Impact of compression on BWA

[murali_balreddy2]

I am trying to understand the impact of infocube compression on BWA index.

When an infocube is in BWA, I have read that compression is not really needed on a infocube at the BW layer as the index is in a compressed form in BWA. Also due to this fact, physical parition by time period does not add any value.

But, if the cube index is compressed, then why do we have the test "Compare Size of Fact Tables with Fact Index". The SAP document says

The following circumstances can result in differences in the numbers of records:
uDBC4uDCBC The InfoCube was compressed after the BW accelerator index was built.
Since the BW accelerator index is not compressed, it may contain more
records than the InfoCube.
uDBC4uDCBC Requests were deleted from the InfoCube after the BW accelerator index was
built. The requests are deleted from the BWA index in the package dimension
only. The records in the fact index are therefore no longer referenced and no
longer taken into account when the query is executed; however, they are not
deleted.

So does that mean,

1) the records in the index match the F and E fact table count at the time of indexing.

2) compression of infocube is always better before indexing to keep the index size small

3) physical parition by time period adds value

Thanks,

Murali

[Former Member]

Hi,

Generally the BIA indexes are filled after the data has been compressed.

Regards,

Rahul

[Former Member]

Murali, you can take a look at this link for bit more info as well

http://help.sap.com/saphelp_nw73/helpdata/en/4b/ec95e2a084324ae10000000a15822b/content.htm

Cheers

Tansu

[Vitaliy-R]

There is lots of confusion around "compression" and "index" in the context of BWA. I think it was discussed few times on this forum as well.

> 1) the records in the index match the F and E fact table count at the time of indexing. 

Correct.

> 2) compression of infocube is always better before indexing to keep the index size small

If you do not see the need for request-based deletion of the data, then this is correct.

> 3) physical parition by time period adds value

It is pretty general statement. Please note that not all underlying databases in BW do physical partitioning of E table. F->E compression of fact table can add as well to performance of initial indexing into BWA, as parallelization of the process is based on time characteristic which is usually better than package dimension used for paralellization of F-table indexing.

As with everything, decissions should be based on analysis and thinking, rather than on "best practices".

Cheers,

-Vitaliy