Memory Issues

[Jason_93]

Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 1703936 68988 1634948 4% 0 - 1 /dev/vg00/lvol2
reserve - 176504 -176504
memory 780648 765276 15372 98%

This is the results of swapinfo. Needless to say we are having system issues. I am unable to identify any one process that is utilizing all the memory. How can I determine if this system is just plain underpowered memory wise?

[Sridhar Bhaskarla · Accepted solution]

Hi,

You have 68 MB sitting on swap which means the system did page out.

Run vmstat and look at 'po' column. If you see a doule digit or more number continuously, then you are experiencing memory issues.

One parameter that you can check immediately is 'dbc_max_pct'. You wouldn't need more than 700-800 MB on 11i and 300-400MB on 11.0. If the above parameter is set to 50% (default), then you may want to consider lowering it.

Also, keep track of the size of each of the processes and see if it is not increasing over the time. If so, then there is a good chance that the process has a memory leak. Run the following command at regular intervals and see if the first column is increasing for any of the processes over the time

UNIX95= ps -e -o 'vsz pid args' |sort -n > /tmp/ps.out

-Sri

[Steven E. Protter]

I suggest some performance monitoring and am attaching my script for that purpose.

swapinfo -tam will give you more data.

SEP

[A. Clay Stephenson]

The most important metric to observe, either with Glance or vmstat, is the page-out rate. If the page-out rate exceeds 10 or so for any length of time then you are definitely swapping too much. The ideal is zero. If you are running the default dbc_max_pct, 50%, then reduce it to no more than about 20%. If you already have reasonable buffer sizes then you need to add more memory or reduce load.

[Jason_93]

The page out, and in is almost always at zero, and never into double digits.