Performance

[Faizer Jameel]

Dear All
I have Oracle ERP 11.5.10.2 installed on a 2 node rp7410 cluster running houx 11.11
I have attached the output of glance and swapinfo.
The users are complaining of real slowness.
Any advise please,
Thanks

[Steven E. Protter]

Shalom,

Sorry you need to provide more information.

sar -d
swapinfo -tam
vmstat output.

You may wish to try this toolset for collecting data:

http://www.hpux.ws/system.perf.sh

SEP

[Patrick Wallek]

It appears that your system is working very hard.

Your glance output shows 100% CPU usage. and 99% RAM usage.

Your sar output (that is sar output, by the way and NOT swapinfo output) shows that you do have some processes being paged out. That is not necessarily good.

What is the output of 'swapinfo -tam'?

How much RAM do you have on this machine? My first reaction to this would be to try to add some more RAM. It appears you may have a shortage.

[Faizer Jameel]

This is the additional information that I have captured and loaded now.

[Hein van den Heuvel]

More data is always needed.
Simple stuff like an indication of the IO subsystem (raid 5?) cpu counts, memory size.

But perhaps more importantly, a little context is needed.

Has it ever worked well?

Any major changes recently? (Cluster, OS, Oracle?)

What is the performance history? You do have a history showing cpu, mem, IO usage by day right?

At first glance there is CPU, IO is a little busy, memory highly used. Decrease the buffer space a little more and give to Oracle?
Buy more memory and give to Oracle as sga/db_buffers.

I suspect that any real understanding of performance issue wil have to come from how well Oracle is behaving. So you best get a statspack report or two going covering peak business windows and/or other windows of identifyable slowness.

Good luck,

Hein van den Heuvel
HvdH Performance Consulting.

[A. Clay Stephenson]

Note that your memory utilization is very high and that your pageout rate is non-zero. Although the amount of data you presented is limited, there is a strong indication that your box is under moderately-high memory pressure. You need to add more RAM, reduce the size of database data structures, reduce your buffer cache size, and/or reduce the number of processes.

[Faizer Jameel]

The system has worked well. Also this system is a one of the 2 node clusters which is connected to a va7410 box.
The memory is 10gb. it has 4 cpus.
2 weeks ago we upgraded Oracle ERP for HRMS implementation though no users still exist for HRMS. We applied a series of patches from Oracle for this upgrade process.

I too think there is a memory pressure, and there are disk bottelnecks too.
Any other information that is required that could be of help.
Like kernel parameters etc. Just one thing I suspect the dbc_max_pct that I have set is the default and I feel it should be in the range of 15-20 with the dbc_min_pct to be 5.

[Patrick Wallek · Accepted solution]

YES, you should definitely reduce dbc_max_pct.

The default of 50 is WAY too high. For a system with 10GB of RAM, I would probably set dbc_max_pct at 10 and would leave dbc_min_pct at 5.

[A. Clay Stephenson]

The default value for dbs_max_pct is 50% and even your 15-20% value is too large for a box with 10GiB of physical memory. In your case, I would run a fixed buffer cache of about 1200MiB or so by setting bufpages = 307200. By going to a static buffer cache, you are able to make that value constant and thus make overall tuning easier. The "sweet spot" for large memory boxes under 11.11 is somewhere in the 800-1600MiB range and 1200MiB should be quite reasonable.

[Faizer Jameel]

Well my opinion I would have the kernel parameters changed and this would help in an improvement.
Next would be to add some more memory.
Well now the system is very quite as all the users have left for the day. Tomorrow morning the system would dance again.
Thanku all round and would get back to the forum if there are any more issues.
Thanks a lot