[CentOS] New Tutorial - RHCS + DRBD + KVM; 2-Node HA on EL6

Digimer linux at alteeve.com
Wed Jan 4 17:30:11 UTC 2012


On 01/04/2012 11:52 AM, Jorge Fábregas wrote:
> On 01/03/2012 10:29 AM, Digimer wrote:
>> Hi all,
>>
>>   I'm happy to announce a new tutorial!
>>
>> https://alteeve.com/w/2-Node_Red_Hat_KVM_Cluster_Tutorial
> 
> Hello Digimer,
> 
> Thanks for sharing this.  I might try it in a couple of months as I'm
> not ready yet (need to grasp some concepts/technologies first).  I also
> haven't used KVM but I have some experience with VMware (vSphere Clusters).
> 
> For vSphere clusters you need a shared storage system:  ideally (in
> preference order) you'll be using a  FC SAN, iSCSI SAN or a NAS (serving
> NFS).  I'm interested in the DRBD part here.  Did you use it because you
> didn't have access to a shared storage system? or is it a requirement
> for a particular functionality you wanted?  Have you done it before with
> a shared system? Any considerable performance difference (DRBD vs
> shared-storage)?
> 
> Thanks!
> 
> Best regards,
> Jorge

When you get a chance to try it out, please feel free to ask for help if
you run into any issues.

I chose DRBD because of it's ease to implement and that it did not
require external storage. I've had very good success with performance of
DRBD, getting near-capacity speeds out of it (that is, near the speed of
the underlying storage). The only limitation is that DRBD is a best fit
at two nodes only. You can do three nodes with stacked configuration,
but I've not played with that so I can't comment on it's effectiveness.

As for external storage as a comparison, I can't say. I don't have
corporate backing or a hardware budget. :) I suspect though that the
real question will not be so much FC SAN vs DRBD as it will be the speed
of the underlying storage and the number and type of VMs hitting that
storage. The consistent issue I have to deal with in production is
storage seek latency. Thankfully, 15k drives and sufficient caching
seems to resolve this in most cases. Also, the distributed locking, by
it's nature, can be a source of slow down. So you need to allocate time
to tune both the storage and the locking when concerned with
performance, more than the details of the storage.

Cheers!

-- 
Digimer
E-Mail:              digimer at alteeve.com
Freenode handle:     digimer
Papers and Projects: http://alteeve.com
Node Assassin:       http://nodeassassin.org
"omg my singularity battery is dead again.
stupid hawking radiation." - epitron



More information about the CentOS mailing list