Some say that partitioning give me performance some say it does not -
what is the truth?

Example:
AS2005: 1 year of transactions. Partitioned by weeks - 52 in all. The
whole cube is loaded into ram.

Would a MDX querying filtered down to a single week perform better when
partitioning is implemented?
I guess if data were on disk and not in ram, the load time would be
considerable less. But when everything is already in ram this should
not be an issue.

A simple test did not show any performance gain. Is that what I should
expect?

Thanks
JE

Queries for a specific set of weeks that miss the in-memory caches will
benefit from partitioning.

Thanks,
Akshai
--
This posting is provided "AS IS" with no warranties, and confers no rights
Please do not send email directly to this alias. This alias is for newsgroup
purposes only.

"jellegaa" <jesper.ellegaard (AT) gmail (DOT) com> wrote

Thanks, Does that mean in case of disk access? (I'm not sure on when I
hit the cache and when I don't)

/JE

We use many partitions with SSAS 2000.
I think it must be the same with SSAS 2005.

Using partitions is a good solution if you have large cube : you can
just process the partition with new data : in your case, just one week
and not the full year.

If you use partitions, something is very important : you must define
Slice for each partition. With slice, the engine access directly to the
partition storing the data. If there is no slice, the engine scans all
the partitions to read the data.

One information more : if you define weekly partitions it isn't worth
to create aggregation at the year level because they will store the
same data than weekly aggragations. If you query the year, the engine
will scan all the partitions to dynamically aggregate the data.

Yes, disk access would be reduced because some partitions would not have any
data for the current slice of the cube. Only partitions for the weeks that
are in the current query would need to be scanned...

Thanks,
Akshai

--
This posting is provided "AS IS" with no warranties, and confers no rights
Please do not send email directly to this alias. This alias is for newsgroup
purposes only.

"jellegaa" <jesper.ellegaard (AT) gmail (DOT) com> wrote

Thanks both for the tips,
but still I'm not sure if I can get performance boost with partitions
if all cube data always is in ram. Do you have any opinion on this?

/JE

If all the cube data is always in RAM then partitions will not help query
performance. Startup queries though will obviously not be in RAM and they
may benefit a little bit...

But generally don't expect perf benefits if the data is small enough to fit
in RAM...

Thanks,
Akshai
--
This posting is provided "AS IS" with no warranties, and confers no rights
Please do not send email directly to this alias. This alias is for newsgroup
purposes only.

"jellegaa" <jesper.ellegaard (AT) gmail (DOT) com> wrote