Hi,

We have cubes with multiple partitions with significant volumes (>10 billion
fact rows with a dimension of >1 million members). We are developing the
cube
on 64-bit Analysis Services Enterprise Edition. We have a 32-bit Analysis
Services with SP3 configured exactly the same as the 64-bit to process the
cubes as per
http://www.microsoft.com/technet/tre...hnet/prodtechn
ol/sql/evaluate/AnSvcs64.asp. Unfortunately, we haven't been able to get the
results as indicated by the statement "However, when you process the
resulting dimensions, cubes, and partitions (the "runtime" environment), the
actual database access is being executed by the msmdsrv service on the
64-bit system.". Instead we get the following:

* Shared dimensions are processed on 64-bit
* Cubes and partitions are processed on 32-bit (we wan't this to also be
processed on 64-bit).

When using Performance Monitor we are finding that only shared dimensions
are "processed" on the 64-bit computer (ie A > B, C is high). However, all
the cubes and partitions are "processed" using resources of the 32-bit
computer (ie B > A, C is zero or low) .

A. Computer = 64-bit, Object = Processor, Counter = % Processor Time,
Instance = _Total
B. Computer = 32-bit, Object = Processor, Counter = % Processor Time,
Instance = _Total
C. Computer = 64-bit, Object = Process, Counter = % Processor Time, Instance
= msmdsrv
D. Computer = 32-bit, Object = Process, Counter = % Processor Time, Instance
= msmdsrv

Tests performed:
Test1. Using Analysis Manager on 32-bit, right-click Shared Dimensions
folder and click Process All Dimensions. Result: A > B.
Test2. Using Analysis Manager on 32-bit, right-click Cubes folder and click
Process All Dimensions. Result: B > A.
Test3. Using Analysis Manager on 32-bit, right-click database, and click
Process the Database. Result: A > B when processing shared dimension, then B
the OLAP Processing Task to process the database. Result: A > B when
processing shared dimension, then B > A when processing cubes and
dimensions.
Test5. On the command line on 32-bit, execute the same DTS package using
dtsrun. Result: A > B when processing shared dimension, then B > A when
processing cubes and dimensions.

What can we do to get cubes and partions processed on 64-bit?
We will eventually use a custom application written in VB (using DSO) to
process multiple partitions in parallel.

Thanks
James Yeo
jameskyeoAThotmailDOTcom

James,

The client application is not capable of doing the actual processing work.
If you have cubes on a 64-bit server then they can only be processed on a
64-bit server. The client application (DSO) is not doing the
processing....it is just sending commands to the server to do the
processing.

That said, what you may have is that the source data from your cube is
sitting on a 32-bit SQL Server. If that is the case, then you will see
activity on the 32-bit server as the data is being selected during the
processing phase.

But what you're describing below in that the 32-bit machine is doing the
cube processing is not possible. If you have source data on the 32-bit box,
the database will use resources. But once the data is sent to the Analysis
Server, it will be Analysis Services on the 64-bit machine that processes
the cubes as long as they are defined on the 64-bit server. You probably
won't see high CPU utilization during the read phase for a single partition,
but during the aggregation phase, you should see CPU utilization increase on
the 64-bit box.

Sean


--
Sean Boon
SQL Server BI Product Unit

--
This posting is provided "AS IS" with no warranties, and confers no rights.
Use of included script samples are subject to the terms specified at
http://www.microsoft.com/info/cpyright.htm.



"James Yeo" <jameskyeo_AT_hotmail_DOT_com> wrote

Sean,

You are right. The peaking on the 32-bit machine is due to the sending of
the commands to the 64-bit server to do the
processing. We have since been able to peak the 64-bit (previously no rows
were being processed due to a join problem). Source data is from Oracle on
Linux which is another computer altogether.

Can you suggest the best set of performance counters to clearly distinct the
processing phase and aggregation phase of the cube processing?

We've used Analysis Services Proc Aggs object. The Proc Aggs object counters
provide information about processing aggregations.Proc Aggs object counters
Description
Current partitions Current number of partitions being processed.


but it is always showing zero (0).

Thanks.

James


"Sean Boon [MS]" <seanboon (AT) online (DOT) microsoft.com> wrote

There could be a bug with that particular perfmon counter. I'd have to try
to repro it in order to find out for sure.

Other counters that you can watch should include Rows Read/second on the
proc object and for processing aggregations you could monitor rows
created/sec. There is a known issue with the temp file bytes written
counter. If you need to monitor temp file bytes written you can use a tool
from sysinternals.com called filemon.

Sean


--
Sean

--
Sean Boon
SQL Server BI Product Unit

--
This posting is provided "AS IS" with no warranties, and confers no rights.
Use of included script samples are subject to the terms specified at
http://www.microsoft.com/info/cpyright.htm.

"James Yeo" <jameskyeoAThotmailDOTcom> wrote