[LU-16368] ZFS RPMs without vectorization Created: 06/Dec/22 Updated: 06/Dec/22 |
|
| Status: | Open |
| Project: | Lustre |
| Component/s: | None |
| Affects Version/s: | Lustre 2.15.1 |
| Fix Version/s: | None |
| Type: | Improvement | Priority: | Minor |
| Reporter: | Patrick Keller | Assignee: | WC Triage |
| Resolution: | Unresolved | Votes: | 0 |
| Labels: | performance, zfs | ||
| Environment: |
RHEL8.6 |
||
| Epic/Theme: | ZFS |
| Rank (Obsolete): | 9223372036854775807 |
| Description |
|
The provided ZFS RPMs for RHEL86 Lustre servers are appearently built without vectorization support for raidz and fletcher4. The available algorithms are benchmarked on module load. The benchmarked implementations can be checked under
cat /proc/spl/kstat/zfs/fletcher_4_bench
implementation native byteswap
scalar 5497759569 4845665644
superscalar 5480937676 4858516672
superscalar4 5488439650 4847575376
fastest scalar superscalar
cat /proc/spl/kstat/zfs/vdev_raidz_bench
implementation gen_p gen_pq gen_pqr rec_p rec_q rec_r rec_pq rec_pr rec_qr rec_pqr
original 254722784 200239743 97275027 1207670510 225639247 35724759 87910053 20776517 20811404 13297656
scalar 1234755379 393704159 159539402 1211544896 385637841 266021529 195891453 153232648 90736440 81828326
fastest scalar scalar scalar scalar scalar scalar scalar scalar scalar scalar
The same hardware with official ZFS RPMs supports further optimizations, e.g.:
cat /proc/spl/kstat/zfs/fletcher_4_bench
implementation native byteswap
scalar 6421841866 4605167866
superscalar 9151122207 5740950514
superscalar4 8273988438 6065318192
sse2 10187627568 2365144114
ssse3 10428775432 7506302223
avx2 24481349937 19454027999
avx512f 35018614756 10724551066
avx512bw 36102325545 30780239769
fastest avx512bw avx512bw
A noticable performance improvement on high throughput scenarios could be observerd when changing to official ZFS RPMs with proper vectorization support. |