Skip to Content
avatar image
Former Member

SAP HANA - Table size in Main Memory after archiving

Hi all,

I'm working on the optimisation of table size on a "small" SAP ECC running on HANA system. It's a small but old system and I decided to test the impact of archiving (SARA). I get followinf results and have some trouble to understand:

1) Starting point: Table BSEG. Size in Total Memory on Hana: 6 Gb for 100 M of entries

2) I run archiving with ECC transaction SAR and archive/delete 27 M of entries

3) I forced MERGE

4) Table BSEG: Size in Total Memory on Hana increased to 9 Gb for 73 M of entries !!!!

5) I forced "Optimize_compression"

6) 4) Table BSEG: Size in Total Memory on Hana decreased to 3,7 Gb for 73 M of entries !!!!

I thought that doing a MERGE was doing an optimization of compression in the same time.

What happened ? I noticed that when archiving objects some tables gets bigger (KONV, BSEG) in memory and some get smaller (CDPOS).

Thanks for your help.

Lilian

Add comment
10|10000 characters needed characters exceeded

  • Get RSS Feed

1 Answer

  • Nov 04, 2015 at 10:19 AM

    Without seeing the data of the system views and the exact commands you used, it's hard to get conclusive data here.

    What's possible though is this:

    After the archiving run not enough data had been deleted to actually have the MERGE happen.

    That's would explain why after the merge the table size was increased.

    The optimize compression step in fact does not perform the compression but evaluates the data distribution to find the optimal compression settings for all columns in a table.

    Deleting a sizeable part of your tables clearly changes the data distribution in the table (e.g. no more values from before 2005), so there is a benefit in running the optimize compression.

    Generally I would also recommend to look into checking whether hybrid lob usage where possible and to get rid of superfluous concatenate index hidden columns.

    Add comment
    10|10000 characters needed characters exceeded