10.2 Dynamic Cubes

[Kracky]

Good afternoon folks - I didn't see any threads covering Dynamic Cubes.  I'm trying to do a POC using Dynamic Cubes in 10.2  I've run the standard sample cube and it works fine.  I've modeled a new cube off some heavily masked customer data and deployed it to my environment.  When I start the cube, the status changes to "partially available" (not the full available like the standard sample cube) and then one of my two Java processes runs away with all the server memory and the CPUs hold steady at about 80-90% utilization, effectively locking up the whole environment.  I even had my hardware guy add 6 more CPUs (total of 8) and get the same results.  My cube isn't that big (per IBM's statement that this thing can handle terabytes of data):  fact table - 5.1 million recs, product dim - 143 recs, store dim - 58 recs, customer dim - 550k recs, calendar dim - 4,100 recs (2002 to present).  I found the below link on IBM's tech site (not very encouraging for a new product...).  Anyone out there have any experience yet with this?  Thank you!!     http://pic.dhe.ibm.com/infocenter/cbi/v10r2m0/index.jsp?topic=%2Fcom.ibm.swg.ba.cognos.rdm_cog_rlp.10.2.0.doc%2Fc_rdm_cog_rlp_releasenotes.html

[pumccg]

Have you seen the Dynamic Cubes Redbook? It provides a lot of additional information, as does the Dynamic Cubes User Guide.

Also, is your data truly in physical star schema form? I believe this is a strict requirement for Dynamic Cubes.

[Kracky]

Hey pumccg - yes, it's definitely in star schema (built it with data manager).  I have a copy of the redbook.  I'll be combing through the troubleshooting section today.  I was just curious if anyone else has been building dynamic cubes yet and what their experience has been.  I found it odd that they have a bug in the JRE requiring a patch/new version.  Correct me if I'm wrong but the JRE comes with the BI server 10.2 software itself, no?  Thank you for your input!

[pumccg]

Cognos do supply a JRE, and it probably had the same security vulnerabilities in it that prompted the many recent updates to Oracle's JRE.

I've built a few dynamic cubes from first principles, and merged them into a virtual cube etc as well, but only using Cognos' sample data. I then used a trigger file to run some standard reports to populate the data cache, and created some in-memory summary tables. Haven't done the in-database summary tables as yet. Even though the data set used is not huge, the improvement in response was most impressive. Can't think what might be causing your problem.

[Kracky]

A follow up for anyone interested. 

The problem I was having with the dynamic cube I created (not the standard sample) was the environment JVM heap size was too small and there wasn't enough memory to load the full cube.  Typically, this is set to 1024 MB.  With dynamic query mode (and dynamic cubes), the memory allotment needs to be upped to take advantage of DQM and also to accommodate the size of the dynamic cube(s).

This article covers the topic:   http://www-01.ibm.com/support/docview.wss?uid=swg21587457

The problem I have with this is trying to predict how much memory a cube will require.  The link above has a sub-link to a sizing recommendation.  In my opinion, it's almost easier to use trial and error than work through all the criteria they mention!

If anyone else is going down this road, I'd love to hear about your experiences. 

Cheers!

[pumccg]

Excellent feedback!

Dynamic Cubes are not set and forget. There are lots of tuning issues in the properties of each DC which need to be tweaked to ensure you provide sufficient resources for the DC to perform to its potential. You also need to monitor them to ensure they are actually using the various caches correctly, and not hitting the database all the time.

[jbertrand]

Hi. Just to let you all know, I had the same problem on 10.2. Cube tried to start, but java.exe just sat there at 90%. Increasing the maximum heap size to 8GB in administration did the trick. I would think there would be some error messages or indicator to say "too little memory allocated" or something. Doesn't seem to be the case  :(