ProcessIndexes at Partition or Cube level

Question

Hi,

I am using SSAS 2005 and have a question regarding processing options.

If a cube partitioned by month is processed using an approach of ProcessAdd for Dimensions (I guess this would also require ProcessIndexes for the Dim's??) and then only the most recent partition i.e. the current Month is reprocessed using either a ProcessFull OR a combination of ProcessAdd and ProcessIndexes; do the other partitions and their bitmap indexes remain unaffected or also require ProcessIndexes in which case it would be necessary/easier to ProcessIndexes at cube level?

Thanks

Craig

Answer 1

Hi Craig

If you are only doing a ProcessAdd to the dimension you shouldn't need to do an additional Processing in that regard.

For the partitions, if you do a process full you wont need to do any additional processing, although it is generally considered better to do a ProcessData and ProcessIndex instead of ProcessFull. If you do a ProcessAdd I don't believe you need to do a ProcessIndex afterwards but if you do it'll only be on the partition.

Hope that helps.

Answer 2

Hi Craig

If you are only doing a ProcessAdd to the dimension you shouldn't need to do an additional Processing in that regard.

For the partitions, if you do a process full you wont need to do any additional processing, although it is generally considered better to do a ProcessData and ProcessIndex instead of ProcessFull. If you do a ProcessAdd I don't believe you need to do a ProcessIndex afterwards but if you do it'll only be on the partition.

Hope that helps.

Hi Michael,

Thanks for the quick feedback.

I understand performing a ProcessAdd doesn't affect existing data (partitions or dim), however if you were to omit ProcessIndexes would the bitmap indexes therefore not include the newly added data and in the case of the partition the aggregations also be out-of-date?  If this is the case I assume this some performance degredation would be incurred until ProcessIndexes was performed on the affected objects?

Thanks

Craig

Answer 3

I think if you do a process add on the dimension, it should only add items so there will be no new items in the index of existing partitions. I guess you are correct though, in where you do a processAdd to the partition, you will need to processindex for that partition to update indexes in that measure group to include the new data.

I guess the question is how long you have to do the processing, verse the degradation in performance that will be permissible?

Answer 4

That clears up my question, guess the only thing left to do is test the performance!

Thanks for your help Michael.