SPM2.1/SP04: Performance: Cache Settings

Question

In BW > t-code RSRT2
We have Cache Parameter set as
Local Size MB = 100
Global Size MB = 200
Do you have a recommendation on what it should be from a perf perspective.
In Perfomance white paper there is a ref on page-5 u201CApplication Specific Performance Considerationsu201D > SPM Simple Cacheu2026 and I am assuming it is the same as what we do via RSRT2 in BW.
Thanks

Answer 1

Hi Donald
For me the question is not what the actual size of cache is but is the cache effective for your purposes.
I know that your implementation involves large number of concurrent analytical users AND huge volumes of data. So the cache will be big. This SPM cache works simply on the basis of caching the result set of any report. If another report (with the same dimensions, measures, filters, conditions and user locale) is executed then this cache is used and the data is pulled from the cache instead of going back to the BW OLAP engine. If most of your users are actually using the ad-hoc capabilities more than commonalized set of reports, then it is quite possible that instead of the cache aiding performance, it is hindering it. Reason being that all reports in this scenario are unique and the overhead of write to the cache + overhead of checking the cache before passing over the request to OLAP will slow down performance and result in a huge cache size. On the other hand of ther eis a lot of 'reusability' of reports, then even though the cache size is large, is actually beneficial because the end user gets the faster response time.
Now the real question is, do all of your users do the same centralized reports (with a smaller percentage of ad-hoc analysis) or do all of them fire unique requests to the OLAP?
If the answer is the latter and all of the requests are unique from each other, then you actually might want to consider turning off the SPM cache completely. That would save the disk space and actually improve the performance.
Regards
Rajesh