Performance Issue with one dialog instance

[Former Member]

Hi All,

Wer are facing very serious performance issues with one of our apps server.......at the time when system was behaving slow ....I am sharing the various parameter values -:

app server is on windows platform

CPU

User %: 9

system%: 1

idle%: 89

system calls 9113

Interrupts: 528

memory

physical mem avail: 16604440 kb

page in/s 0

page out/s 15

SWAP

COMMIT CHARGE LIMIT (kb) 65145060

COMMIT CHARGE FREE (kb) 32083152

max swap space :49152000

actual swap space: 49152000

LAN

Packets in/s : 2140

Packets out/s : 1507

At this point of time report which was taking max dialog response time:

ZLOGTRPL :90268 (ms) dialog response time

There is there anything which could be the cause of slowness.......

but same time overall dialog response was showing around >95227(approx)

idealy this value should be less that 1200

In st02:

swap for export/import is coming in red mark with value > 100000

Regards,

Prashant

[pavan_kumargali · Accepted solution]

Hi Prashant,

RED marked swaps in ST02, which means swapping is high in those buffers. You must need to increase the memory of only those buffers (Which are highlighted RED in ST02) to 10% more (As a guideline).

However free memory is very less KB 226,548. (Also try increasing the Physical memory if possible.)

So you should be optimally increase the size of those buffers.

For any other information, feel free to reply me.

Thanks and Regards,

Pavan Kumar Gali.

[former_member251078]

Hello Prashant,

I believe that you are facing performance issue with the transaction/report ZLOGTRPL? Is it so ??

Are you facing performance issue only in one application server? How many application servers are in your landscape ?? Are you using load balancing between application servers if you have more than one instance in your landscape ??

If you are facing only problem with ZLOGTRPL, i would suggest you to take ST12 trace and find out where exactly time is being spent. If it is high CPU time, then problem with the coding.. contact your developer to fine tune the statement ?? if the database time is high, then you have tune the expensive sql statement.

Yours Sincerely

Dileep

[Former Member]

Hi Dilip,

We have Hetrogeneous landscape, CI is on HP UNIX platform and all four Apps servers are on windows NT platform,yes we are using load balancing.As you have suggested us we will try to analyze the Zcodes using ST12 and will let you know the results.

Regards,

Prashant

[Former Member]

What is your DB? SAP has recenetly recommended some new Oracle 10g parameters for HP - UX OS. If your DB is oracle 10g then implement these parameters for better performance.

HPUX_SCHED_NOAGE = 178 (without RAC)

DBBLOCK_NUMA = 1 (see Note 1225732)

ENABLENUMA_OPTIMIZATION = FALSE (see Note 1225732)

[Former Member]

Hi All,

Last weekend we restarted the production server since then we are not facing any performance related issue , I think prob lies with swaps in ST02 . After the server reboot values in ST02 became zero and system started behaving fine. I think we need to increase the buffer values in corresponding red marked parameters. But when I checked those parameters values are far higher than the recommended SAP values.What should I do now?

Regards,

Prashant

[Former Member]

Hi Prashant,

You can increase the parameters additional 20% of the current values which you found suffering with swaps.

Mostly the parameters cand have the following values: But subject to condition you must have enough memory space.

abap/buffersize 500000

rsdb/ntab/entrycount 39000

rsdb/ntab/ftabsize 77000

rsdb/ntab/irbdsize 9500

rtbb/buffer_length 60000

rsdb/cua/buffersize 10000

zcsa/table_buffer_area 100000000

zcsa/db_max_buftab 10000

zcsa/presentation_buffer_area 20000000

sap/bufdir_entries 10000

rsdb/obj/buffersize 40000

rsdb/obj/max_objects 20000

Thank you,

Tilak

[Former Member]

Hi Tilak,

Current values in my system for those buffer's which are suffering due to access swaps are-:

abap/buffersize 900000

rsdb/ntab/irbdsize 13000 KB

rsdb/ntab/entrycount 52000

rsdb/ntab/ftabsize 80000 KB

rsdb/ntab/entrycount 52000

rsdb/cua/buffersize 10000 Kb

zcsa/table_buffer_area 200000512 Byte

zcsa/db_max_buftab 20000

rsdb/obj/buffersize 40000

rsdb/obj/max_objects 20000

rsdb/obj/large_object_size 8192

rsdb/obj/mutex_n 0

Pls suggest the increased values for these parameters

Memory

Physical mem avail Kb 33,417,940 Physical mem free Kb 226,548

Pages in/s 0 Kb paged in/s 0

Pages out/s 0 Kb paged out/s 0

Swap

Configured swap Kb 20,971,520 Maximum swap-space Kb 54,389,460

Free in swap-space Kb 8,486,456 Actual swap-space Kb 54,389,460

waiting for ur quick response

Regards,

Prashant

Edited by: Prashant Shukla on Sep 23, 2008 8:42 AM

[Former Member]

Hi All,

Waiting for your expert opinions, One more thing is it normal to have red marked swaps in ST02 for any production enviornment because we have already set the high values for buffer parameters.

Please Suggest

Current values in my system for those buffer's which are suffering due to access swaps are-:

abap/buffersize 900000

rsdb/ntab/irbdsize 13000 KB

rsdb/ntab/entrycount 52000

rsdb/ntab/ftabsize 80000 KB

rsdb/ntab/entrycount 52000

rsdb/cua/buffersize 10000 Kb

zcsa/table_buffer_area 200000512 Byte

zcsa/db_max_buftab 20000

rsdb/obj/buffersize 40000

rsdb/obj/max_objects 20000

rsdb/obj/large_object_size 8192

rsdb/obj/mutex_n 0

Pls suggest the increased values for these parameters

Memory

Physical mem avail Kb 33,417,940 Physical mem free Kb 226,548

Pages in/s 0 Kb paged in/s 0

Pages out/s 0 Kb paged out/s 0

Swap

Configured swap Kb 20,971,520 Maximum swap-space Kb 54,389,460

Free in swap-space Kb 8,486,456 Actual swap-space Kb 54,389,460

Regards,

Prashant

[Former Member]

Hi

I think that there's some mistake in server profile, you've got a lot of physical memory and you're using almost all of your swap ( 55 Gb ).

When I had this problem, it wasn't due to buffer memory parameters, but extended memory parameters.

I haven't got the list of parameters here, but check those that start with em/

My problem was that when the system started, it was configured to take a lot of memory from the beginning. My experience is that it's better to take a few memory when the system starts up and let it increase when working. If you don't do it this way, sometimes buffers are taken to swap instead of physical memory and all the problems begin.

If you can't find the parameters, tell me and I'll look for them.

Hope this helps

Regards

[Former Member]

Hi Prashant,

Ask your developer to do a ST12 trace and analyize it. Once you get the result you can check for identical selects or full table scans or unnecessary selects, with the findings optimize the code. 80% of the time performance can improved by optimizing the code.

Regards

KVR

[Former Member]

Hi KVR,

In ST02 swaps values for Buffer program CUA and export/import are coming in red colour .....just wanted to know whether this can affect the system performance?

Today i will ask abapers to analyse the programs through ST12, and will let u know the outcome.

Regards,

Prashant

[Former Member]

Hi Prashant,

I believe it is not a issue with cua buffer size. It definately looks there is a problem with Z program.

Do you regularly update your oracle statistics? What is your DB and OS versions?

[Former Member]

Hi KVR,

yes, we have already doing it from DB13 and daily its getting successfully completed.We are using HP-UX B.11.23 U ia64.

Regards,

Prashant

[Former Member]

Prashant,

as this is a Z program (ZLOGTRPL) I'd speak to the developer(s) about it and it's performance.

[Former Member]

Is there somthing else which seems to be fishy?

If you see the LAN parameters alues.....do you think its fine enough ..or this could be the reason behind the slowness.

Regards,.

Prashant

[Former Member]

Hi Prashant,

If your business process permits and you feel that its a really long running transaction why dont your schedule it in your non-dialog hour because with high response times of one transaction falsifies the overall average values in the system.

In general, long-running background jobs should be scheduled to time windows of low system activity to avoid a negative influence to the dialog response times. In doing so, the available hardware resources are not occupied additionally occupied by background load.

Thank you,

Tilak