DAOS-18988 mercury: add NA UCX patch to fix bulk binding performance issues - #18946
DAOS-18988 mercury: add NA UCX patch to fix bulk binding performance issues#18946soumagne wants to merge 3 commits into
Conversation
Signed-off-by: Jerome Soumagne <jerome.soumagne@hpe.com>
…issues Signed-off-by: Jerome Soumagne <jerome.soumagne@hpe.com>
|
Ticket title is 'Performance degradation and growing memory use by DFS writer' |
|
Test stage Unit Test with memcheck completed with status UNSTABLE. https://jenkins-3.daos.hpc.amslabs.hpecorp.net/job/daos-stack/job/daos//view/change-requests/job/PR-18946/2/testReport/ |
|
Test stage Functional on EL 9 completed with status UNSTABLE. https://jenkins-3.daos.hpc.amslabs.hpecorp.net/job/daos-stack/job/daos//view/change-requests/job/PR-18946/2/testReport/ |
Signed-off-by: Jerome Soumagne <jerome.soumagne@hpe.com> Priority: 2
| + | ||
| + /* Reuse cached address when one already exists for this worker, so that | ||
| + * repeated deserializations do not create a new EP for every operation */ | ||
| + if (!na_ucx_addr) { |
There was a problem hiding this comment.
my main concern is this part.
this works fine when it is the same client sending bulks over, so we can re-use this cached entry
what happens when it is a new client which happens to reuse old worker address? are we safe using this entry then or will it result in some host unreachable errors down the line?
There was a problem hiding this comment.
verified that the entry gets correctly removed on disconnect event
|
Test stage Functional on EL 9 completed with status UNSTABLE. https://jenkins-3.daos.hpc.amslabs.hpecorp.net/job/daos-stack/job/daos//view/change-requests/job/PR-18946/3/testReport/ |
5 participants
Steps for the author:
After all prior steps are complete: