Calculations competing for resources when they run in parallel

Question

I have an allocation process where I run 5 different calcs in 5 different
cubes (With exactly same structure and settings) to speed up my process.
-None of them are using CALCPARELLEL
-We have 6 CPU's with 4 cores each.(Total 24 processors)
-Enough RAM for 5 cubes
- AIX 64 bit OS
Even though I have adequate resources to run the calcs in parallel, For some
reason calcs are waiting for other to complete. 5 calcs are starting at same
time, which are almost should take same time based on what they are
calculating, but they seem like waiting for some resources to free up. Run
times for these calcs are very random, for Ex: If calc1 takes 1 hour calc5
takes 20 mins in first test, Calc 1 takes 20 mins and Calc 5 takes 1 hour in
the next test.
Did anybody see this scenario before? Any idea on what resource the calcs
are competing for? Please shed some light on it.
Thanks,
SR.

Answer 1

OK Let me first confirm some things from your previous emails.  When you say ASO Calcs you do mean "custom calcs" (why not post your maxl scripts) ?  Hopefully you are not referring to aggregations as calcs.
You say you are doing calcs on multiple cubes concurrently - please post the script
You say you are loading from BSO calcing and then using a report to extract from ASO.  Are you doing and aggregation after the data load?  Remember that when aggregations run the whole cube is "frozen" and the agg is given exclusive use of the cube.  I suspect that you should have no aggregations defined for these cubes.
Finally, you say you are doing multiple loads/calcs/extracts in each of your 5 cubes.  How - post the scripts please.  You have read the section of the dbag below regarding load buffers and custom calcs right?
Understanding Data Load Buffers for Custom Calculations and Allocations
When performing allocations or custom calculations on an aggregate storage database, Essbase uses temporary data load buffers. If there are insufficient resources in the aggregate storage cache to create the data load buffers, Essbase waits until resources are available.
Multiple data load buffers can exist on a single aggregate storage database. The data load buffers that Essbase creates for allocations and custom calculations are not configurable. You can, however, configure the data load buffers that you create for data loads and postings.
If you want to perform allocations and custom calculations concurrently with data loads and postings, set the resource usage for the data load buffers that you create for data loads and postings to a maximum of 0.8 (80%). The lower you set the resource usage setting, the greater the number of allocations and custom calculations that can run concurrently with data loads and postings. You can also configure the amount of time Essbase waits for resources to become available in order to process load buffer operations.
To configure data load buffers, use the alter database MaxL statement with the initialize load_buffer grammar and ASOLOADBUFFERWAIT configuration setting.
I question whether this section of the documentation is correct.  I would think that a calculation would need exclusive access to the cube otherwise it would be calculating against a moving target.   I will try to reach out to someone at Oracle to ask the question.