Background: In order to support atomics each btl needs to provide support
for communicating with self unless the btl module can guarantee global
atomicity. Before this commit bml/r2 discarded any BTL with lower
exclusivity than an existing send btl. This would cause the BML to
discard any btl other than self.
The new behavior is as follows:
- If an exisiting send btl has higher exclusivity then the btl will not be
added to the send btl list for the endpoint.
- If a btl provides RDMA support then it is always added to the rdma btl
list.
- bml_btl weight for send btls is now calculated across all send btls.
- bml_btl weight for rdma btls is now calculated across all rdma btls.
With this change self should still win as the only send btl for loopback
without disqualifying other btls (ugni, openib) for atomic operations.
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
of existing BTL fo an endpoint, all the others are just partial list.
Thus, all the cleaning should first be done in the btl_send array,
and them in the other arrays (btl_eager and btl_rdma).
This commit was SVN r31834.
This commit fixes two issues:
- This intent of the code @ bml_r2.c:486 is to prevent calling the
btl_del_procs more than once for a given proc. Gilles correctly
identified there was a problem in this code but r31786 we not the
correct fix.
- Fix a segmentation fault in r2 finalize revealed by the fact we
actually call del_procs now.
cmr=v1.8.2:reviewer=ggouaillardet:ticket=trac:4645
This commit was SVN r31829.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
r31786 --> open-mpi/ompi@fc96b0a7b8
The following Trac tickets were found above:
Ticket 4645 --> https://svn.open-mpi.org/trac/ompi/ticket/4645
This commit fixes leaks of bml endpoints in finalize. A summary of the
bugs/fixes is below.
1) ompi_mpi_finalize used ompi_proc_all to get the list of procs but
never released the reference to them (ompi_proc_all called
OBJ_RETAIN on all the procs returned). When calling del_procs at
finalize it should suffice to call ompi_proc_world which does not
increment the reference count.
2) del_procs is called BEFORE ompi_comm_finalize. This leaves the
references to the procs from calling the pml_add_comm
function. The fix is to reorder the calls to do omp_comm_finalize,
del_procs, pml_finalize instead of del_procs, pml_finalize,
ompi_comm_finalize.
3) The check in del_procs in r2 checked for a reference count of
1. This is incorrect. At this point there should be 2 references:
1 from ompi_proc, and another from the add_procs. The fix is to
change this check to look for a reference count of 22. This check
makes me extremely uncomforable as nothing will call del_procs if
the reference count of a procs is not 2 when del_procs is
called. Maybe there should be an assert since this is a developer
error IMHO.
cmr=v1.8.2:reviewer=bosilca
This commit was SVN r31782.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
This commit fixes three leaks:
- bml/r2: fix leak of del_procs in mca_bml_r2_del_procs
- Release the modex data in btl/scif, btl/ugni, and btl/vader
- ompi_mpi_finalize: close the allocator framework
cmr=v1.8.2:reviewer=jsquyres
This commit was SVN r31778.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
configure-time dynamic allocation of flags. The net result for platforms
which only support BTL-based communication is a reduction of 8*nprocs bytes
per process. Platforms which support both MTLs and BTLs will not see
a space reduction, but will now be able to safely run both the MTL and BTL
side-by-side, which will prove useful.
This commit was SVN r29100.
* add a new MCA param orte_hostname_cutoff to specify the number of nodes at which we stop including hostnames. This defaults to INT_MAX => always include hostnames. If a value is given, then we will include hostnames for any allocation smaller than the given limit.
* remove ompi_proc_get_hostname. Replace all occurrences with a direct link to ompi_proc_t's proc_hostname, protected by appropriate "if NULL"
* modify the OMPI-ORTE integration component so that any call to modex_recv automatically loads the ompi_proc_t->proc_hostname field as well as returning the requested info. Thus, any process whose modex info you retrieve will automatically receive the hostname. Note that on-demand retrieval is still enabled - i.e., if we are running under direct launch with PMI, the hostname will be fetched upon first call to modex_recv, and then the ompi_proc_t->proc_hostname field will be loaded
* removed a stale MCA param "mpi_keep_peer_hostnames" that was no longer used anywhere in the code base
* added an envar lookup in ess/pmi for the number of nodes in the allocation. Sadly, PMI itself doesn't provide that info, so we have to get it a different way. Currently, we support PBS-based systems and SLURM - for any other, rank0 will emit a warning and we assume max number of daemons so we will always retain hostnames
This commit was SVN r29052.
This creates a really bad scaling behavior. Users have found a nearly 20% launch time differential between mpirun and PMI, with PMI being the slower method. Some of the problem is attributable to poor exchange algorithms in RM's like Slurm and Alps, but we make things worse by calling "get" so many times.
Nathan (with a tad advice from me) has attempted to alleviate this problem by reducing the number of "get" calls. This required the following changes:
* upon first request for data, have the OPAL db pmi component fetch and decode *all* the info from a given remote proc. It turned out we weren't caching the info, so we would continually request it and only decode the piece we needed for the immediate request. We now decode all the info and push it into the db hash component for local storage - and then all subsequent retrievals are fulfilled locally
* reduced the amount of data by eliminating the exchange of the OMPI_ARCH value if heterogeneity is not enabled. This was used solely as a check so we would error out if the system wasn't actually homogeneous, which was fine when we thought there was no cost in doing the check. Unfortunately, at large scale and with direct launch, there is a non-zero cost of making this test. We are open to finding a compromise (perhaps turning the test off if requested?), if people feel strongly about performing the test
* reduced the amount of RTE data being automatically fetched, and fetched the rest only upon request. In particular, we no longer immediately fetch the hostname (which is only used for error reporting), but instead get it when needed. Likewise for the RML uri as that info is only required for some (not all) environments. In addition, we no longer fetch the locality unless required, relying instead on the PMI clique info to tell us who is on our local node (if additional info is required, the fetch is performed when a modex_recv is issued).
Again, all this only impacts direct launch - all the info is provided when launched via mpirun as there is no added cost to getting it
Barring objections, we may move this (plus any required other pieces) to the 1.7 branch once it soaks for an appropriate time.
This commit was SVN r29040.
ompi_show_help, because opal_show_help is replaced with an
aggregating version when using ORTE, so there's no reason to
directly call orte_show_help.
This commit was SVN r28051.
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.
(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.
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.
Often, orte/util/show_help.h is included, although no functionality
is required -- instead, most often opal_output.h, or
orte/mca/rml/rml_types.h
Please see orte_show_help_replacement.sh commited next.
- Local compilation (Linux/x86_64) w/ -Wimplicit-function-declaration
actually showed two *missing* #include "orte/util/show_help.h"
in orte/mca/odls/base/odls_base_default_fns.c and
in orte/tools/orte-top/orte-top.c
Manually added these.
Let's have MTT the last word.
This commit was SVN r20557.
got a whole lot smaller, decreasing the memory footprint of the
running application. How much it's a good question. Here is a
breakdown:
- in mca_bml_base_endpoint_t: 3 *size_t + 1 * uint32_t
- in mca_bml_base_btl_t: 1 * int + 1 * double - 1 * float
+ 6 * size_t + 9 * (void*)
The decrease in mca_bml_base_endpoint_t is for each peer and the
decrease in mca_bml_base_btl_t is for each BTL for each peer.
So, if we consider the most convenient case where there is only
one network between all peers, this decrease the memory foot print
per peer by
9*size_t + 9*(void*) + 2 * int32_t + 1 * double - 1 * float.
On a 64 bits machine this will be 156 bytes per peer.
Now we access all these fields directly from the underlying BTL
structure, and as this structure is common to multiple BML endpoint,
we are a lot more cache friendly. Even if this do not improve the
latency, it makes the SM performance graph a lot smoother.
This commit was SVN r19659.
There was an argument that was barely used, and on return at the PML
level it contained nothing usable. It has been removed, so now we're
using less memory ...
This commit was SVN r19657.
Thanks to George and Jeff for pointing out a better way to do this.
This commit was SVN r19573.
The following SVN revision numbers were found above:
r19566 --> open-mpi/ompi@351c3a3a86
that by decreasing the btl_inuse if there was already a registered BTL we basically reset
the changes for this new BTL to register it's progress function, even if it was supposed to
handle another peer.
This commit was SVN r19080.
