Cognos Express CPU Sizing

[jameswebber]

Guys,
I wanted to pick your brains about CPU sizing of your cognos express instances.

We have purchased the Reporter, Advisor & Xcelerator Products so according to this link (page 18):
http://ndb.bg/ndb/home-new.nsf/0/2dec95 ... xpress.pdf
(and this link http://ebookbrowse.com/cognos-express-9 ... d107036637)
The server requirements are:
Minimum CPU* and RAM: 4 cores / 4GB per installed product**
Recommended CPU* and RAM: 8 cores / 8GB per installed product*

I.e. minimum 12 cores and 12 GB recommending a whopping 24 cores 24 GB RAM (ouch) .
Our cores are much better that the 2GHz Intel Pentiums and surely no mid-size client can come up with 24 cores?
In fact our infrastructure team tell me the technology to provide a virtualised 24 core box is very new. Currently we have Vmware vSphere standard ESXi 4.1 installed on dual socket quad core xeon Intel SR2600UR server platform connected to Netapp FAS2050 SAN over an iSCSI dedicated network.

Anyone got any experience with performance?
Will a dual socket quad core suffice?

Just wondering what other TM1 peeps are running there CE installations on?

[jim wood]

I asses the number of cores based on the number of users you expect to hit concurrently. If your only going to have 5 users total a quad core would be more than enough. As for memory, the TM1 element will want most of that. Again that depends on the size of the model. If you know you are going to build something pretty big then I would look to have a little bit too much rather than to little.

If you are installing on 64-bit I would say 8GB is a minimum requirement. (As 64-bit applications use roughly twice the memory)

I hope that helps,

Jim.

[qml]

64-bit applications use roughly twice the memory

Not to challenge your whole assertion, but generally it tends to be closer to a 20-40% difference caused purely by the switch from a 32-bit to a 64-bit architecture (not all machine code needs to double in size, just the addressing bits). Of course software generally gets bigger and its memory usage larger for a number of reasons, two of them being cheaper hardware and lazy software developers.

[jim wood]

That's why I put roughly.

[jameswebber]

Thanks guys for the very prompt reply.

IBM advice I got back yesterday was 1 core dedicated to each connector, with 4 GB of ram per connector.
I.e. 3 connectors thats 3 cores and 12GB of ram with a caveat it is dependant on the workload for each connector, and the age and performance of the processors etc.

I think we will give it 4 dual cores (maximum a node on our virtual blade can handle) and 24GB since our current TM1 itself is running on 16GB.
Looks like with 9.5.2 we would make use of more cores with parallelisation

[tomok]

Not to challenge your whole assertion, but generally it tends to be closer to a 20-40% difference caused purely by the switch from a 32-bit to a 64-bit architecture (not all machine code needs to double in size, just the addressing bits).

I've done quite a few conversions for clients moving from 32-bit to 64-bit and I've never had one that came in at less than around 75% more memory needed in the 64-bit world.

[jim wood]

Me to. I think QML was coming theoretical point of view. But as I always say in theory bumble bees shouldn't be able to fly.

[stephen waters]

Looks like with 9.5.2 we would make use of more cores with parallelisation

Note that Cognos Express currently uses TM1 9.5.1 so you won't get the benefits of Parallel interaction yet, or probably for some time.

This also cause a problem for CX customers who want to move (or plan to move ) to Excel 2010 which is not suported by TM1 9.5.1

[jameswebber]

Yeah our infrastructure team was laughing at me for asking for 24 cores for an application that doesn't support office 2010.

Lack of 2010 support and no release date is a huge IBM failing IMHO.
Although I hear first half of 2012 is a possibility.