From Glance to See Performance

[Krishna Prasad]

Steven,

What is your Oracle SGA set to. If it is to low it could be causing your paging becuase Oracle's needs more shared memory. It is also possible that it is set to high not leaving enough physical memory for Oracle clinet connections.

You can find out by looking at your initSID.ora file.

Also, what is your Oracle hit ratio this is a good indicator for the correct setting for your SGA. If you are getting less then 90% hits the SGA needs to be adjusted.

[Bill Hassell]

For Oracle, the size of SGA (specifically, Oracle's buffer cache, shared pool, temp pool, etc) directly affects performance. Page out is the ONLY stat you pay attention to since page in also counts every new process that starts.

If page out is 5-10 in Glance, you're fine. If it starts pushing 30-50, then performance will suffer. Since you are limited in RAM by budget constraints, there is nothing you can do as a sysadmin. However, most sysadmins and seasoned DBAs will agree that rewriting poor SQL statements and/or adding new indexes will help performance more than kernel tweaking. Note that Oracle silently ignores indexes that are badly out of balance...it's a good idea to rebuild those indexes regularly when lots of row insertions are being done. These indexes can be created very fast by using the old index to build a new one rather than deleting and creating the index.

[Steven Chen_1]

Again thanks all for the in-depth analysis.

Question for Bill & Harry:

1) among columns in mem Glance graph, which worths looking on "page out", Current, Cumulative, or High Rate...?

2) dbc_min_pct is set to 5% and at peak it keeps on there of 51M. So changing dbc_max_pct from current 15% to 10% does little about it. Is that right? Or put it this way, hpux11i needs 280M to work in this case?

Thanks,

Steven

[Sridhar Bhaskarla]

Hi Steven,

As I mentioned before, we can see some deactivations happening on your system. On a 11.0, the deactivation occurs typically after reaching the "minfree" limit. So, you would want to check the rate of page outs and deactivations when your system is heavily used. Also, your swapinfo -t and the KB used (only on the swap devices not the reserve or memory rows) in it will give you how much is sitting in your swap areas.

I don't know how much it is going to help you but you can bring down some of the processes like NFS, sendmail, lp daemon etc if you are not using them. Also install the lastest possible patches and diagnostics.

-Sri