One of my cubes uses 3 large dimensions. Each dimension has 1 million
members. After the dimensions and cube are processed, MSAS is taking 1.5 GB
memory. Does that mean MSAS would not be able to handle 6 million dimension
members, considering it has only 3GB to play with?

How does MSAS manage its memory when caching dimension members? It seems
MSAS just keeps allocating memory. Does it have some kind of LRU algorithm
to manage limited amount of cache based on memeory limitation?

A future cube of mine would involve a dimension with 9 million members. It
would be interesting to see how MSAS deals with it.

- Yubo Fan

Forgot to mention: all of my dimensions are parent-child dimensions with no
extra member properties.

"Yubo Fan" <yubo.fan (AT) sas (DOT) com> wrote

Hi Yubo,

Analysis Services uses the Very Large Dimension Manager (VLDM) to move
large dimensions out of the Analysis Services Server process space in an
attempt to improve the scalability of Analysis Services. The threshold for
large dimensions is ~64MB, this can be configured in the Registry by
setting the VLDMThreshold value. Any dimension that exceeds the
VLDMThreshold is automatically moved to the VLDM process.

You may check the following article.
http://msdn.microsoft.com/library/de...us/dnolap/html
/olapperf.asp

Memory
Memory consumed during processing includes both the memory used for
processing dimensions, and memory used for processing the facts on a
segment-by-segment basis.

Memory for processing dimensions + Memory for processing facts

When dimensions are processed, the intermediate structures are also kept in
memory. This means the memory requirement is potentially doubled. This can
make a difference when processing large dimensions. Currently, we estimate
taking about 100 bytes per member. This memory requirement can increase
tremendously depending on whether member properties are used and what their
sizes are.

When the fact table is processed, memory is consumed on a
segment-by-segment basis, therefore the maximum memory used would be for
storing a segment's worth of data in memory. As discussed above, changing
the ReadAheadBufferSize and ProcessReadSegmentSize will affect the memory
requirements.


Bill Cheng
Microsoft Online Partner Support

Get Secure! - www.microsoft.com/security
This posting is provided "as is" with no warranties and confers no rights.
--------------------
e.de!news-lei1.dfn.de!news-fra1.dfn.de!npeer.de.kpn-eurorings.net!news-out.n
uthinbutnews.com!propagator2-sterling!In.nntp.be!vienna7.his.com!attws1!ip.a
tt.net!lamb.sas.com!newshost!not-for-mail

Bill,

Thanks for the reply. Here are some follow up questions:

1. Where can I find more information about VLDM?

2. Where's the registry entry you mentioned?

3. If the VLDM threshold is 64MB and each member takes 100 bytes, all of my
large dimensions (1 million member) should have been moved to the VLDM
process. However, why was I seeing the MSAS process (msmdsrv.exe) took 1.5
GB memory after the dimensions were processed? (It was taking 150 MB memory
before those dimensions were processed)

4. Even with the consideration of doubling of the memory usage during
dimension processing. The memory usage seems to indicate that each dimension
member would take 250 bytes.

5. What is the process name for this VLDM process?

Thanks again.

Best regards,

Yubo Fan

"Bill Cheng [MSFT]" <billchng (AT) online (DOT) microsoft.com> wrote