stacks of encountered errors and associate related individual errors with each other.
* allow registration of custom callbacks to intercept error events.
For more information, refer to
https://svn.open-mpi.org/trac/ompi/wiki/ErrorMessages
This commit was SVN r23158.
tests ;-)) Turn them off again, didn't have time to look into them
Also, the test-program on eddie.osl.iu.edu, detects the rpc_pipefs
mounted on /var/lib/nfs/rpc_pipefs, required for NFS.
This commit was SVN r22607.
If file does not exist, check the directory it lives in...
Maybe used by caller, trying to open mmap() on NFS, Lustre or
Panasas (thanks Sam).
For now, this is used to warn about the usage of mmap on such FS.
Please note, that Ralph mentioned the orte_no_session_dir parameter.
The help message includes a reference to this.
Tested on NFS and Lustre on Linux on
smoky: mpirun --mca orte_tmpdir_base $HOME/tmp -np 2 ./mpi_stub
jaguar: mpirun ... --mca orte_tmpdir_base /tmp/work/$USER ...
Fixes trac:1354
This should cmr:v1.5 once it has soaked and is shown to work on
Solaris
This commit was SVN r22604.
The following Trac tickets were found above:
Ticket 1354 --> https://svn.open-mpi.org/trac/ompi/ticket/1354
In CMake 2.6 and earlier, this function add dependencies for targets and also link the target libraries automatically, but in CMake 2.8,this behavior has been changed, i.e. it will only add the dependencies but no link, which will cause linking errors at compilation time.
This commit was SVN r22405.
Re-enable "./autogen.sh -no-ompi" again. If you -no-ompi, the entire OMPI
configury is skipped and the entire ompi/ subtree is not built. There's
some simple m4-isms that prune out the relevant parts.
I added ompi/config/, orte/config/, and opal/config/ directories. I moved a
bunch of m4 files from the top-level config/ dir into ompi/config/, and a few
into orte/config/.
Note that all 3 <project>/config directories have a config_files.m4 file. This
file contains the AC_CONFIG_FILES list for that project. The AC_CONFIG_FILES
call cannot be in an AC_DEFUN macro and conditionally called -- if it is
included at all, Autoconf will process it. Hence, these config_files.m4 files
don't AC_DEFUN -- they just have AC_CONFIG_FILES. m4_ifdef() is used to
conditionally include the files or not.
I moved a bunch of obvious OMPI-only m4 files from config/ to ompi/config/,
but I'm sure that there's more that could go. A ticket will be filed with
thoughts on future work in this area.
This commit was SVN r22113.
#if defined (c_plusplus)
defined (__cplusplus)
followed by
extern "C" {
and the closing counterpart by BEGIN_C_DECLS and END_C_DECLS.
Notable exceptions are:
- opal/include/opal_config_bottom.h:
This is our generated code, that itself defines BEGIN_C_DECL and
END_C_DECL
- ompi/mpi/cxx/mpicxx.h:
Here we do not include opal_config_bottom.h:
- Belongs to external code:
opal/mca/backtrace/darwin/MoreBacktrace/MoreDebugging/MoreBacktrace.c
opal/mca/backtrace/darwin/MoreBacktrace/MoreDebugging/MoreBacktrace.h
- opal/include/opal/prefetch.h:
Has C++ specific macros that are protected:
- Had #if ... } #endif _and_ END_C_DECLS (aka end up with 2x
END_C_DECLS)
ompi/mca/btl/openib/btl_openib.h
- opal/event/event.h has #ifdef __cplusplus as BEGIN_C_DECLS...
- opal/win32/ompi_process.h: had extern "C"\n {...
opal/win32/ompi_process.h: dito
- ompi/mca/btl/pcie/btl_pcie_lex.l: needed to add *_C_DECLS
ompi/mpi/f90/test/align_c.c: dito
- ompi/debuggers/msgq_interface.h: used #ifdef __cplusplus
- ompi/mpi/f90/xml/common-C.xsl: Amend
Tested on linux using --with-openib and --with-mx
The following do not contain either opal_config.h, orte_config.h or
ompi_config.h
(but possibly other header files, that include one of the above):
ompi/mca/bml/r2/bml_r2_ft.h
ompi/mca/btl/gm/btl_gm_endpoint.h
ompi/mca/btl/gm/btl_gm_proc.h
ompi/mca/btl/mx/btl_mx_endpoint.h
ompi/mca/btl/ofud/btl_ofud_endpoint.h
ompi/mca/btl/ofud/btl_ofud_frag.h
ompi/mca/btl/ofud/btl_ofud_proc.h
ompi/mca/btl/openib/btl_openib_mca.h
ompi/mca/btl/portals/btl_portals_endpoint.h
ompi/mca/btl/portals/btl_portals_frag.h
ompi/mca/btl/sctp/btl_sctp_endpoint.h
ompi/mca/btl/sctp/btl_sctp_proc.h
ompi/mca/btl/tcp/btl_tcp_endpoint.h
ompi/mca/btl/tcp/btl_tcp_ft.h
ompi/mca/btl/tcp/btl_tcp_proc.h
ompi/mca/btl/template/btl_template_endpoint.h
ompi/mca/btl/template/btl_template_proc.h
ompi/mca/btl/udapl/btl_udapl_eager_rdma.h
ompi/mca/btl/udapl/btl_udapl_endpoint.h
ompi/mca/btl/udapl/btl_udapl_mca.h
ompi/mca/btl/udapl/btl_udapl_proc.h
ompi/mca/mtl/mx/mtl_mx_endpoint.h
ompi/mca/mtl/mx/mtl_mx.h
ompi/mca/mtl/psm/mtl_psm_endpoint.h
ompi/mca/mtl/psm/mtl_psm.h
ompi/mca/pml/cm/pml_cm_component.h
ompi/mca/pml/csum/pml_csum_comm.h
ompi/mca/pml/dr/pml_dr_comm.h
ompi/mca/pml/dr/pml_dr_component.h
ompi/mca/pml/dr/pml_dr_endpoint.h
ompi/mca/pml/dr/pml_dr_recvfrag.h
ompi/mca/pml/example/pml_example.h
ompi/mca/pml/ob1/pml_ob1_comm.h
ompi/mca/pml/ob1/pml_ob1_component.h
ompi/mca/pml/ob1/pml_ob1_endpoint.h
ompi/mca/pml/ob1/pml_ob1_rdmafrag.h
ompi/mca/pml/ob1/pml_ob1_recvfrag.h
ompi/mca/pml/v/pml_v_output.h
opal/include/opal/prefetch.h
opal/mca/timer/aix/timer_aix.h
opal/util/qsort.h
test/support/components.h
This commit was SVN r21855.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
OMPI
and a language agnostic part in OPAL. The convertor is completely
moved into OPAL. This offers several benefits as described in RFC
http://www.open-mpi.org/community/lists/devel/2009/07/6387.php
namely:
- Fewer basic types (int* and float* types, boolean and wchar
- Fixing naming scheme to ompi-nomenclature.
- Usability outside of the ompi-layer.
- Due to the fixed nature of simple opal types, their information is
completely
known at compile time and therefore constified
- With fewer datatypes (22), the actual sizes of bit-field types may be
reduced
from 64 to 32 bits, allowing reorganizing the opal_datatype
structure, eliminating holes and keeping data required in convertor
(upon send/recv) in one cacheline...
This has implications to the convertor-datastructure and other parts
of the code.
- Several performance tests have been run, the netpipe latency does not
change with
this patch on Linux/x86-64 on the smoky cluster.
- Extensive tests have been done to verify correctness (no new
regressions) using:
1. mpi_test_suite on linux/x86-64 using clean ompi-trunk and
ompi-ddt:
a. running both trunk and ompi-ddt resulted in no differences
(except for MPI_SHORT_INT and MPI_TYPE_MIX_LB_UB do now run
correctly).
b. with --enable-memchecker and running under valgrind (one buglet
when run with static found in test-suite, commited)
2. ibm testsuite on linux/x86-64 using clean ompi-trunk and ompi-ddt:
all passed (except for the dynamic/ tests failed!! as trunk/MTT)
3. compilation and usage of HDF5 tests on Jaguar using PGI and
PathScale compilers.
4. compilation and usage on Scicortex.
- Please note, that for the heterogeneous case, (-m32 compiled
binaries/ompi), neither
ompi-trunk, nor ompi-ddt branch would successfully launch.
This commit was SVN r21641.
OMPI_* to OPAL_*. This allows opal layer to be used more independent
from the whole of ompi.
NOTE: 9 "svn mv" operations immediately follow this commit.
This commit was SVN r21180.
- Delete unnecessary header files using
contrib/check_unnecessary_headers.sh after applying
patches, that include headers, being "lost" due to
inclusion in one of the now deleted headers...
In total 817 files are touched.
In ompi/mpi/c/ header files are moved up into the actual c-file,
where necessary (these are the only additional #include),
otherwise it is only deletions of #include (apart from the above
additions required due to notifier...)
- To get different MCAs (OpenIB, TM, ALPS), an earlier version was
successfully compiled (yesterday) on:
Linux locally using intel-11, gcc-4.3.2 and gcc-SVN + warnings enabled
Smoky cluster (x86-64 running Linux) using PGI-8.0.2 + warnings enabled
Lens cluster (x86-64 running Linux) using Pathscale-3.2 + warnings enabled
This commit was SVN r21096.
Adapt orte_process_info to orte_proc_info, and
change orte_proc_info() to orte_proc_info_init().
- Compiled on linux-x86-64
- Discussed with Ralph
This commit was SVN r20739.
mode, because some functions are not inlined.
This commit was SVN r20736.
The following SVN revision numbers were found above:
r20701 --> open-mpi/ompi@b440c92455
opal layer.
Add a check against a maximum (actually get rid of ifs internally to
opal_bitmap.c) -- the functionality to set the current maximum size
opal_bitmap_set_max_size() is currently only used in attribute.c
to set the maximum OMPI_FORTRAN_HANDLE_MAX...
Tested on linux/x86-64 with intel-tests with all_tests_no_perf_f
run with 6 procs.
Let's look into MTT as well...
This commit was SVN r20708.
* compute mmap-file size more wisely and pass requested size to allocator
* change MCA parameters:
- get rid of mpool_sm_per_peer_size
- get rid of mpool_sm_max_size
- set default mpool_sm_min_size to 0
* no longer pad sm allocations to page boundaries
* have sm_btl_first_time_init check return codes on free-list creations
Have mca_btl_sm_prepare_src() check to see if it can allocate an EAGER fragment
rather than a MAX fragment if the smaller size works.
Remove ompi/class/ompi_[circular_buffer_]fifo.h and references thereto.
Remove opal/util/pow2.[c|h] and references thereto.
This commit was SVN r20614.
Moreover, I could not figure out why <time.h> would need to be
included twice. So, I substituted the former for the latter,
in the superfluous instantiation.
This commit was SVN r19859.
Only one place used the user name field - session_dir, when formulating the name of the top-level directory. Accordingly, the code for getting the user's id has been moved to the session_dir code.
This commit was SVN r17926.
(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.
