TM1 Forum

I have been playing around with dimension order and was wondering how the SYSTEM optimization option actually works. I did not see anything in the docs about it (I could be blind)

My main questions are

1) How is it optimizing? Does it just use the standard rules that they give as guidelines? Or does it evaluate the options for best performance? I know from testing that it does not evaluate it based on final cube size.

2) How often does this optimization occur? When meta data is added? When the system loads? Only when you go back to the reorder screen?

3) Probably based on the answer to number 2, but is there any noticeable performance issues/locking associated with the choice of SYSTEM optimization?

Hi George:

My assumption is that the System optimization option is just a best guess at dimension order based on some algorithm that isn't currently published by IBM. This is probably not unlike Cubewise's Vizier product, which I think had a similar feature and its own algorithms for arriving at a more optimal cube dimension order. Based on how fast TM1's native System re-order comes back (and the fact that it doesn't come up with the smallest cube footprint always), I'm not sure how thorough/robust its algorithms are.

Optimization only occurs when you go into the cube optimizer, select a different dimension order, and hit OK. In other words, if you think selecting System will cause TM1 to regularly examine and re-optimize automatically for you as data populations in the cube change then I don't think that's the case or intention of the option currently, but I could be wrong. I really think this is just a shortcut option to save you manually testing different dimension order selections manually and allowing you to rely on its own algorithms.

The above probably answers your 3rd question - I would assume performance/locking is an issue only when testing a different dimension order (where TM1 will be creating a copy of the alternatively-structured cube in RAM - note the additional RAM overhead that could be needed for this) and when committing to a modified dimension order for that cube when hitting "OK". Certainly this isn't the kind of thing where you want to be trying out many different dimension orders in a live production environment.

Hope that helps.

Regards,
Mike

Thanks.

Those were the answers I was coming to in testing but wanted to get some external confirmation since the option does not seem to be explained anywhere.

It's always looked to me as if the System order is simply smallest to largest (with Measures last). I haven't noticed it attempting to show any intelligence as to sparse and dense. As a consequence, I never bother using it any more and use the old grey matter instead.