MCSG VLAN issue

[TMcB]

Hello
I am looking for some advice regarding MC Service Guard and subnetting.
We have 2 new rp4410 servers and MCSG.
The 2 servers will be located in separate IT rooms approx 1/2km apart.
The issue that has arisen is that the 2 computer rooms are on separate VLANs.

I am aware that for MCSG to work, they have to be on the same VLAN, as the failover IP has to be able to migrate from one server to another, but our comms team have asked me to find out :
1. why cant the connection between the 2 servers be routed?

2. What options do we have to get the 2 servers into a cluster.
EG - will we have to upgrade MCSG to continental or campus cluster.
Or would an option be to use a direct fibre connection between the 2 servers.

There will only be 1 package, with one of the servers active, and the other in standby.
We will be connecting our data disks to an EMC storage array, and mirroring the local boot disks.

Your advice is greatly appreciated

[Duncan Edmonstone]

1. why cant the connection between the 2 servers be routed?

There are 2 reasons:

i) Serviceguard heartbeat operates at data-link level - below the IP stack. That's not a routed protocol.
ii) Serviceguard expects to failover an IP address with your package. So the same subnet/VLAN must be present at both sites.


2. What options do we have to get the 2 servers into a cluster.

To be honest it would be cheaper to get the LAN sorted out than implement ContinentalClusters.

HTH

Duncan

[TMcB]

thanks Duncan

To change our whole lan infrastructure would be a major project.
Thats why I was wondering if there is any work around leaving the 2 vlans in place.

Can I directly connect the 2 servers using fibre?

Thanks

[Steve Lewis]

When docs.hp.com comes back on-line, find the document entitled "Introduction to Serviceguard" and locate the section on types of cluster.

You are stretching things if you try to implement a local cluster over 500m. There are timing issues.

You need to be able to linkloop to the remote server over all heartbeat interfaces.

You can go for a campus cluster if you have a direct fibre link (or preferably 2 links along different trenches in case somebody sticks a JCB bucket through one of them like my cousin once did).

[Rita C Workman · Accepted solution]

I'll just address your question about trying this via a Continental Cluster.
The answer is No.

Yes, Continental Clusters are on different subnets...but NO they are not automatic failovers. They are manual failovers that require root access to run the command to enact the failover. And prior to running it,(at least in this shop) we would have to adjust DNS to resolve the package to the failover VLan subnet. Again confirming this is strictly manual intervention.

What you want is automatic failover; i.e. "local" failover. Steve suggests using the Campus Cluster which requires some networking to make it happen.

Kindest Regards,
Rita

[Court Campbell]

>To change our whole lan infrastructure would be a major project.

It shouldn't be a major project to change a switch port to be on a particular VLAN. If it is for some odd reason, just put the servers in the same room.