-
Notifications
You must be signed in to change notification settings - Fork 307
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
DAOS-16469 engine: bind helper - b26 #15307
base: release/2.6
Are you sure you want to change the base?
Conversation
Ticket title is 'IOR Easy performance low with EC_16P2GX' |
9627e2c
to
175d853
Compare
Test stage Functional Hardware Medium completed with status FAILURE. https://build.hpdd.intel.com//job/daos-stack/job/daos/view/change-requests/job/PR-15307/2/execution/node/1550/log |
f10ce8a
to
54e3af4
Compare
Test stage Unit Test bdev on EL 8.8 completed with status FAILURE. https://build.hpdd.intel.com/job/daos-stack/job/daos/job/PR-15307/5/display/redirect |
Test stage NLT on EL 8.8 completed with status FAILURE. https://build.hpdd.intel.com/job/daos-stack/job/daos/job/PR-15307/5/display/redirect |
Test stage Unit Test on EL 8.8 completed with status FAILURE. https://build.hpdd.intel.com/job/daos-stack/job/daos/job/PR-15307/5/display/redirect |
54e3af4
to
27fcff8
Compare
48baa9b
to
f68d151
Compare
When a main IO XS wants to add some task to helper XS (in spite of via chore queue or creating new ULT), it needs to take lock on related helper XS. Under the dss_helper_pool mode, if there are a lot of ULTs from different main IO XS doing that concurrently, the overhead caused by lock contention may be high. To reduce such overhead, we will divide main IO XS into small groups, and bind each group to a helper XS: the ones in the same group will share one helper, different groups use different helpers. Then above lock contention will be restricted inside the group. Depends on the system configuration, such solution may be unnecessary, for example, there is only single helper per engine, or very limited targets (less than 4 or about) per engine. So this feature will be configurable via environment variable DAOS_BIND_HELPER that is unset by default. NOTE: 1. Binding helper is only available when the count of helper XS is non-zero and smaller than the count of main IO XS. 2. If multiple sockets mode is enabled (dss_numa_nr > 1), then do not bind helper for simplification. 3. To be load balanced, if the count of main IO XS is not the integral multiple of the helper XS, then binding helper will be disabled automatically. Allow-unstable-test: true Signed-off-by: Fan Yong <[email protected]>
f68d151
to
ad6a15c
Compare
When a main IO XS wants to add some task to helper XS (in spite of via chore queue or creating new ULT), it needs to take lock on related helper XS. Under the dss_helper_pool mode, if there are a lot of ULTs from different main IO XS doing that concurrently, the overhead caused by lock contention may be high. To reduce such overhead, we will divide main IO XS into small groups, and bind each group to a helper XS: the ones in the same group will share one helper, different groups use different helpers. Then above lock contention will be restricted inside the group. Depends on the system configuration, such solution may be unnecessary, for example, there is only single helper per engine, or very limited targets (less than 4 or about) per engine. So this feature will be configurable via environment variable DAOS_BIND_HELPER that is unset by default.
NOTE:
Allow-unstable-test: true
Before requesting gatekeeper:
Features:
(orTest-tag*
) commit pragma was used or there is a reason documented that there are no appropriate tags for this PR.Gatekeeper: