Continental Cluster and package fail over question....

[Rita C Workman]

I'm looking at 'maybe' going with a Continental Cluster (as opposed to a Metro Cluster). My question is this:
I would have two sites: Primary site/cluster with 3 nodes and a Secondary site/cluster with 2 nodes.
Question:
Can I manually stop a package that is on 1 node in the Primary Cluster...and then start that package on a node in the Secondary Cluster? But bear in mind I do not want the Primary Cluster to go down...the other 2 nodes in the Primary Cluster would remain up and running where they are....I just want to move the package of one node to the other Cluster....Is this do-able?

Thanks,
Rita

[melvyn burnard · Accepted solution]

Rita
This is "do-able" but I would question why you wish to use this, as it not as simple a fail-over as normal package failover scenario in MetroCluster, or even normal SG clusters.

Apart from the cost factor, the process is automated but NOT automatic, and once begun generally means that the cluster that was running the package needs a lot oif work done to be able to recoevrm, which in itsef is not simply a switch back.

[Rita C Workman]

The problem I have is that the one node (server) in Primary Cluster has 2 Juke boxes scsi connected to it. We require constant access to the data in those Juke boxes (CD's). At the other site on the Secondary Cluster is another server...with Juke box(s) scsi connected and the backup copies of the Primary CD's. So even though that node (server) is connected as part of the Primary Cluster...failing over to another node (server) in THAT cluster is worthless; I have to be able to move that package to the other Secondary site. And since we are setting up dual EMC disk arrays with SRDF the rest of the data & apps is covered...it's those pesky CD's in those Juke boxes that has been creating this '...oh so weird situation'.

Thanks Melvyn,
Rita

[melvyn burnard]

Hmm, not really a good candidate for SG failover, but you may want to look at MetroCluster rather than Continental Cluster, if the two sites are within the same subnet and a distance that would allow this, which it sounds like it is if you are using EMC and SRDF.

[Rita C Workman]

We make it just under the 10mile limit for that Metro Cluster....Same subnet, well I probably could work that part out with very few changes. But the headache came up with quorum. Cluster A has 3 nodes (including that one with those scsi connected juke boxes) Cluster B has 2 nodes (with one having a juke scsi connected to it). Now you can't fail over unless you have & maintain quorum....arghhhhh!!! So there-in lied the problem, that, and the fact that now they say we need two more small boxes to act as arbitrators, preferably in a third location....arghhhh!!!
Talk about building a headache...that's why the Continental Cluster started looking far more appealing. Since both seperate cluster are failover sites of each other, I felt that 'manually' one package could be brought down on one cluster and 'manually' brought up on the other cluster. Bear in mind again I said manually...I'm not looking at mechanizing this package. It would be a node on the first cluster (A), but I will disable it from failing over anywhere in the event of a problem. I fully intend to always have this one done manually! This is a govt site...we are not e-commerce..so a few minutes down is not that big a deal (Thankfully..) Since you have to packages copied on ALL nodes in both clusters...I felt this was do-able.

What d'ya think????

/Rita

[melvyn burnard]

Rita, I agree that it is "do-able", but the cost of CC over MC is quite large, and may have other implications. ( please bear in mind I am unable to do site design etc).

As far as the MC side of things goers, yes you need an equal number of nodes at the primary site and secondary site, along with Arbitration node(s) on a third site ( which coul dbe at the same site as the production side.)

Also, these nodes need not be huge systems, all they need is the network connectivity to allow for heartbeat arbitration.
The deicsion is yours and will obviously need a lot of investigation.
Hope you get it right.

[Rita C Workman]

Thanks Mel...

...me too ....

Rita