DSXchange

The "Architect" here has decided that we will run a cluster configuration running the engines on 2 x 8 cpu linux boxes.

He seems to think that this will give us high availability.

I was wondering if this this is likely to be more performant than running the engine on a 16 cpu box.

Also which would be cheaper for the license cost, I assume they are the same.

I am guessing GRID would be more.

Any advice, will we get High or Higher Availability with a cluster?

If you have two boxes you can have true high availability, in either active-active or active-passive configuration. Read all about this in the Planning, Installation and Configuration Guide. However, that would give you high availability of an eight core configuration.

Active-active is the highest cost. Active-passive is cheaper. Single 16 core might be about the same as active-passive or slightly cheaper, depending upon the processor type (research processor value units (PVUs), IBM's basis for licensing charges).

Grid configuration ought not to make any difference to your licence cost for the same maximum number of processors, again depending upon how you decide to configure (e.g. number of head nodes vs number of compute nodes).

I believe that "active-active" is what the doc refers to as "fully clustered." You license according to what is active.

Not really. Both active-active and active-passive refer to how failover is managed.

Thanks for the replies. Just to clarify:

With Grid the workload is shared over the available machines. If one goes down the work continues on the available machines.
This has a cost though.

With two boxes, there is the performance option, or the HA option.

HA: Run on one box and the other box is standby if it goes down.

Performance: The job runs on nodes on both machines. There is no HA, although the boxes could be re-configured to run on one machine.

Is this the case?

Regards

The license costs and performance depends on which tier or tiers you are talking about, and what type of HA you might configure (lower cost active-passive vs. an alternative). There are so many topology and HA variations to choose from. One HA consideration is automatic vs. manual failover and whether or not you're willing to wait 10-15 minutes for the failover to take place (HA with a 15 minute failover vs. 24x7 HA). It may help to clarify which tiers will land where or if you're only talking about the engine tier.

Check out this IBM Redbook and share it with your architect.

IBM InfoSphere Information Server Deployment Architectures

Hi Eric

Thanks for that, we are talking about just the engine tier.

From what I can work out, an MPP Topology setup allows all the engine to be used at once.
But if one goes down it says "A failure of single server will have a corresponding fractional impact on the total server resources that are available."

This sounds better than the Active Passive Cluster (and the Active - Active Cluster?). It appears only one engine is used at once and it says "In the event of a hardware failure of the active server, the HA software unmounts the external storage from the previously active server, and mounts it on the passive server".

However I guess with the MPP you are paying license for all engine servers, where as with Active Passive you are only paying for one.

blewip wrote:But if one goes down it says "A failure of single server will have a corresponding fractional impact on the total server resources that are available."

In an active-active cluster, or MPP, or grid setup, you do have to license the max that is active. In your scenario, if one engine system goes down then the fraction is 50%. If you switched things up a bit and and 8 x 2 cpu boxes, then if one goes down, you're in better shape.

The active-passive cluster would cost less and give you some HA assurance, with 8 cores max active at once.

One thing you have to watch out for these days is that servers have so many layers of virtualization that it can get more confusing. For instance, if you allocated 2 x 8 cpu servers or 8 x 2 cpu servers, etc., you would not want to find out after a physical failure that all such LPARs were running on the same physical unit.

qt_ky wrote:you would not want to find out after a physical failure that all such LPARs were running on the same physical unit.

That never happens.

We recently found out that our wonderful hardware partner has almost all of our virtual servers - dev, test *and* production - on a single physical server.

I can imagine.

Everyone really hopes you are right!