topic Re: Where to look for applications performance issues on hpux with regard to sap and peoplesoft in Operating System - HP-UX

Where to look for applications performance issues on hpux with regard to sap and peoplesoft

Shivkumar — Wed, 09 Aug 2006 19:02:31 GMT

Hi,

I would be given the additional task of supporting peoplesoft and SAP applications running on HPUX 11i.

I would be required to support these applications from system admin standpoint of view.

I appreciate if anyone can suggest some tips where should i be looking if these applications experiences issues, performance and errors ?

Thanks,
Shiv

Re: Where to look for applications performance issues on hpux with regard to sap and peoplesoft

A. Clay Stephenson — Wed, 09 Aug 2006 19:41:51 GMT

Just like any other applications narrow the bottlenecks to disk, memory, and cpu and attack the biggest bottlenecks first. Now having said this, in general, the real performance enhancements are much more likely to come from making the SQL more efficient then tuning the database itself and finally tuning the OS. Unless you go out of your way to terribly tune the kernel, terribly layout your disks, or leave tunables like dbc_max_pct at the default of 50% -- you are actually going to find that the OS is actually a small component of the overall application performance.

What you really better concentrate on is making sure that you can restore a blank machine and get the application running again.

Re: Where to look for applications performance issues on hpux with regard to sap and peoplesoft

Michael Steele_2 — Wed, 09 Aug 2006 20:51:56 GMT

glance is probably the best realtime monitoring tool. If you want historical data then you'll have to redirect the glance data inro an output file. Refer to the 'm' (memory) report in glance. Pay attention to page outs. Here's an example of a messed up server and the (m) memory report from glance. All the percents utilized are near 100%. Page faults are astronomical, as is swap utilized.

B3692A GlancePlus C.03.72.00 11:34:22 rodan 9000/800 Current Avg High
--------------------------------------------------------------------------------
CPU Util S SU U | 68% 73% 100%
Disk Util FV | 3% 17% 78%
Mem Util S SU UB |100% 99% 100%
Swap Util U UR R | 98% 97% 98%
--------------------------------------------------------------------------------
MEMORY REPORT Users= 5
Event Current Cumulative Current Rate Cum Rate High Rate
--------------------------------------------------------------------------------
Page Faults 18372 1033322 3749.3 2100.2 9925.6
Page In 1208 76733 246.5 155.9 723.8
Page Out 0 2573 0.0 5.2 107.4
KB Paged In 52kb 17.6mb 10.6 36.5 o/f
KB Paged Out 0kb 48.1mb 0.0 100.0 2565.8
Reactivations 5 493 1.0 1.0 4.5
Deactivations 5 493 1.0 1.0 2.2
KB Deactivated 0kb 48kb 0.0 0.0 6.6
VM Reads 9 1293 1.8 2.6 147.3
VM Writes 32 7900 1.8 16.0 306.7

Total VM : 12.4gb Sys Mem : 1.38gb User Mem: 6.42gb Phys Mem: 8.00gb
Active VM: 6.76gb Buf Cache: 163.8mb Free Mem: 41.9mb Page 1 of 1

Re: Where to look for applications performance issues on hpux with regard to sap and peoplesoft

Michael Steele_2 — Wed, 09 Aug 2006 20:54:13 GMT

This was from an BW SAP server, aka application server, riding atop of oracle.

Re: Where to look for applications performance issues on hpux with regard to sap and peoplesoft

V.Manoharan — Wed, 09 Aug 2006 22:48:48 GMT

Hi shiva,
1. your need to check the bottelneck pertaining to disk i/o, cpu, Memory.
This can be isolated using top,sar and galnce.

2.Check for the OS related tunable parameters to optimize the server.
3.Check for the tunables parameters as recomended by SAP.
kernal tuning can be done through sam or kmtune command.

Thanks and regards
V.Manoharan

Re: Where to look for applications performance issues on hpux with regard to sap and peoplesoft

Steven E. Protter — Wed, 09 Aug 2006 23:08:34 GMT

Shalom,

The applications themselves can be memory hogs. That needs to be watched.

The back end database SGA must not be too large or too small.

dbc_max_pct low arond 5
dbc_min_pct 3 or 4

ninodes set high enough to deal with file usage
vx_ninodes set to 90% of ninode

Watch the database alert logs and syslog.

SEP