Date: Tue, 26 Apr 2016 22:33:44 -0400 From: PK1048 <paul@pk1048.com> To: Andy Farkas <andyf@andyit.com.au> Cc: freebsd-fs@freebsd.org Subject: Re: How to speed up slow zpool scrub? Message-ID: <56C0A956-F134-4A8D-A8B6-B93DCA045BE4@pk1048.com> In-Reply-To: <571FEB34.7040305@andyit.com.au> References: <698816653.2698619.1461685653634.JavaMail.yahoo.ref@mail.yahoo.com> <698816653.2698619.1461685653634.JavaMail.yahoo@mail.yahoo.com> <571F9897.2070008@quip.cz> <571FEB34.7040305@andyit.com.au>
next in thread | previous in thread | raw e-mail | index | archive | help
> On Apr 26, 2016, at 18:27, Andy Farkas <andyf@andyit.com.au> wrote: >=20 > On 27/04/2016 02:34, Miroslav Lachman wrote: >> DH wrote on 04/26/2016 17:47: >>>> 5GB of RAM >>>=20 >>> That seems to be an insufficient amount of system ram when employing = zfs. >>>=20 >>> Take a look at this: >>>=20 >>> http://doc.freenas.org/9.3/freenas_intro.html#ram >>=20 >> I know 5GB is not much these days but is memory used for scrubbing a = lot? Because I am satisfied with working performance. The only concern = is slow scrubbing and I am not sure that more memory helps in this case. I don=E2=80=99t expect memory to make a big difference in scrub or = resilver performance. Rememeber the way ZFS uses memory is as both write = buffer and read cache (all in the ARC). So insufficient memory will hurt = real performance but it should not have any real effect on a scrub or = resilver (which are both operations that read all the data that has been = written to the zpool and check it against the metadata for consistency). Scrubs (and resilver) operations are essentially all random I/O. Those = drives are low end, low performance, desktop drives. The fact that the scrub _repaired_ anything means that there was damage = to data. If all of the data on the drives is good, then a scrub has = nothing to repair. What does an `iostat -x 1` show during the scrub ? How about a `zpool = iostat -v 1` ? How hard are you hitting those drives and are they all = really good ? I have seen svc_t values differ by a factor of two among = drives of all the same make and model. Is one drive slower than the rest = ? Perhaps that drive is on the way out.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?56C0A956-F134-4A8D-A8B6-B93DCA045BE4>