SG Extended Cluster - How many nodes?

[Marlou Everson]

I currently have a SG cluster with 2 nodes in one data center and 2 nodes in another data center across campus and a Quorum server.

I have documentation that says that data centers have to have the same number of nodes.

We are thinking about consolidating systems. I always see diagrams with 2 nodes in each data center. I was wondering if it is okay (HP supported) to have only 1 node in each data center?

Marlou

[Sridhar Bhaskarla · Accepted solution]

Hi Marlou,

Yes. You can have one node in one datacenter and the other in other datacenter. You can either have quorum server or an arbitrator node.

Look at the following document for supported configuration in the metrocluster section.

http://docs.hp.com/hpux/onlinedocs/B7660-90015/B7660-90015.html

-Sri

[Steven E. Protter]

One node in each data center is certainly possible. The high availability of your systems is based on the two nodes taking over applications.

Here is a general workplan.

1) Figure out how you are going to maintain heartbeat. Network congestion between the two data centers can cause split brain syndrom and TOC of one of the nodes.
2) Make sure all users have access to both data nodes. That way your failover IP addres will make the transition look quite seamless when one node or the otehr goes down.
3) What kind of connection is there to shared disk resources. In general SG failes over the application and gives up the volume group when one node fails. The second node takes over the volume group as part of the cluster configuration. Depending on distance you need fiber or a number of T1 lines to maintain access to the shared volume groups.
4) Come up with a test plan that roughly covers the following areas:
a) failover of individual applications and services when the application is failed over but the primary node is still running
b) test a full shutdown and running of the entire server suite of services on one system. Make sure you have the capacity in CPU and memory to pull it off.

General note: Your setup is only as strong as the heartbeat. If that is not rock solid, its not going to work, no matter how many nodes you have in each data center.

SEP

[IT Response]

Availability is the number one reason our customers use Serviceguard.
Availability should be the major determining factor for node location.
HP does not write a support specification for the number of nodes "supported" per site in a campus (extended) cluster.
Since a cluster may be comprised of 1 - 16 nodes, the cluster may certainly consist of just 2 nodes.
Having them located at different sites (and power circuits) is more "available" than if they were co-located (so long as redundant heartbeat networks are configured).

StephenD.

[Richard Perez]

Marlou
the short answer to your question is: yes, it's supported.

The long answer is: talk with a HP representative to know if your desired setup is supported.

Regards

[Marlou Everson]

Thanks for the information. I can continue planning consolidation of systems.

Our current cluster has a VA7400 in each data center with separate paths (separate trenches between buildings) for the data and network. Each node has 3 LAN connections: one for primary heartbeat/data, one for standby and one for dedicated heartbeat. Now it is time to plan for fewer/newer boxes.

Thanks,
Marlou