Slow resilvering with mirrored ZIL
Daniel Kalchev
daniel at digsys.bg
Fri Jul 5 10:30:47 UTC 2013
On 05.07.13 02:21, Steven Hartland wrote:
>
> To give a concrete example which may well be of use for others, we
> had a mysql box here with dual 60GB SSD L2ARC's whic after continuous
> increases in query write traffic we ended up with total IO saturation.
>
> As a test we removed the L2ARC, partitioning them into a 10GB SLOG and
> 40GB L2ARC and the machine was utterly transformed, from constant 100%
> disk IO to 10% as the SLOG's soaked up the sync transfers from mysql.
Also, you removed the ZFS fragmentation issue, which can severely impact
performance.
Unfortunately, this removes fragmentation for newly written data only,
old data remains fragmented --- one can only dream for the day when the
"block pointer rewrite" promise becomes reality... it could do wonders
to ZFS.
Alternatively, my proposal is for an background rewrite of ZFS blocks,
that could at least achieve: re-distributing blocks to newly added
vdevs, "fixing" compression/dedup of datasets, cleaning up fragmented
data etc.
It should be relatively easy to be implemented... (I am, unfortunately
not offering coding help due to lack of time)
Daniel
More information about the freebsd-fs
mailing list