Arbitrators for Metro Cluster

[robert meredith_2]

We are thinking about metro clusters. It seems we will have to have a seperate arbitrator nodde for each cluster. So if we have 20 clusters at the moment in local SG config we will have to buy 20 new machines. Anyone got any experience of this.

On another note, anyone got any experience of LVM mirroring over fibre over metro distances which would mean we could run campus cluster instead.

Cheers

Rob

[melvyn burnard]

There seems to be a little confusion here, as MetroCluster is one cluster, with a requirement for arbitration nodes, or a Quorum Server.
You cannot have 20 clusters in one MetroCluster

[robert meredith_2]

Melvyn

I realise that you can't have lots of clusters inside one metrocluster. We will have lots of metroclusters. (Can you do this all connected to one XP, I hope you can!!). So the point is that to change our current cluster setup to Metro we have to buy lots of new arbitrator nodes.

Cheers,

Rob

[melvyn burnard · Accepted solution]

Well that is not strictly true. We now support MetroCluster and the Quorum Server.
A separate node outside the cluster can run the free Quorum Server software, and serves up to 100 nodes, or 30 clusters, whichever limit is reached first.
Of course, you need to be aware of the limitation of using a QS, and that is there can only be one, so there is a n issue there, but you might have the same issue with only one arbitrator node.

May I point you to:
http://docs.hp.com/hpux/onlinedocs/B8467-90001/B8467-90001.html
for the Quorum server notes and

http://docs.hp.com/hpux/onlinedocs/B8109-90010/B8109-90010.html

for MetroCluster with XP arays or
http://docs.hp.com/hpux/onlinedocs/B6264-90013/B6264-90013.html

for MetroCluster with EMC/SRDF

Hope that helps

[robert meredith_2]

Melvyn, good stuff, I didn't know about the QS product.

A couple more questions:

a) It is OK to run lots of metroclusters off the same XP isn't it? I assume you can partition the CA software so that when one cluster flips over the R/W R/O settings are only changed for the area of disk space assigned to that cluster.

b) There is a lot of stuff in the docs about taking equal numbers of nodes down in each data centre for planned maintenance. If we have a 2 node cluster this isn't much good!! I assume if we have planned maintenance we disable package switching while we are doing it? Thus no worse off than our 2 node clusters at the moment??

Cheers,

Rob

[melvyn burnard]

Rob

AFAIK, you cannot share the array between clusters, but I stand to be corrected on that point.

As far as the 2 nodes etc, the re is no major issue with this, excpet to understand the consequences of actions taken when you do halt a node.
So yes, for planned work, you would want to have pkg switching disabled.

You may also want to look at the free ServiceGuard Manager software availbale at http://software.hp.com and manuals are also on the docs web page.

[melvyn burnard]

Hmm, doing some good old RTFM, I believe the rule is no SHARED devices may be attached to anything OUTSIDE the Cluster.
Therefore if you have a part of the array explicitly for one cluster, and another part for another cluser, this should be ok.

However for your own sanity I would check this with your local HP office first.

[melvyn burnard]

Another thought on the XP side. It may be that you cannot share the array over multiple clkusters due to the CA software activation etc.
I would strongly advise you discuss this with your HP SalePerson before committing to it.

HTH

[robert meredith_2]

cheers for your thoughts melvyn, i'm off to harrass our local people.