[zfs-discuss] Slow read performance
g.danti at assyoma.it
Mon Apr 2 15:27:22 EDT 2018
Il 02-04-2018 16:53 Richard Elling ha scritto:
> 1MiB / 8 = 128kiB
> -- richard
Sure, but the OP stated that both using 128K recordsize and 16M
recordsize changed nothing.
Moreover, with working prefetch, I (naively?) expect more blocks to be,
well, pre-fetched, leading to I/O coalescing (ie: 2x128K reads becoming
1x256K read). Is this expectation wrong?
Finally, how ashift=9 produce better streaming results?
Assyoma S.r.l. - www.assyoma.it
email: g.danti at assyoma.it - info at assyoma.it
GPG public key ID: FF5F32A8
More information about the zfs-discuss