TM1 is not caching data after it has calculated it

[redsince78]

Hi All,

Hoping someone can help here we have recently upgraded the Servers that our TM1 Application runs on but we are now experiencing the following issue

TM1 is not caching data after it has calculated it, even after a fresh reboot of the service.
So if nothing changes and a cube view is refreshed, it should be instant, but is not; eg TM1 recalculates the view.
Physical system memory consumption appears to be ok and working as expected. (eg server memory usage grows with use)
Internal TM1 performance stats confirm the cache is not taking.
We have tested on a default TM1 training model and get the same result

The same model, software version (different OS as below) have been loaded on a laptop and run successfully with correct behaviours suggesting the issue is not inherent to TM1. That the default training model bundled with TM1 is not behaving either is further confirmation.

[declanr]

Have you tried playing about with the VMM and VMT settings?
Might be worth setting the stargate loggers to record more info for you as well. Its most likely just that the views are not within the size or time constraints of VMM/VMT and the loggers will help you trace that.

[redsince78]

Hi Declan,

Thanks for the reply I believe they are but will get it checked to be sure

Cheers
Gareth

[declanr]

Should also point out that from memory I think the default demo instances from IBM would all be so small in their datasets that they probably wouldn't cause any caching due to the time taken. Might be wrong in that though; its been a while since I have looked through them.

[tomok]

Should also point out that from memory I think the default demo instances from IBM would all be so small in their datasets that they probably wouldn't cause any caching due to the time taken.

You are correct. They won't cache unless you monkey with VMT. The default VMM setting is a joke for any cube of decent size. If you don't have an adequate setting for each cube in your model then they'll hardly ever cache.

[redsince78]

Thanks Guys,

We have looked at the VMT and VMM settings and the base models will cache one view per cube (of tiny data sets) each but no more, despite attempting different full sized views on planning sample. If we change the view/data ranges it should cache a second view but it doesn’t seem to want to

[tomok]

We have looked at the VMT and VMM settings

OK. So you looked at them. What did you change them to?

[redsince78]

Hi

VMT = 1, VMM = between 1,000,000 & 2,000,000 depending on cube just to be safe.

Stored calc cells & memory used for calcs drops off as soon as view is rendered, and it never caches.

Screenshot attached as well

Cheers

[Steve Rowe]

Hi,
Suggest you check the following.

What is the actual behaviour you are seeing on recalculation of the view, i.e. is it slow on 2nd refresh? => The stats cubes are not always reliable, it is hopefully more likely there is a bug in stat collection than in caching. I recently had an issue in 10.2.2 FP5 (I think from mem) where the stats cube were giving me wrong info, you don't mention what release you moved to?

Confirm that this behaviour is not expected, i.e. You have data being loaded that is forcing the cache to be cleared.

Cheers

[redsince78]

Hi Steve,

Second Recalculation results in a full recalculation – effectively no caching of the previous calculation, for any cube in the instance, regardless of the VMM/VMT settings.
There has been no change to the underlying data.
In effect at 0m38 above, the view is rendered and the system appears to dump the memory.

We have been running on 10.2.2 FP6/FP7

[Steve Rowe]

Hi redsince78, (ManU, Communism? We need to know )

Just to confirm explicitly the recalculation of the view takes longer on the second pass and you are not relying on the stats cubes as your only indicator of an issue?

i.e.
Restart server or change data in cube A
Time to calculate big view of cube A= X
Time to re-calculate big view of cube A= Y

X=Y?

Cheers,

[redsince78]

Hi Steve,

Your 25 miles out YNWA !

After a Restart/data change of cube A, X =Y.
In effect a complete recalculation taking ~ the same amount of time.

Also this is a dev environment at the moment so there are no other TM1 systems running, no users etc.