during the IPv6 patch. The most important is the multi BTL support. There
was a quite interesting bug. Instead of setting up the multiple connections
over different physical devices, based on the time when these connections
were created most of the time they were all using the same physical network.
Which, of course, was not the intended goal, as we top at the maximum
bandwidth available over one device instead of gathering all available
bandwidth from all devices.
Second, the IPv6 RFC suggest to use sockaddr_storage as a holder for the
IP information, but use a sockaddr* when we pass it to functions. This is
only partially corrected by this patch.
Some other minor cleanups.
This commit was SVN r14544.
This "feature" is disabled by default and it should not affect the current performance.
In case when the message size is large and segment size is smaller than eager size for particular interface,
the leaf nodes in generalized reduce function can overflood parent nodes by sending all segments without
any synchronization. This can cause the parent to have HIGH number of unexpected messages (think 16MB
message with 1KB segments for example). In case of binomial algorithm root node always has at least one
child which is leaf, so this can potentially affect the root's performance significantly [Especially in
large communicators where root may have quite a few children (binomial tree for example)].
When the segment size is bigger than the eager size, rendezvous protocol ensures that this does
not happen so it is not necessary.
Originally, the problem was exposed in "infinite" bucket allocator clean up time for "small" segment sizes
(which may explain some "deadlocks" on Thunderbird tests).
To prevent this, we allow user to specify mca parameter "--mca coll_tuned_reduce_algorithm_max_requests NUM"
this limits number of outstanding messages from a leaf node in generalized reduce to the parent to NUM.
Messages are sent as non-blocking synchrnous messages, so syncronization happens at "wait" time.
The synchronization actually improved performance of pipeline and binomial algorithm for large message sizes
with 1KB segments over MX, but I need to test it some more to make sure it is consistent.
Since there is no easy way to find out what is "the eager" size for particular btl, I set the limit to 4000B.
If message/individual segment size is greater than 4000B - we will not use this feature. This variable may
or may not be exposed as mca parameter later...
I did not have any problems running it and both "default" and "synchronous" tests passed Intel Reduce* tests
up to 80 processes (over MX).
This commit was SVN r14518.
- make opal_sockaddr2str() take a sockaddr_storage instead of a sockaddr_in6
so that it works for IPv4 and IPv6 addresses, and remove a whole bunch
of #ifs in the OOOB code.
- Fix a compiler warning in the TCP BTL due to run-time determined
array size by making it a dynamicly allocated array.
- Fix the unpacking code of IPv4 addresses when using IPv6 support, so
that the address is in the correct location (instead of in an IPv6
structure, use an IPv4 structure). Refs trac:1005.
This commit was SVN r14514.
The following Trac tickets were found above:
Ticket 1005 --> https://svn.open-mpi.org/trac/ompi/ticket/1005
Make sure that the wrapper selection is compiled out if not enabling FT. Before the
logic would skip over it since the conditional if statements would not be satisfied,
now there are no additional if statements when compiled out.
With this modification the selection logic looks nearly identical to pre-r14051
with the exception of the non-FT related improvements.
This commit was SVN r14491.
The following SVN revision numbers were found above:
r14051 --> open-mpi/ompi@dadca7da88
There is a binomial algorithm in the code (i.e., the HNP would send to a subset of the orteds, which then relay it on according to the typical log-2 algo), but that has a bug in it so the code won't let you select it even if you tried (and the mca param doesn't show, so you'd *really* have to try).
This also involved a slight change to the oob.xcast API, so propagated that as required.
Note: this has *only* been tested on rsh, SLURM, and Bproc environments (now that it has been transferred to the OMPI trunk, I'll need to re-test it [only done rsh so far]). It should work fine on any environment that uses the ORTE daemons - anywhere else, you are on your own... :-)
Also, correct a mistake where the orte_debug_flag was declared an int, but the mca param was set as a bool. Move the storage for that flag to the orte/runtime/params.c and orte/runtime/params.h files appropriately.
This commit was SVN r14475.
It's a bit longer, but much more clear in it's implementation I believe.
Fundamentally it is the same, but is much more solid in the implementation.
I created quite a few directed tests that this version of the implementation
now passes.
This commit was SVN r14457.
compiler suite. The rule is that ompi_info.h is supposed to be the
''first'' file included so that it can affect system header files if
necessary (and it is sometimes necessary, such as with the PGI
compiler suite).
If this breaks VC on Windows, we'll have to find another fix. More on
the mailing list...
This commit was SVN r14453.
The following SVN revision numbers were found above:
r14435 --> open-mpi/ompi@a20b43ace9
finally brings in functionality that is already on the 1.2 branch, and
was developed and tested in the v1.2ofed branch (and other places).
Short version of new features:
* Support for ibv_fork_init()
* Automatically fill in the openib BTL bandwidth value by
querying the HCA port
* Installdirs functionality
* Fixes to always use -I in the Fortran wrapper compilers (#924)
* Gleb's mpool updates
* Remove some kruft in btl/openib/configure.m4, therefore
fixing the harmless warnings noted in #665
* Bunches of updates to the Linux RPM spec file
I.e., effectively the same thing that r14411 brought to the v1.2
branch.
Also effectively brought in r14432 and r14433 (some fixes on top of
the original r14411 commit to v1.2). Still need to bring in the moral
equivalent of r14445 after this commit (fixes to installdirs).
This commit was SVN r14449.
The following SVN revision numbers were found above:
r14411 --> open-mpi/ompi@83b31314ae
r14432 --> open-mpi/ompi@a48f160595
r14433 --> open-mpi/ompi@68f346d2bc
r14445 --> open-mpi/ompi@13d366b827
- Move the PML Modex stuff out of the BML -- Abstraction violation.
- Also fix the location of the add_procs with respect to the stage gates.
This commit was SVN r14422.
to move it into the unexpected queue. Instead pack the data in
only one buffer. Now the code look more optimized and clear, but
I have a doubt about who's using this functionality. I think that
all BTLs always return only one memory segment attached to the
matching fragment (i.e. there is no unexpected iov type receive).
This commit was SVN r14416.
function to make it more scalable. The memory fragmentation is still high, but
at least in most of the cases (where all ressources are correctly released
before the cleanup) the code is now highly efficient. Before the code execute
in (N * (N-1))!, which take a while when the number of allocated ressources
increase (which is the case when a lot of unexpected messages are created).
The fix consist of checking if all items are freed and if it's the case
then do not recreate the free items list (as we know that everything will
be released). If this condition is not true, we fall back on the
original execution path (which is still sub-sub-sub ... optimal).
This commit was SVN r14406.
same "restrict" check as the top-level OMPI configure.ac script so
that it will guarantee to always get the same result. Therefore, the
#define for restrict will always have the same value in both
opal_config.h and romioconf.h, and we get 7 less warnings (6 in the IO
ROMIO component, 1 in ROMIO itself) when compiling with icc on Linux
(because PAC_C_RESTRICT and AC_C_RESTRICT would get different values
for the "restrict" #define in this case).
This commit was SVN r14387.
Fix for memory corruption in the restarted process stack. This stemed from
the brute force method we were previously using. This commit fixes this by
using a lighter weight solution focused in the r2 BML instead of above the PML.
This is a more efficient and flexible solution, and it solves the original
problem.
In the process I pulled out the ft_event function in the tcp BTL and r2 BML
into a set of *_ft.[c|h] files just to keep any updates to these code paths
as isolated as possible to make merging easier on everyone.
This commit was SVN r14371.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
The following Trac tickets were found above:
Ticket 977 --> https://svn.open-mpi.org/trac/ompi/ticket/977
Per discussions with Brian and Ralph, make a slight correction in
where components are installed. Use $pkglibdir, not $libdir/openmpi,
so that when compiled in the orte trunk, components are installed to
the right directory (because the component search patch is checking
$pkglibdir).
This commit was SVN r14345.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r14289
the real interest is for small to middle size unexpected messages. The unexpected messages are copied
by the PML in it's own unexpected buffers. Therefore, there is no reason to make a first copy in the
TCP BTL. The BTL can handle to the PML it's own buffer, and can be sure that once the callback
completed it can reuse the buffer, no matter what happened with the fragment.
This commit was SVN r14320.
time difference...
1) The PML makes an assumption on local/remote completion semantics of the BTL
which Self BTL does not obey, nor should it, so we fix the PML
2) The Get protocol must handle the case when sender and reciever do not agree
on wheter the data is contiguous
This commit was SVN r14313.
Sorry for configure changes during the day; I totally forgot about
that. :-(
This commit was SVN r14288.
The following SVN revision numbers were found above:
r14286 --> open-mpi/ompi@0083eba18e
The top-level OMPI configure script already checks for "restrict" and
will issue a #define for it. PAC_C_RESTRICT would also check for
restrict, but sometimes come up with a different answer than the
top-level OMPI configure script, thereby resulting in conflicting
#define's for "restrict" (e.g., icc 9.0/9.1 on linux x86-64).
So it's easiest just to remove this test from ROMIO's configure.in
script.
This commit was SVN r14286.
Remove a redundant statement in the r2 BML.
This commit was SVN r14228.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
- Remove an old comment from crcp_base_fns.c
- Let ob1 have its very own ft_event function (which I'll fill in shortly)
- Make sure ob1 finalizes the bsend stuff so we don't leave a bunch of memory sitting around
- PML base - destruct the array upon finalize. Shrink the include search so it stops after finding a match
This commit was SVN r14222.
creating ompi_mpi_comm_null, since the destructor of ompi_mpi_comm_null will
decrease the reference counter of ompi_mpi_group_null twice according to the
last fix of Mohamad.
Added also a lengthy comment in ompi_comm_finalize about why we do
not decrease the reference counters for ompi_mpi_comm_null,
ompi_mpi_group_null etc. for the parent
communicator, although we do increase it in ompi_comm_init
This commit was SVN r14210.
when the size of the data is zero. Now they are not updated, which leave
us with the correct memory layout in all situations (so far). Update all
the comments to reflect exactly the supported behavior of the DDT engine.
This commit was SVN r14202.
computation of the current location on the pack/unpack process. This can
be used both for retrieving the pointer to the first byte (in the special
case of the cached RDMA protocol) and for getting the current
position (for the pipelined protocol).
I modified all BTLs, but most of them are still untested.
This commit was SVN r14180.
some time ago (last summer) that included checking for M_TRIM_THRESHOLD and
M_MMAP_MAX, unfortunately we didn't include <malloc.h> which is where these
are define, so disabling sbrk for the registration cache has been busted for
some time.
This commit was SVN r14169.
mca_mpool_base_page_size_log. They are exported by the mpool/base/base.h,
if some other code need them, then it should include this file
instead of having it's own redefinition of these externals.
This commit was SVN r14156.
Back out r14073 - it speeds up TCP latency / bandwidth but at the same time
it kills ROMIO and one-sided performance when using only TCP. The problem
is that it only allows those two to be progressed every couple of seconds,
leading to what looks like hangs in the one-sided tests (and the ROMIO stuff,
although people seem to not notice that at this point).
This commit was SVN r14144.
The following SVN revision numbers were found above:
r14073 --> open-mpi/ompi@64fbbc20b8
r14142 --> open-mpi/ompi@241545a098
it kills ROMIO and one-sided performance when using only TCP. The problem
is that it only allows those two to be progressed every couple of seconds,
leading to what looks like hangs in the one-sided tests (and the ROMIO stuff,
although people seem to not notice that at this point).
This commit was SVN r14142.
The following SVN revision numbers were found above:
r14073 --> open-mpi/ompi@64fbbc20b8
it limits the number of circular buffers allocated between each pair of peers.
This allows for more tight memory usage control.
This commit was SVN r14120.
waist slightly more memory, but prevents problem when fifo cannot be allocated
later during a job run when memory resource is exhausted.
This commit was SVN r14119.
if less than or equal pml_ob1_unexpected_limit just buffer in the PML level recv
fragment else allocate a buffer via the bucket allocator
This commit was SVN r14117.
development trees since last year (had to wait for some intel tests to
run yesterday, so I finally took the time to finish this work):
* Improve MPI API argument checking by also checking for NULL values
(especially helps when invalid Fortran MPI handles are passed,
because the various MPI_*f2c functions are supposed to return an
"invalid" MPI handle [meaning NULL] when this happens). So now
OMPI will generate an MPI exception rather than a segv.
* Removed a few redundant DATATYPE_NULL checks.
* Also check for some other forms of "invalid" handles (e.g., already
been freed, etc.) in some cases. We could probably be a bit more
stringent in this regard if we really wanted to.
* Change MPI_Get_processor_name to zero out the string up to
MPI_MAX_PROCESSOR_NAME characters, because the MPI spec says that
the string must be at least that long. We were already passing
that length to gethostname(), anyway.
This commit was SVN r14100.
latency is high and the network relatively fast. This will allow for more kernel
level buffering, which allow overlap between system calls and communications.
Somehow, even on fast clusters there is an improvement (non significant).
This patch create multiple modules for the same device, which in turn will
create multiple sockets between the peers. By default the number of BTL by
device is set to 1, so there is no fundamental difference with the current
version. Change the value of btl_tcp_links to enable multiple links between
peers.
This commit was SVN r14076.
when we precalculate most of the addresses there is no point to have separate
BTL for this. The sm_progress() code become much more simple as a result.
This commit was SVN r14071.
This merge adds Checkpoint/Restart support to Open MPI. The initial
frameworks and components support a LAM/MPI-like implementation.
This commit follows the risk assessment presented to the Open MPI core
development group on Feb. 22, 2007.
This commit closes trac:158
More details to follow.
This commit was SVN r14051.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r13912
The following Trac tickets were found above:
Ticket 158 --> https://svn.open-mpi.org/trac/ompi/ticket/158
In that case, sendcount and sendtype are not valid and we need to use
recvcount and recvtype.
This commit fixes trac:943. Reviewed by Jelena Pjesivac-Grbovic.
This commit was SVN r14022.
The following Trac tickets were found above:
Ticket 943 --> https://svn.open-mpi.org/trac/ompi/ticket/943
Queue_empty is determined by the reader, and is it's local view.
However, the writer may continue writing to this queue. The decision
to go on to the next cb_fifo is done in an atomic region, checking the
writer's view. The writer also "changes it's view" in an atomic
region protected by the same lock.
This commit was SVN r13968.
test, Sun's darray test, and an internal LANL test code. I would not
assume it will work properly on other codes, as I'm still not sure I
completely understand what the standard says this function is supposed to
do.
Refs trac:65
This commit was SVN r13967.
The following Trac tickets were found above:
Ticket 65 --> https://svn.open-mpi.org/trac/ompi/ticket/65
- fixing line lengths and some of the comments
- possible bug fix (but I do not think we exposed it in any tests so far)
temporary buffers were allocated as multiples of extent instead of
true_extent + (count -1) * extent.
Everything is still passing Intel tests over tcp and btl mx up to 64 nodes.
This commit was SVN r13956.
The original code was not compensating for the space used by the header.
When memory got tight, the allocator would return a pointer to memory that
did not exist resulting in a SEGV for the application. This is a partial
fix for ticket #929.
Reviewed by Rich Graham.
This commit was SVN r13950.
create a duplicate type, because any duplicate type lose the PREDEFINED flag.
An MPI_LB (respectively MPI_UB) without the PREDEFINED tag is useless, as it's
not the a marker anymore. The solution is to return the same pointer, but once
the reference count has been increased. In order for this to work, I allowed
the destruction to check for the reference count of an object before complaining
about destroying a predefined type.
This fixed ticket #317.
This commit was SVN r13942.
Currently 3 algorithms are available:
- non-overlapping, reduce + scatterv, (works for non-commutative operations)
- recursive halving algorithm (copied from basic module)
- ring algorithm (similar to allreduce ring, for large messages)
This commit was SVN r13929.
MCA_BTL_SM_FRAG_SEND) and status success/fail in low bits of pointers we
are passing through circular buffer. The rank that receives ACK doesn't need
to look into data it received and this is a big win since this data is not in
the cache of the rank's CPU. (Note that we can use low bits of pointers because
free_list always return pointers aligned at least to cache line size).
This commit was SVN r13922.
allocated from mpool memory (which is registered memory for RDMA transports)
This is not a problem for a small jobs, but for a big number of ranks an
amount of waisted memory is big.
This commit was SVN r13921.
- mca_base_param_file_prefix
(Default: NULL)
This is the fullname of the "-am" mpirun option. Used to specify a ':'
separated list of AMCA parameter set files.
- mca_base_param_file_path
(Default: $SYSCONFDIR/amca-param-sets/:$CWD)
The path to search for AMCA files with relative paths. A warning will be
printed if the AMCA file cannot be found.
* Added a new function "mca_base_param_recache_files" the re-reads the file
configurations. This is used internally to help bootstrap the MCA system.
* Added a new orterun/mpirun command line option '-am' that aliases for the
mca_base_param_file_prefix MCA parameter
* Exposed the opal_path_access function as it is generally useful in other
places in the code.
* New function "opal_cmd_line_make_opt_mca" which will allow you to append a
new command line option with MCA parameter identifiers to set at the same
time. Previously this could only be done at command line declaration time.
* Added a new directory under the $pkgdatadir named "amca-param-sets" where all
the 'shipped with' Open MPI AMCA parameter sets are placed. This is the first
place to search for AMCA sets with relative paths.
* An example.conf AMCA parameter set file is located in
contrib/amca-param-sets/.
* Jeff Squyres contributed an OpenIB AMCA set for benchmarking.
Note: You will need to autogen with this commit as it adds a configure param.
Sorry :(
This commit was SVN r13867.
and another for dst descriptors. This provide partial solution to OB1 protocol
deadlock problem. We can limit number of RDMA descriptors (by setting
btl_openib_free_list_max to something different from -1) and if we will be
lucky to hit this limit before we fail to register more memory the protocol
will not deadlock. When we had only one list for src/dst descriptors we
deadlocked when we reached max limit for the list.
This commit was SVN r13844.
Algorithm allows user to specify the segment size to be used for computation/communication overlap.
The additional memory requirement for the algorithm is 2 x segment size.
It performed well for (really) large message sizes over MX and it passed intel Allreduce_c and Allreduce_loc_c tests.
This commit was SVN r13832.
Long unexpected messages were not generating PUT_START events
because the MD for long unexpected messages was configured to
ignore start events. When a long unexpected message arrived, it
traversed the match list, and ended up in the long unexpected MD.
As the long message is being consumed, the code called PtlMDUpdate()
to look for the message, but there was no event that indicated
that it had arrived. So, the update succeeded. Once the long
unexpected message was consumed, the PUT_END event showed up in the
event queue -- except the code wasn't looking for it anymore.
The PUT_START events exist specifically to handle ordering between
short and long unexpected messages, so PUT_START events can't be
ignored on long unexpected messages.
Modified the code to generate PUT_START events for both long and
short unexpected messages and handle matching up START and END
events appropriately.
This commit was SVN r13746.
- the block sizes are computed in more uniformn way.
The first k blocks may be 1 element larger than the remaining blocks.
The algorithm passed Intel Allreduce_c and Allreduce_loc_c tests, and
IMB-3.2 Allreduce, over TCP and both btl and mtl MX (up to 128 processes).
The algorithm still only supports commutative operations.
This commit was SVN r13738.
- Implement the BML/r2 finialize funciton
- Cleanup the btl close routine
- Wire up a pml_base_verbose MCA parameter so you can actually watch the PML selection logic if you really want to.
- Fix a potental segfault in the selection logic.
ompi_pointer_array_get_item() may return NULL, so we have to check for it
This commit was SVN r13734.
The following SVN revision numbers were found above:
r2 --> open-mpi/ompi@58fdc18855
outstanding requests can be limited using mca parameters.
The implementation passed Intel, IMB-3.2, and mpi_test_suite tests over
TCP and MX up to 128 processes (64 nodes), on both 32-bit and 64-bit machines.
It is not activated by default, but it should be useful for really large
communicator sizes.
This commit was SVN r13720.
middle of a struct). Now we properly define and name the union
outside the struct and simply create an instance of it inside the
struct.
This commit was SVN r13709.
support SEND_IN_PLACE causes badness because the BTL tries to use the
not-exactly-complete convertor. Don't need it in this situation anyway.
This commit was SVN r13700.
Implementation passed intel: MPI_Reduce_c , MPI_Reduce_loc_c, and MPI_Reduce_user_c tests
over TCP, BTL MX, and MTL MX, as well as, mpi_test_suite Reduce tests (up to 64 nodes).
The algorithm is still not activated by decision function (will be in the near future).
This commit was SVN r13657.
when on the Cray machine (aka when the NULL GPR is in use).
This commit was SVN r13638.
The following SVN revision numbers were found above:
r13582 --> open-mpi/ompi@041beeb1b6
buffer fails. If cb is already allocated, but it is full and allocation of
additional cb fails, we spin waiting for receiver to free space in existing
cb.
This commit was SVN r13635.
* Update exactly how we handle MPI exceptions, particularly with
respect to MPI-1 section 3.2.5, and how error handlers are only
invoked for the ''first'' request that generates an exception.
* Update the "see also" section to be consistent across all 8
MPI_Test* and MPI_Wait* functions.
* Fixes trac:560
This commit was SVN r13619.
The following Trac tickets were found above:
Ticket 560 --> https://svn.open-mpi.org/trac/ompi/ticket/560
adding new procs that the remote proc's pml is the same as our local pml.
Turns the hangs from mismatched PMLs into an abort, which is better,
I think.
This commit was SVN r13582.
open MTL MX and BTL MX and initialize them at the same time. The problem is
that both call mx_init and mx_finalize, solution is to add an external entity
that does the init and finalize (based on ref counting).
This commit was SVN r13576.
The C++ bindings were not tracking keyvals properly -- they were
freeing some internal meta data when Free_keyval() was called, not
when the keyval was actually destroyed (keyvals are refcounted in the
C layer, just like all other MPI objects, because they can live for
long after their corresponding Free call is invoked). This commit
fixes this problem and several other things:
* Add infrastructure on the ompi_attribute_keyval_t for an "extra"
destructor pointer that will be invoked during the "real"
constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This
allows calling back into the C++ layer to release meta data
associated with the keyval.
* Adjust all cases where keyvals are created to pass in relevant
destructors (NULL or the C++ destructor).
* Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype:
* Move several functions out of the .cc file into the _inln.h file
since they no longer require locks
* Make the 4 Create_keyval() functions call a common back-end
keyval creation function that does the Right Thing depending on
whether C or C++ function pointers were used for the keyval
functions. The back-end function does not call the corresponding
C MPI_*_create_keyval function, but rather does the work itself
so that it can associate a "destructor" callback for the C++
bindings for when the keyval is actually destroyed.
* Change a few type names to be more indicative of what they are
(mostly dealing with keyvals [not "keys"]).
* Add the 3 missing bindings for MPI::Comm::Create_keyval().
* Remove MPI::Comm::comm_map (and associated types) because it's no
longer necessary in the intercepts -- it was a by-product of being
a portable C++ bindings layer. Now we can just query the C layer
directly to figure out what type a communicator is. This solves
some logistics / callback issues, too.
* Rename several types, variables, and fix many comments in the
back-end C attribute implementation to make the names really
reflect what they are (keyvals vs. attributes). The previous names
heavily overloaded the name "key" and were ''extremely''
confusing.
This commit was SVN r13565.
The following Trac tickets were found above:
Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
investivating #817:
* Remove use of legal_numbits member and always just use the full
size of the array. There was a corner case where legal_numbits was
not an even multiple of the number of bits in the array where bits
would not get freed properly, ususally causing wasted fortran
MPI handles, or, as in the case of #817, wasted attribute keyvals
(i.e., the user freed them, but the bitmap didn't reflect the
free).
* Re-order some error checks to ensure that we don't segv (we don't
currently trigger this problem anywhere; I just noticed it while
doing the other attribute keyval and legal_numbits work).
Since this change affects all Fortran MPI handles, I ran all the intel
and ibm tests and all still pass with this change.
This commit was SVN r13561.
