Cluster nodes at different sites.

[Poornachandran SM]

Hi,

We would like to have a a two node Serviceguard cluster split in two different sites (nodeA in city A & nodeB in city B). We will not have any storage devices in the cluster and we can use Quorum server as cluster lock.

What is the recommendation wrt network (mainly SG heartbeat nw) to have this setup? How far the sites can be? What should be the network bandwidth? What can be the network latency?

I could see only one reference wrt our requirement in a doc named "Understanding and Designing Serviceguard Disaster Tolerant
Architectures", which mentions that the Cluster network latency should not exceed 200ms. Pointers/suggestions to reference or recommendations will be very much helpful.

[Rita C Workman]

That would be a Bob Sauers document.

Let me just say to have "clusters" at different sites (city) involves either Metro or Continental clusters (depending on distances).
But without storage replicating data - I do not understand what you're trying to achieve. You fail over the data that is on external disk (i.e. what makes up your "packages") because the failover occurs when you can no longer run on a particular server. So if all your data is on internal disk - then failing over even locally won't qualify for HA.

Cluster failover requires "shared" HA disk. Failover between sites requires replicating data between storage arrays.

Read anything by Bob Sauers on the subject. It will help you.

Rgrds,
Rita

[melvyn burnard]

Well I don't quite understand how or why you would want to have a cluster with no shared storage!!
Saying that, the limitation of sites for an extended or campus cluster come down to one of three criteria:
1) Maximum cluster size is 4 nodes with two data center architecture,
Maximum cluster size of 16 nodes with three data center architecture
Maximum distance determined by networking or Fibre Channel link type used, whichever is shorter (maximum range is from 10km to 100km)

2) There must be less than 200 milliseconds of latency in the heartbeat and Fibre Channel data links between the data centers

3) the two sites must be on the same subnet, UNLESS using the new functionality provided for SG 11.18 allowing cross-subnet clustering.

But I still fail to see why or how you could use a cluster without sharing ANY storage!

[Poornachandran SM]

Hi,

Thank you for the reply.
Yes. Our platform require neither data replication nor shared disk. We use this cluster purely to achieve HA of an application, which doesn't use any data from shared disk.

Considering these reqs, shall we take points 2 (n/w latency) & 3 (subnet) as the recommendations for our platform?

[Rita C Workman]

Poornachandran,

I think your missing the point. HA (high availability) can NOT be acheived without shared storage.

When that server stops - so will your application on it's internal disk. The application can not fail over to another servers internal disk. The only thing you could do is to restore the data via some other method or media to the second server. Doing this does not qualify it for HA High Available and defeats the whole purpose of a clustered failover environment.

Regards,
Rita

[Stephen Doud]

Patched Serviceguard 11.18 will permit you to create a cross-subnet cluster, meaning that you can operate a cluster across distances further than a single network/subnet will permit. Since you indicate that you will be using a quorum server, and don't need storage arrays, this might be the solution you need.

Be aware that though package failover can occur, the fact that different subnets are in use at each site precludes the use of the same floating (relocatable) IP across sites.

See pages 17, 18 and 33 in the April 2008 edition of the Serviceguard A.11.18 Release Notes, at http://docs.hp.com/en/B3935-90119/B3935-90119.pdf
Look for references to cross-subnet configuration.