topic Multiple Clusters using same Heartbeat LAN in Operating System - HP-UX

Multiple Clusters using same Heartbeat LAN

Lynne Dawson — Tue, 20 Feb 2001 16:44:26 GMT

Does anyone know if you can have two separate clusters, which both use the same subnet for a Heartbeat LAN?

For example, Cluster Letter contains nodes Alpha and Bravo, and Cluster Number contains nodes One and Two. Alpha and Bravo have an HEARTBEAT_IP configured for 10.245.1.10 and 10.245.1.11 respectively (with a netmask of 255.255.255.0). Can nodes One and Two, which also have LAN cards configured with IP addresses 10.245.1.20 and 10.245.1.21, (with netmask 255.255.255.0) also have an HEARTBEAT_IP configured on these cards as well?

Or will both clusters get confused with which heartbeats belong to which cluster. Are heartbeats sent point-to-point with each node in a cluster sending a heartbeat to every other node in the cluster (or just to the cluster co-ordinator)? Or are heartbeats sent out as broadcast packets, and so go to every device on that subnet?

Look forward to your comments!

Lynne

Re: Multiple Clusters using same Heartbeat LAN

Emil Velez — Tue, 20 Feb 2001 17:29:30 GMT

It does not seem like a problem since the heartbeat messages have MAC addresses in them. In a training environment I have experienced no problems with 9 clusters on the same HB subnet.

Be careful about your HB interval and HB timeout.

Good luck.

Re: Multiple Clusters using same Heartbeat LAN

MARTINACHE — Tue, 20 Feb 2001 17:42:49 GMT

Hi,

No problem, this will work.

But I suggest you to use different subnet on each cluster :

Cluster Letter : 10.245.1.
Cluster Number : 10.245.2.

Regards,

Patrice.

Re: Multiple Clusters using same Heartbeat LAN

melvyn burnard — Wed, 21 Feb 2001 08:05:43 GMT

There are no issues or problems with clusters on the same subnets.
As the Mac address and ip address are in each in cluster binary, each node in the cluser knows exactly who to "talk" to, and the network sensor will only monitor those ip's/mac addresses that it knows about from the binary.
This is why if a lanic gets replaced, the binary needs to be updated with the replaced MAC address.
Fairly simple really.