Anyone using HAST in production / performance?
Karl Pielorz
kpielorz_lst at tdx.co.uk
Sun May 4 13:11:28 UTC 2014
--On 29 April 2014 16:55:04 +0100 Pete French <petefrench at ingresso.co.uk>
wrote:
> A belated followup to this, but yes, I have been using it in
> production for a while now - it replaces an older stack
> we constructed based on gmirror and ggated. Much mucy better and
> far more repliable.
I've used ggated before - I never got it working reliably enough to use :(
- But it was very, very fast! ;)
> Performance wise I havent done any real testing I am
> afraid, aside form looking at raw disc bandwidth using
> gstat when we are doing a heavy operation. There we get
> about 50 meg/second, which is more or less what I would expecrt
> over gig ether.
My main concern is - if I take 4 'local' disks and run ZFS atop of them (2
mirrors of 2) I get 130Mbyte/sec for writes.
If I switch that to 4 disks over HAST (all local, no remote node - i.e.
remote set to 'none') - that drops to 31Mbyte/sec. for the same setup.
It doesn't get any lower with the remote node enabled - but that's a heck
of a drop :( By the time you've setup a ZVOL and exported as iSCSI (or just
CIFS/NFS) there's really nothing left performance wise :(
> The setup is used to run mysql on top of - initially using myISAM tables,
> but now using innodb. Performance appears to be fine - i,.e. I havent
> noticed any issues, but as I said, we havent actually treid to measure
> it.
I guess if you're just using the zool 'locally' - that 31Mbyte/sec may be
close to the performance you're getting [estimated]?
What version of FreeBSD are you using? I'm just wondering if that's making
a difference...
-Karl
More information about the freebsd-geom
mailing list