The btl_recv.h:lookup_sender() function uses the hashed ORTE proc name
to determine the sender of the packet. With add_procs_cutoff>0, the
usnic BTL may not have knowledge of all the senders.
Until the usNIC BTL can be adjusted to do something like the
openib/ugni BTLs (i.e., use opal_proc_for_name() to lookup unknown
sender proc names), set MCA_BTL_FLAGS_SINGLE_ADD_PROCS, which means
that ob1 will only all add_procs() once -- with all the procs in it.
Also in this commit, adapt the connectivity checker to not rely on
knowing all the senders (which is a bit easier than adapting the main
BTL send path): the receiving connectivity agent will simply echo back
the same PING message (which contains the sender's IP address+UDP
port) back to the sender without checking that it knows who the sender
is. If the sender receives the echoed PING back on the expexted
interface, it will find a match in the pending pings list. If the
sender receives the echoed PING back an unexpected interface, a match
will not be found, and the incoming PING message will be dropped.
Fixesopen-mpi/ompi#1440
This commit fixes an inconsistency between btl_openib_receive_queues,
btl_openib_max_send_size and btl_openib_eager_limit. Before this
commit if the ini file specified a set of default receive queues that
happen to not contain one large enough for the default max_send_size
of eager_limit users would see an error like:
WARNING: The largest queue pair buffer size specified in the
btl_openib_receive_queues MCA parameter is smaller than the maximum
send size (i.e., the btl_openib_max_send_size MCA parameter), meaning
that no queue is large enough to receive the largest possible incoming
message fragment. The OpenFabrics (openib) BTL will therefore be
deactivated for this run.
Local host: somehost
Largest buffer size: 65536
Maximum send fragment size: 131072
This commit adds code that detects the source of the max_send_size and
eager_limit values and sets either or both of them to the size
supported by the largest queue pair if both 1) the value is larger
than the largest queue pair size, and 2) the value was not set by the
user or a MCA configuration file.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit adds two m4 macros: OPAL_SUMMARY_ADD, OPAL_SUMMARY_PRINT.
OPAL_SUMMARY_ADD adds an item to a section in the summary. For example
OPAL_SUMMARY_ADD([[Transports]],[[Foo]],...,[yes]) will add the
following to the summary:
Transports
-----------------------
Foo: yes
With this commit two sections are added: Transports, Resource Managers.
The OPAL_SUMMARY_PRINT macro is called after AC_OUTPUT and prints out
some information about the build (version, projects, etc) and then
the summarys sections. It will additionally print a warning if
internal debugging is enabled.
Example output:
Open MPI configuration:
-----------------------
Version: 3.0.0 a1
Build Open Platform Abstration project: yes
Build Open Runtime project: yes
Build Open MPI project: yes
Build Open SHMEM project: no
MPI C++ bindings (deprecated): no
MPI Fortran bindings: mpif.h, use mpi, use mpi_f08
Debug build: yes
Transports
-----------------------
Cray uGNI (Gemini/Aries): no
Intel Omnipath (PSM2): no
KNEM Shared Memory: no
Linux CMA IPC: no
Mellanox MXM: no
Open UCX: no
OpenFabrics libfabric: no
OpenFabrics Verbs: no
portals4: no
QLogic Infinipath (PSM): no
tcp: yes
XPMEM Shared Memory: no
Resource Managers
-----------------------
Cray Alps: no
Grid Engine: no
LSF: no
Slurm: yes
Torque: yes
INTERNAL DEBUGGING IS ENABLED. DO NOT USE THIS BUILD FOR PERFORMANCE MEASUREMENTS!
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
This commit fixes several threading bugs:
- Add an additional lock to the btl_base_endpoint_t structure to lock
the list of pending frags. This allows the progress function to
attempt to send pending frags without needing to drop/reaquire the
lock. This should provide a small improvement in performance and
fixes a potential race between adding an removing items from the
pending list.
- Ensure fast boxes are only set up once by updating the send count
using atomics when needed and do not set the fast box buffer
pointer until the fast box is set up.
Closesopen-mpi/ompi#1408
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit fixes a bug that can occur when communicating via XRC to
peers on the same node. UDCM was not saving the SRQ numbers on the
loopback endpoint (which shares its ib_addr info with all local peers)
so any messages to local peers use an invalid SRQ number.
Fixesopen-mpi/ompi#1383
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
This bug fixes two issue with the ib_addr lock:
- The ib_addr lock must always be obtained regardless of
opal_using_threads() as the CPC is run in a seperate thread.
- The ib_addr lock is held in mca_btl_openib_endpoint_connected when
calling back into the CPC start_connect on any pending
connections. This will attempt to obtain the ib_addr lock
again. Since this is not a performance-critical part of the code
the lock has been changed to be recursive.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit fixes a bug that occurs when attempting a get or put
operation on an endpoint that is not already connected. In this case
the remote_srqn may be set to an invalid value as the rem_srqs array
on the endpoint is not populated. This commit moves the usage of the
rem_srqs array to the internal put/get functions where it is
guaranteed this array is populated.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit ensures ib_addr->remote_xrc_rcv_qp_num value is set when
creating the loopback queue pair. This is needed when communicating
with any other local peer.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit fixes two bugs in XRC support
- When dynamic add_procs support was added to master the remote
process name was added to the non-XRC request structure. The same
value was not added to the XRC xconnect structure. This error was
not caught because the send/recv code was incorrectly using the
wrong structure member. This commmit adds the member and ensure the
xconnect code uses the correct structure.
- XRC loopback QP support has been fixed. It was 1) not setting the
correct fields on the endpoint structure, 2) calling
udcm_xrc_recv_qp_connect, and 3) was not initializing the endpoint
data.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
mca_btl_openib_put incorrectly checks the qp inline max before
allowing an inline put. This check will always fail for an endpoint
that has not been connected. The commit changes the check to use the
btl_put_local_registration_threshold instead.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Commit open-mpi/ompi@400af6c52d
introduced a regression in XRC support. The commit reversed the
ordering of shared receive queue (SRQ) and completion queue (CQ)
completion. CQ creation must always preceed SRQ creation when using
XRC as the CQs are needed to create the SRQs. This commit fixes the
ordering so that CQs are always created before SRQs.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Cleanup the configury so we properly check for Singularity under the various typical use-cases
Bring the Singularity support online. We have to turn "off" the sm BTL as it segfaults from inside the container - root cause remains unclear. Also turned "off" the various OPAL shmem components in case they are involved and someone else tries to use them. Happily, the vader BTL works just fine!
These changes fix issue https://github.com/open-mpi/ompi/issues/1336
- improve abstractions: opal/memory/linux component should be single place that opeartes with
Memory Allocation Hooks.
- avoid collisions in case dynamic component open/close: it is safe because it is linked statically.
- does not change original behaivour.
The send code in the ugni btl has an optimization that enables it to
return 1 (fragment gone) in some cases. This optimization involved
removing the btl ownership and callback flags to ensure the fragment
stuck around long enough for its completion flag to be checked. This
works fine for the single-threaded case but not in the multi-threaded
case. It is possible that a fragment will be completed by another
thread while a thread is in mca_btl_ugni_send. This competition can
lead to a leaked fragment, missed callback, or both. To fix the issue
without removing the optimization a reference count has been added to
the fragment. Callbacks and fragment release will not be made until
the fragment reference count has reach 0. The count is incremented
before sending the frag and decremented after the completion flag has
been checked. The fix has been verified to work using a multi-threaded
RMA benchmark with the osc/pt2pt component.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit fixes a race condition that can cause an endpoint to be
added to the wait list multiple times. To fix the issue an additional
check has been added to ensure the endpoint is not on the wait list
after the wait list lock is held. The wait list processing code has
also been updated to keep the wait list lock until all wait listed
endpoints have been handled. This reduces the chance that an endpoint
that is being processed by the wait list code is not re-added to the
list by a competing send.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Three minor updates from the code review of
https://github.com/open-mpi/ompi-release/pull/933:
* Remove an extra blank line a show_help message
* We no longer allow -1 for the MCA param btl_usnic_av_eq_num, so
change the flag to REGINT_GE_ONE
* Change "num_blocks" definition to be in terms of block_len (not
eq_size)
A bunch of empirical testing has shown that increasing the retranmit
timeout from 1ms to 5ms doesn't adversely affect performance, yet
decreases the number of gratuitious retransmissions.
Add endpoints in a blocked manner so that we don't overrun the
fi_av_insert() event queue. Also make the AV EQ length an MCA param,
and report it in mca_btl_base_verbose >=5 output.
Sequence numbers will wrap around; it is not sufficient to check for
(seq-1) -- must use the SEQ_DIFF macro to properly handle the
wraparound.
This bug wasn't serious; it just meant we might retransmit one or two
extra times when retransmits were triggerd and the sequence numbers
wrapped around their sliding windows.
`cm_message_event_active == 1` but main thread has already stopped
processing messages and thus we will have the situation where one
message was left unhandled leading to a hang.
Per open-mpi/ompi#1230, add a comment explaining why we write to a
temporary file and then rename(2) the file, just so that future code
maintainers don't wonder why we do this seemingly-useless step.
