[Gluster-users] How beefy does an arbiter box have to be?
ravishankar at redhat.com
Wed Mar 30 01:33:23 UTC 2016
On 03/30/2016 12:12 AM, William Kern wrote:
> We have a few replica 2 clusters using Gluster 3.4 for various projects.
> They have worked very well, but we always had to be careful about
> split brain when doing maintenance or in the event of a failure.
> So now we are looking to do a forklift upgrade to 3.7.x and add
> Arbiter box(s) into the new setup (i.e replica 2 + arbiter)
> Can we get away with using some older machines from the bone pile (i.e
> core2 cpu 2/4GB) or should we be using proper server kit?
> Would SSDs on the arbiter help or are they even necessary?
Not an authoritative answer but I think it should be manageable. I
don't think SSDs for arbiter alone (and not the other 2 bricks of the
replica) would improve performance in any way because we don't do read()
or write() on the files of the arbiter brick.
Another option would be to reserve a brick on the existing nodes itself
when creating a volume. (Sorry, converting an existing replica 2 volume
to arbiter is not yet supported but is on the cards.)
> Note: we are still using 1G (jumbo mtu) networking as the 10G stuff
> hasn't trickled down to our group yet <grin>. That is why we are
> mostly interested in using Arbiter rather than a true replica 3
> Gluster-users mailing list
> Gluster-users at gluster.org
More information about the Gluster-users