We deploy a customized .net reporting solution that includes the ability for users to generate a cube and then view it through a grid viewer. Users also have the ability to take a cube they generated and distribute it to other users. Due the nature of the security model in place, a cube is generated for every user. For example, one distributed cube going to 50 users would generate 50 cubes based on the user’s security.
These cubes are very small (2 MB on average) and normally only a 2-3 dimension and 2-3 facts. Cube generation is done through the .net api by generating a data source view to a finished table in the SQL database and processing using MOLAP-regular and process full command.
A client utilizes this ability heavily and now have around 11k cubes stored in one database. The query performance to the cubes is still good but the time related to processing a new or existing cube has gone up significantly. What would take 30 seconds to process now takes 5-6 minutes. After checking a numerous settings to optimize SSAS, I finally backed up the existing SSAS database and replaced it with a blank SSAS database. Processing time went back down to expected levels. So assume I have hit some threshold where the number of objects impacts processing performance.
Based on the object and performance relationship, I think that the answer is less cubes with a different approach to security. This approach however would cause some major code change within the application. We have thought about suggesting the client use a dedicated SSAS server, however I am not confident that would resolve the issue within the scope of our software design approach.
So, I have a few questions I am hoping the community may be able to shed some light on.
1. Why would the processing time increase with the number of cubes in a database?
2. What if any suggestions are there to improve performance in an environment like this?
Hardware specs
Windows Server 2008
SQL server Enterprise Edition 2008 R2 SP2 running db (memory cap 22GB), AS (lower memory 10/Upper memory 30), and RS (1GB memory cap).
2 12 core Intel Xeon CPU
System Memory 48 GB
SAN through Qlogic fibre channel adapter
Other SSAS non default settings
Process\Max Threads 240
Query\Max Threads 48