Whamcloud - gitweb
LU-11290 osc: Batch gang_lookup cbs
authorPatrick Farrell <paf@cray.com>
Wed, 14 Feb 2024 08:44:47 +0000 (00:44 -0800)
committerAndreas Dilger <adilger@whamcloud.com>
Sun, 3 Mar 2024 10:16:24 +0000 (10:16 +0000)
commit17ec47e55c605ef9ae3a75ddc2e73873480a4ce9
tree1cc7b681bd7d3357326813f2ea1e1b15a34d96be
parent8909e5a09428da7e72609501dbd33cb4dd534528
LU-11290 osc: Batch gang_lookup cbs

The osc_page_gang_lookup call backs can be trivially
converted to operate in batches rather than one page at a
time.  This improves cancellation time for locks protecting
large numbers of pages by about 10% (after landing
another optimization (LU-11290 ldlm: page discard speedup)
it shows 6% for canceling a lock for 30GB cached file ).

Truncate to zero time (with one lock protecting many pages)
was improved by about 5-10% as well.  Lock weighing
performance should be improved slightly as well, but is
tricky to benchmark.

Lustre-change: https://review.whamcloud.com/33089
Lustre-commit: 0d6d0b7bc95a82dee02d35d0a8a41d24692cad45

HPE-bug-id: LUS-6432
Change-Id: Ib30594ae97182cbeb18051d6cee860c97ae7e119
Signed-off-by: Patrick Farrell <paf@cray.com>
Signed-off-by: Alexander Zarochentsev <alexander.zarochentsev@hpe.com>
Reviewed-by: Bobi Jam <bobijam@hotmail.com>
Reviewed-on: https://review.whamcloud.com/c/ex/lustre-release/+/54031
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
lustre/include/lustre_osc.h
lustre/mdc/mdc_dev.c
lustre/osc/osc_cache.c
lustre/osc/osc_io.c
lustre/osc/osc_lock.c