TI and Parallel Interaction

[image2x]

Given IBM's best practice note on PI on TI's:

Parallel Interaction allows for simultaneous execution of TI processes reading and/or writing to the same cube – none of the data loading processes blocking the others – without needing to use Batch Update Start / Batch Update Wait Finish. Remember that TI’s having meta data updates will result in blocking actions.

(1) Is it appropriate to expect to be able to do a full export of a rules-heavy cube with asciioutput to a file AND not have the TI operation block read or writes to the same cube? I have had no luck getting TM1 not to block with even with the simpliest of TI export code (i.e. a using a static source view).

(2) Does the caveat about metadata updates apply to MDX subset/view creation code? My guess/hope is that only true dimension changes/additions block ops.

(3) Functionally speaking, can Rules prevent Parallel Interaction from working?

Thanks again folks.

-- John

[jim wood]

(1) I have noticed a marked improvement in this in 9.5.2 I ran 9.1.4 for many years and aside from the known issues I always found that parralel running was not an option.

(2) I think it refers more to dimension builds. People have posted recently that they have experienced issues while trying to do things while running TI processes that update dimensions.

(3) If you save them or reload them via TI then yes. You may notice slower pefromance mind if you have heavy rule usage.

I hope that helps. Are you running 9.5.2? You didn't mention it in your post.

[image2x]

Yup, 9.5.2. I should update my profile.

So have parallel exports have become the norm for you since 9.5.2?

Regarding rules, I’m wondering if parallel interaction may have practical limitations when rule complexity / data size thresholds are exceeded and reverts to hard lock behavior.

Anyone else have observations on lock status during exports (or TI in general) in 9.5.2?

Thanks.

[David Usherwood]

Few observations:
a PI is new in 9.5.2 and needs an explicit parameter turning it on in the CFG file.
b You need to check out the CubeDependency writeups to get PI working.
c I have found it to _function_ (I'm not - yet - saying function _well_) with pretty complex rules (34 pages enough?) and large datasets (the app I've mentioned a number of times which is too large to calculate fully on a 400gig server - big enough?)
d For this app, with PI on, a copy process, run singly, takes 85 times longer to run with PI turned on than with PI turned off.
e Running two jobs in parallel did not lock but caused memory to rise beyond the available RAM (64gig) of our test box, so we have no final stats.
f IBM are aware of the above.