SSAS Tabular : MDX query goes OutOfMemory for a larger dataset

Question

Hello all,
I am using SSAS 2012 Tabular to build the cube to support the organizational reporting requirements. Right now the server is Windows 2008 x64 with 16GB of Ram installed. I have the following MDX query. What this query does is get the member caption of the
“OrderGroupNumber” non-key attribute as a measure where order group numbers pertain to a specific day and which occurs in specific seconds of a day. As I want to find in which second I have order group numbers, I cross the time dimension’s members with a specific
day and filter the tuples using the transaction count. The transaction count is a non-zero value if an Order Group Number occurs within a specific second of a selected day.
At present “TransactionsInflight].[OrderGroupNumber].[OrderGroupNumber]” has 170+ million members (Potentially this could grow rapidly) and time dimension has 86400 members.
WITH
MEMBER [Measures].[OrderGroupNumber]
AS IIF([Measures].[Transaction Count] > 0, [TransactionsInflight].[OrderGroupNumber].CURRENTMEMBER.MEMBER_CAPTION,
NULL)
SELECT
NON EMPTY{[TransactionsInflight].[OrderGroupNumber].[OrderGroupNumber].MEMBERS}
ON COLUMNS,
{FILTER(([Date].[Calendar Hierarchy].[Date].&[2012-07-05T00:00:00], [Time].[Time].[Time].MEMBERS),
[Measures].[Transaction Count] > 0) } ON
ROWS
FROM [OrgDataCube]
WHERE [Measures].[OrderGroupNumber]
After I run this query it reaches to a dead-end and freezes the server (Sometimes SSAS server throws OutOfMemory exception but sometimes it does not). Even though I have 16GB of memory it uses all the memory and doing nothing. I have to do a hard-reset against
the server to get the server online. Even I limit the time members using the “:” range operator still the machine freeze. I have run out of solutions to fine-tune the design. Could you guys provide me some guidelines to optimize this query? I am willing to
do a design change if it is necessary.
Thanks and best regards,
Chandima

Answer 1

Hi Greg,
Finally I found the problem why the query goes out of memory in tabular mode. I guess this information will helpful for others and I am posting my findings.
Some of the non-key attribute columns in the tabular model tables (mainly the tables which form dimensions) do not contain pretty names. So for the non-key attribute columns which I need to provide pretty names I renamed the columns to something else.
For an example, in my date dimension there is a non-key attribute named “DateAltKey”. This is the date column which I am using. As this is not pretty to the client tools I renamed this column as “Date” inside the designer (Dimension
design screen). I deployed the cube, processed the cube and no problem.
Now here comes the fun part. For every table, inside the Tables node (Tabular SSAS Database > Tables) you can view the partition details. You have single partition per dimension table if you do not create extra partitions. I opened the partitions screen
and clicked on the “Edit” icon and performed a Syntax Check. Surprisingly it failed. It complains about the renamed column. It complained “Date” cannot be found in source. So I realized that I cannot simply rename the columns like that.
After that I created calculated columns (with a pretty name) for all the columns which complained and all the source columns to the calculated columns were hid from the client tools. I deployed the cube, processed the cube and performed a
syntax check. No errors and everything were perfect.
I ran the query which gave me trouble and guess what... it executed within 5 seconds. My problem is solved. I really do not know who did this improve the performance but the trick worked for me.
Thanks a lot for your support.
Chandima