Re: CPU time << elapsed time (trace): Why?

Rick Denoire <100.17706_at_germanynet.de> wrote in message news:<r7bqfus4qdkd5jnmdc43tgguk0uh5gv9kr_at_4ax.com>...
> >How about network time? OK. What's in v$session_wait for the session
> >in question? Can you run truss -p [pid] and show us portion of it?
>
> The application in question does not exchange huge amounts of data
> over the network.
>
> OK, thanks for the hint using truss -p. What do you expect to see?
> I am afraid that the elapsed time (about 29) can't be truss'ed that
> way, and that I will get to much data that I won't be able to deal
> with. Truss would yield a long, long file, and trying to interpret
> what is happening would be just a mess.

Hi, Rick,

If you want to find out what a specific session is doing, you must at least look at v$session_wait (not v$sys% views). If you're not willing to tell us what that view says about your session, how can we help you?

Occasionally truss at the OS level reveals something interesting. For instance, if your shadow process keeps calling yield(), it may be because this session has hogged the CPU too long and the system is busy so other processes have to take more and more CPU slices. If you keep seeing poll(), maybe the network or the client that that talks to the Oracle server needs some performance tuning.

Yong Huang Received on Tue Jun 04 2002 - 22:53:41 CDT