[Bugs] [Bug 1467614] Gluster read/write performance improvements on NVMe backend

bugzilla at redhat.com bugzilla at redhat.com
Fri Oct 27 06:35:27 UTC 2017


--- Comment #36 from Manoj Pillai <mpillai at redhat.com> ---

While comment #35 separately establishes what the limits are on the client-side
and brick-side, it does not require us to change course. The priority should be
fixing the client-side bottleneck, which is what previous comments are dealing
with anyway. But when we do analyze brick-side bottlenecks we should use a
multi-client test to ensure enough IOPs are bring sent in to push the brick to
the limit.

For the single-client, single-brick test, the best result I saw was with
io-thread-count=2 (not much improvement from io-thread-count=4, but still).
That ideal number might change as we remove existing bottlenecks, but that's
where things stand as of now. In contrast, for the brick to reach its limit,
io-thread-count needs to be at least 8. I think the investigation here would
benefit from being able to separately control no. of io-threads on the
client-side and brick i.e. from a fix for bz #1499649.

You are receiving this mail because:
You are on the CC list for the bug.
Unsubscribe from this bug https://bugzilla.redhat.com/token.cgi?t=qGYp4lSaHD&a=cc_unsubscribe

More information about the Bugs mailing list