On 6/23/07, Jon Gabrielson <[EMAIL PROTECTED]> wrote:
> If you want a highly-available 2-node cluster then Lustre probably isn't
> what you want. If either node fails it is considered a "double failure"
> from the Lustre POV (both MDS, OST, client failing at once) and recovery
> from this will not work.
I'm not wanting a HA 2 node cluster, I'm actually wanting a
HA 4 node cluster. DRBD by itself works quite nicely as a 2 node
cluster but it doesn't support 4way only 2way.
That's the reason I was asking what the minimum lustre cluster size was.
The docs mentioned that people are using a 4node cluster in production.
What configuration would you use for a 4 node cluster?
1 MDS node
1 MDS failover node
1 OSS node (however many osts)
1 OSS node for ost failover -- do note that the osts can be primary
on either of the OSS nodes.
This would be the absolute minimum I think -- clients are of course
separate. This is for 1.4.x, I am not sure how requirements and
prerequisites have changed for 1.6.
--
Mustafa A. Hashmi
[EMAIL PROTECTED]
_______________________________________________
Lustre-discuss mailing list
[email protected]
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
