AGGREGATE detailed log file POUTFILEUNIT?

Question

Hi,
While trying to tune the performance of a "one variable, 7 dimensions" OLAP cube aggregation with the AGGREGATE command, I've set the POUTFILEUNIT option to a file to see what was happening with the aggregation.
Beside generating a huge (6 MB) log file, I can't do much with it because I cannot find any information regarding specific labels such as "ctuphiers" as in:
11:31:14.951 [   ctuphiers] [0] 279022
11:32:52.213 [   ctuphiers] [382] 8519917
or
various informations on other commands such as "parents" in [Aggregate]:
16:02:43.909 [   Aggregate] aggregate A7_VARIABLE using A7_AGGMAP over CMP_PF_TMPS op SUM
16:02:43.909 [     AgWlist] Generating worklists
16:02:43.909 [     AgWlist] start=0 end=47 len=47
16:02:44.409 [     AgWlist] done
16:02:44.409 [   Aggregate] wlstats worklist=1 parents=33630
My question: Where can I find any documentation about this detailed information in the logs so I can better tune my - way too long - aggregations ?
As you can figure out in the previous examples, we are using a named COMPOSITE, CMP_PF_TMPS. I've played with the definition of it (by adding / removing dimensions) while recreating AWs. I also modified the AGGMAPs definitions but found on Metalink bug reports about AGGREGATE with AGGMAPs composed of 4+ RELATION commands.
Many thanks in advance for any insight,
Pierre
P.S.: We are using Oracle 9i OLAP v.9.2.0.4.1 on Win2K Server.

Answer 1

Pierre, there's no specific documentation on POUTFILEUNIT because it's an unsupported debugging aid - there's no guarantee that the output won't change in the future (in fact, it changes probably at least with every major release). It's mostly meant to be used when working in conjunction with a support person who can engage development if they need. Specifically, it does NOT provide the sort of output that makes it easy to guess steps to take to improve performance unless you have a good understanding of both OLAP in general and the entire AGGREGATE system.
It's hard to provide specific advice given the information; I can tell you that the line
11:32:52.213 [ ctuphiers] [382] 8519917
says that at that point in time the composite has grown to 8.5 million composite tuples, which is a fairly good size composite, but certainly not unheard of. I can also tell you that worklists are sets of parent tuples which don't depend on each other, the line
16:02:44.409 [ Aggregate] wlstats worklist=1 parents=33630
means that for the first set of independant parents to be rolled up consists of 33630 tuples.
But as I said before, if you're looking for a magic line that says "you're doing this wrong," you're not going to find it (that's not what POUTFILEUNIT does). Rather than trying to learn to read the log, if you're having problems with performance, could you go into more detail about your situation? How your variable is set up, what your aggmap looks like, input data volumes, how long your job is currently taking and what your needed window is, etc, are all helpful to debug performance problems. The most common "beginner" problems we see would probably be
o Make sure you have a dense, fast varying dimension (ie, not all of your dimensions in the composite)
o Use dynamic calculation to only materialize part of the cube
o If you're using dynamic calculation, make sure AGGINDEX OFF is set in your aggmap
Whatever bug report you found about aggregate with > 4 relations should also mention that it is fixed by the 9.2.0.4.1 patch (this is only a problem with vanilla 9.2.0.4.0). Assuming you have the patch, there is no reason to limit yourself to 4 relations.
Feel free to follow up with more questions, hope this helps.