[Gluster-users] iowait issues

[Ben Turner]

----- Original Message -----
> From: "Lindsay Mathieson" <lindsay.mathieson at gmail.com>
> To: "gluster-users" <gluster-users at gluster.org>
> Sent: Monday, November 3, 2014 12:41:59 AM
> Subject: [Gluster-users] iowait issues
> 
> Using glusterfs (2 nodes, one disk each, replicated) to serve up my
> kvm disk images. Wehn 4+ vm's are running, I'm seeing iowait upto 25%,
> which is causing problems.
> 
> Oddly my nas, which has crappy network performance compared to the
> gluster cluster doesn't display this problem, but it does use a RAID10
> array.
> 
> Would my iowait problems because by the gluster network replication or
> by the single disks (WD Red 3TB).
> 
> If the latter, would adding a ssd cache help? something else?

What is you NW bandwidth?  If you are using gigabit then I can see this as the problem.  I have found that without at very least 20 MB / sec of throughput per VM they get difficult to use with a GUI.  So with gigabit and replication:

Gigabit = 120 MB / sec, divide that by 2 for replication:

60 MB / sec / 4 VMs = 15 MB / sec per VM

That is under the at least 20 that I like to see.  With gluster you are using 1/2 you bandwidth for replication and with the NAS you are probably getting the full gigabit.  Also having the RAID on the NAS defiantly helps, but I think in this case you are probably hitting the NIC as the bottleneck.  I would look at NW bandwith first and then look at RAID, you need to have a fast enough back end to service the bandwidth of your NICs.  Also the more spindles you have the better IOPS / smallfile / random IO will be.  What about 10G NW cards or NIC bonding?  How many VMs are you looking to run?  What sort of workload?

-b

> thanks,
> 
> --
> Lindsay
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>