[Linux-cluster] GFS2 locking in a VM based cluster (KVM)

[Thomas Sjolshagen]

 On Tue, 04 Jan 2011 10:24:03 +0000, Steven Whitehouse 
 <swhiteho at redhat.com> wrote:
> Hi,
>
> On Tue, 2011-01-04 at 05:00 -0500, Thomas Sjolshagen wrote:
>> Hi,
>>
>> Posted this on IRC just now, but was recommended to try another 
>> approach.
>>
>> I'm wondering if the locks/sec rate I'm seeing between two virtual
>> machines is what I should expect - details below:
>>
>> I've got a 2-node cluster that is actually two KVM VMs (Fedora 14 in
>> the guests/vms). Between the two VM's, I'm sharing - iSCSI based
>> external array using virtio_net drivers in the guests/vms - two GFS2
>> file systems. After setting <dlm> and <gfs_controld> as listed below
>> in cluster.conf (and, of course, rebooting the whole cluster), I'm
>> still only seeing 600-650 locks/sec when using the ping_pong 
>> utility.
>>
>> <dlm plock_ownership="1" plock_rate_limit="0" protocol="sctp"/>
>> <gfs_controld plock_rate_limit="0"/>
>>
>> The same settings & configuration on the physical hosts (a separate
>> cluster, but using the same iSCSI array) is seeing 5.5-6K locks/sec.
>>
>> I'm also curious as to why, when starting ping_pong on the 
>> virtualized
>> node1, I see ~20-40K locks/sec, but if I start and then stop 
>> ping_pong
>> on the virtualized node2, node1 never returns to anything close to 
>> the
>> 20-40K locks/sec. Is that (also) expected?
>>
> Did you have the second node mounted when you got the faster locking
> rates on the first node?
>
 Yes, both nodes mounted (before and after starting ping_pong on the 2nd 
 node, no umount inbetween)

 // Thomas