Model Optimisation

[Eric]

I was reading something online and it went right over my head. Anyone know what this means and more importantly how to change and check it?

Finally two start-up parameters, VMM and VMT control how TM1 cache’s views. Caching provides performance benefit but absorbs RAM. Tuning these parameters can provide additional available RAM. Again the trade-off for performance needs to be considered and is site-specific.

Srouce:http://www.spfplus.com/newsletter03.html#tm1

[Olivier]

Hello Eric,

Try to have a look in this Cube : prod:}CubeProperties
(you have to tick "display control object" in your server explorer view menu)

There are several items for cube settings including the two Items VMT and for VMM.

I tryed to tweak the settings for VMM and VMT but never really successfully sized the impact of changing this setting.
If you check the manual about stargate views, you will find out the default value for VMM and VMT is left blank.


Hope this helps,


Olivier

[Steve Vincent]

It's in the PDF Administrators Guide, which should have been installed with TM1, probably in the directory C:\Program Files\Applix\Manuals\eng or something similar.

I can't copy / paste from the PDF (scrambling characters...) but the basics are VMM is the amount of RAM per cube RESERVED for stargate views, and VMT is the amount of time a cube view is allowed to calculate before it starts to commit the data to a stargate. Default VMM is 65,536 bytes, default VMT is 5 seconds. I'm guessing that if you increase the VMM on certain high useage, high RAM cubes you should see a performance increase as long as the data is not changing. You'd need to ensure the server has enough RAM to deal with it tho - if you tell it to reserve 3GB for a cube but don't have 3GB available, it's gonna kill your server and it won't even start up.

Might have a play about with those on the dev server and see if i can get some better recalculation on one of our bigger cubes, thanks for bringing it up tho - never knew it existed!

[Steve Vincent]

Did some testing on this yesterday and didn't see any difference what so ever. Set 2 cubes to 50MB and their sizes on server startup were identical, and after calculating their default views they were also identical both with and without the parameters. Either i've misunderstood what they are trying to achieve, or they aren't doign what they are supposed to be. Hmmm..

[Olivier]

I am happy with your comment because now i am not alone to see no effect throught those settings VMM and VMT configuration...

If anyone already notice any impact please post a short comment on how to test it ?

regards,

[jim wood]

I was taking about with Cognos Support. We could possibly see some benfit as we have some very large cubes. I think your cubes have to be a decent size to feel any benefit. If you set it 50MB and your average view is smaller than this then you will see no difference. Also bare in mind that once you calculate the vuiew it is cached. The only way to track real benefit is to calculte a view, timwe it, then change the setting, restart the server and then track if the new view takes any longer,

Jim.

[Olivier]

Thanks for the comment Jim !

I will try to test it on a large cube with this procedure and see if i can track any changes...

Olivier

[bihints.com]

Also bare in mind that once you calculate the vuiew it is cached. The only way to track real benefit is to calculte a view, timwe it, then change the setting, restart the server and then track if the new view takes any longer,

Isn't unloading just that cube enough to free up the memory of its cached views?

[Steve Vincent]

Yes. I tried both and saw no difference anyway...