Performance Problems

[Giannina Villa]

Hi;
We have 2 server HP-UX 11.11 with the same
disk configuration (in both we have installed
Oracle 8.1.7.2.0 and SAP 4.6C), one of them
has only 1 processor PA RISC (750 Mhz) and 2Gb
of memory and the other has 2 Processors PA RISC (650 Mhz) and 1,5 Gb.of memory in this last one we have performance problems, also to recover the application and the data base time is double compared with the first server.
We have found some kernel parameters that are
differents:
SERVER1 (1 Processor / 2Gb.Memory)
bcvmap_size_factor --
max_thread_proc 64
msgmni 50
ncsize 13312
semmap 514
semmni 512
st_san_safe 0
vps_ceiling 16
vx_maxlink --
SERVER2 (2 Processor / 1.5Gb.Memory)
bcvmap_size_factor 2
max_thread_proc 256
msgmni 2560
ncsize 23552
semmap 522
semmni 520
st_san_safe --
vps_ceiling 64
vx_maxlink 32767
Which could be the cause of this problem?
Thanks for your cooperation.

[Sundar_7]

If both the servers are running similar application and expected to have identical load, I dont see a reason why the kernel parameters should be different.

First thing I would do is to make sure the kernel parameters are in sync.Especially the IPC parameters (msgmni,semmni,semmap) can make a difference. And if your application is not threaded, you could as well reduce the max_thread_proc to match SERVER1.

Even if the disk configuration is same, resources available to the user processes/application need not necessarily be same in both the servers.

I would suspect some problem with the swap/virtual memory configuration in SERVER2.

Compare the swapinfo output of SERVER1 and SERVER2.

and also run vmstat command regularly and observe pages-in and pages-out values to see if the SERVER2 is facing any memory pressures.

If only the kernel paremeters list above are different between the machines, I tend to think the kernel parameter difference is not the culprit.

As far as I can reach, except for the difference in msgmni I dont see a significant difference (I am not aware of vx_maxlink parameter).

[Con O'Kelly]

Hi

I'd say its unlikely that the kernel parameters are the problem.

When you say you have performance problems - what do you mean?
Extremely high memory utilisation or CPU utilisation?
You need to try and pinpoint where your performance problem lies.

Provide output of:
# sar 2 5
# sar -q 2 5
# vmstat 2 5
# swapinfo -tam
# ipcs -mob

Is oracle configuration the same on both servers?

Cheers
Con

[Hein van den Heuvel]

Step one: forget about kernel parameters for now. They may make or break an applicaiton, but they are rarely the cause of significant performance differences.

Step two: try and put numbers onto 'performance problems'. I know you know some details. Maybe an SAP measurement or even an end user feedback. Share that and provide details. Overall Tx time from x to y? DB response time average changed? How does the cpu (user/system) compare under similar loads? How about the IO and Page rates?

WAG... the memory on the second systems is overcommited. Is SAP (em/initial_size_MB) and Oracle (SGA) per chance configured identical? How much memory are you commiting/using? vmstat?

Good luck,
Hein.

[Bill Hassell]

Both of these machines are very small for Oracle and especially SAP. Oracle and SAP can be easily configured to use more memory than you have which causes extremely slow performance due to page outs. The only fix is to bring the installed memory to 4Gb as a minimum. You'll need more processors (4 total will help a lot), especially if your Oracle and SAP specialists configure the apps to use a lot of memory.

[Giannina Villa]

Hi,
we know of the SAP and Oracle requeriments about the memory but even so as we mentioned previously in the Server1 we do not have greater problems of performance and response time for the own transactions of the application neither, the oracle configuration is the same in both servers.
You could find the details about the sar, vmstat, ipcs, swapinfo commands in the attachment.
Thanks again.

[Hein van den Heuvel]

The memory in the secodn box is totaly over commited. It is somewhat overcommitted on teh first box.


Look at the free clumn in vmstat... nothing left.
Look at the pi and po activity... constant.


You have a 1GB SGA in a 1.5 GB box. That's not going to work. Not even talking about the sap memory. Try SQLPLUS "/ as sysdba" ... SHOW SGA.
Upgrade to 4Gb as suggested, or drastically reduce the Oracle SGA to get in workable space. Work with your DBA. Have them try 400MB pool + 200 MB buffers, or even lower?



T ID KEY MODE OWNER GROUP NATTCH SEGSZ
Shared Memory:
:
m 19717 0x8d3acb48 --rw-rw---- orappr dba 34 1061937152


Good luck,
Hein.