HAST local read performance?
Mark Felder
feld at FreeBSD.org
Wed Mar 5 16:13:11 UTC 2014
On Wed, Mar 5, 2014, at 8:08, Karl Pielorz wrote:
>
> Hi,
>
> I have a couple of FreeBSD 10 systems I've setup HAST on - this seems to
> work well, but I've noticed the read performance of local /dev/hast/*
> devices is around half of the normal raw device?
>
> e.g. a dd from an SSD as /dev/da0 nets around 220Mbyte/sec - the same dd
> run against the same device when setup with hast (i.e. /dev/hast/disk1)
> only nets around 110Mbyte/sec.
>
> I realise another layer is going to affect performance (and this may just
> be the price you have to pay) but is there anything likely tunable for
> this?
>
> If I fail the other node (so I'm running the test against a 'degraded
> primary') it's still as slow [as you'd kind of expect, as reads were/are
> only local]
>
> I've setup ZFS on top of the hast devices - again this seems to work OK,
> but you can really see the performance difference when doing a pool scrub
> on hast backed devices vs. the raw ones...
>
Which HAST replication mode are you using? Fullsync? Do you have atime
enabled on that filesystem? I'd expect it to act like a local disk
without any significant penalties unless you're doing writes and they're
waiting to be synced before the next read.
More information about the freebsd-geom
mailing list