LSMW - Very Slow

[Former Member]

Hi All

We are using LSMW to update function locations in SAP PM using LSMW and Batch Input Sessions,

When we upload the data using tx LSMW we are updating at approx 3 records per minute.

The file that needs to be uploaded has approx 53 000 records, hence it would take days to upload this and furthermore we have another 9 files of similar size to upload using LSMW.

We have also switched off table logging (rec/client=ALL) and switch off archive logging to enhance performance.

CPU's are 99 -100 % idle, no paging, memory is 20 out 32GB beig used, basically the system is performing the way it is supposed to but the uploads via LSMW are way to slow.

Are there any more Basis Settings that we can probably tweak / change / edit to get more performance on the LSMW uploads to run quicker.

Please help, we are on ERP 6.0, Oracle, and Solaris using LSMW Version 4.0.0

Thanks

Solly

[Former Member · Accepted solution]

Solly,

First thing to do is run txn STAD and to determine where the response time is being spent.

Then choose the proper analysis tool:

1) If DB time high => SQL trace with ST05

2) If CPU time high => ABAP runtime analysis (SE30) => choose "execute in parallel session"

3) If "processing time" high but CPU time low => other cause (not probable given that you server seems far from overloaded)

Regards,

Mark

[Former Member]

I feel silly for stating the obvious, but if you want to increase your throughput you should look into parallel processing. I.e. split your input file into smaller chunks and process them in parallel (ideally via some convenient wrapper program that does the trick for you).

[Former Member]

Hello All

Thanks for all the advice, Mark was right, check the stats out and we found that the bottleneck was the database.

The LSMW runs have a V2 update component to as well as using the normal update work processes, additionally we done a db update stats and swtiched to a very heavy batch, update and update2 type operation mode, reducing dialog processes at the same time.

We then ran in parallel alot of lsmw jobs and monitored it closely while extending tablespaces etc,

We are currently processing about +- 175 records a minute which roughly translates up to max 2 records a second.

System idle is averaging around 96% and peaking at 92% idle.

We are already 70% done with the uploads and are predicting to be done in the next 3 days.

We are also running this on the QA instance, the production instance is an M4000 so we are expecting better throughput.

We think this is acceptable for us, do you guys have any similar stats that we could compare against.

I am sure with alot more tweaking especially in the oracle shared mem etc we could better throughput.

Thanks

Suleman

[Former Member]

Hi,

Run LSMW, open a new session with tcode SM50 check the table column and revert back

Regards

[Former Member]

additionally we done a db update stats

Yes! I have seen an troughput increase +10 times faster, after update statistics. The problem here is, that if some tables where empty (no lsmw used so far), then you experience expensive access paths. So after (or during) the first lsmw run, you should update stats.

Cheers Michael

[Former Member]

I would check how is I/O on disk and if Disks and bus speed to disk is fast enough. and Data files. Are no of Datafiles and distribution optimal?

Adding Battery backed memory cache to hardware will improve performace by 3 fold if System is workng fine for normal operations other than LSMW.