Slow Aggs in new 11.1.2.4 environment

Slow Aggs in new 11.1.2.4 environment

JS
JS

January 11th, 2017, 8:54 pm #1

This is kind of a vague question I know, but here goes.

We are in the middle of an upgrade from 11.1.2.3.500 to 11.1.2.4 (w/ .013) patch. We have a few large cubes, one at almost 100 GB and one around 50GB. Both of these are actually planning applications. We have noticed that our aggregations are significantly slower in the new .4 environment than they were in .3 One cube has a daily calc takes roughly 15 minutes in .3 but is taking almost 45 in .4. The other cube has a very long nightly process that takes roughly 8 hours in .3 but is taking about 24 hours in .4.

All things being equal, outline, db settings, config file settings, etc…I can’t understand why this is so much different. The new hardware is “supposed” to be better.

Has anyone run across this issue? Any ideas on what else I can check other than adjusting my cache settings for the billionth time?

Thanks.
Quote
Share

Pete
Pete

January 12th, 2017, 2:00 am #2

Could it be parallel calculation \ essbase.cfg settings?

I'm guessing you've checked all that already though.

That's a massive change to be explained by hardware (in either direction)

Raised an SR?
Quote
Share

JS
JS

January 12th, 2017, 2:30 am #3

Yeah, I submitted and SR yesterday.

There's just a lot of inconsistency. I have an allocation script that I ran manually multiple times today and it took about 20 minutes each time. That same script is run as part of a sequence that I kicked off and it took over 2 hours. I ran it again this evening and it took an hour.
Quote
Share

Joined: March 4th, 2014, 5:57 am

January 12th, 2017, 11:21 am #4

latest essbase work diferently in parallel calcualtion (if u use calcparallel ) .
Now u need to move up "non agg" dimension from bottom outline
Quote
Like
Share

Joined: April 13th, 2011, 5:00 pm

January 12th, 2017, 2:06 pm #5

Yeah, I submitted and SR yesterday.

There's just a lot of inconsistency. I have an allocation script that I ran manually multiple times today and it took about 20 minutes each time. That same script is run as part of a sequence that I kicked off and it took over 2 hours. I ran it again this evening and it took an hour.
Are you running them as business rules? I wonder if you are hitting a timeout setting that is causing the business rules to be re-launched?
Quote
Like
Share

JS
JS

January 12th, 2017, 4:25 pm #6

Javier - Yes I am running them as business rules. If I hit a timeout, would that be logged somewhere? One thing I thought may have been an issue is running the individual rule vs running a ruleset. When I ran the rule individually (from Business rules in Planning) it was running pretty consistently around 20 minutes. When I execute a sequence which calls a ruleset that includes that rule, it was taking longer. But last night I ran the individual rule again and it took over an hour.

ER - that's interesting and I'm testing that now. I have updated my outline to a traditional hourglass, related the data and agg'd. I'm rerunning the calc now which has been running for 40 minutes or so. I will have to do some additional testing with new calc parallel and calctaskdims settings based on thew new outline structure.
Quote
Share

Joined: April 13th, 2011, 5:00 pm

January 12th, 2017, 5:09 pm #7

Check the Job Console. If you see your BR listed multiple times (with runs that are yours but aren't) that's probably it.
Quote
Like
Share

Joined: November 16th, 2003, 11:27 pm

January 12th, 2017, 5:58 pm #8

This is kind of a vague question I know, but here goes.

We are in the middle of an upgrade from 11.1.2.3.500 to 11.1.2.4 (w/ .013) patch. We have a few large cubes, one at almost 100 GB and one around 50GB. Both of these are actually planning applications. We have noticed that our aggregations are significantly slower in the new .4 environment than they were in .3 One cube has a daily calc takes roughly 15 minutes in .3 but is taking almost 45 in .4. The other cube has a very long nightly process that takes roughly 8 hours in .3 but is taking about 24 hours in .4.

All things being equal, outline, db settings, config file settings, etc…I can’t understand why this is so much different. The new hardware is “supposed” to be better.

Has anyone run across this issue? Any ideas on what else I can check other than adjusting my cache settings for the billionth time?

Thanks.
There were architectural changes between 11.1.2.3.500 and 11.1.2.4 that made cube performance change. We had to go back and retune most of our apps. In some cases we had to change Stored and Dynamic dimension settings. Where in the past stored was always better, we found a few cases where a dimension as dynamic was better. You may need to play around with them a bit. Ultimately we got better performance when tuning was complete, but lift and shift was not going to work.
Quote
Like
Share


Confirmation of reply: