Re: "Tuning" small tasks ...

Thank You, Niall, for Your reply. Now I can see some ways to go further towards tracing ...

> Don't measure the size of the table in rows but in blocks below the HWM. A
> 100 row table that is heavily inserted/deleted could in fact be a 100mb
> table on disk.....

... I knew the HWM problem, but in this case, the table never was big ...

> > - desc some_view or desc some_table may also tale 10 seconds ...
> > on other times the responses come up at once, just as it should be ...
>
> See if you can trace the session and find out what the session is spending
> its time doing. It will probably be waiting for something and a trace of
the
> session will show this. If you have metalink access look for docid 21154.1
> on level 10046 tracing with waits which you can enable from sqlplus with
>
> DBMS_SUPPORT.START_TRACE(); for your own session and
> the unsurprisingly named
> DBMS_SUPPORT.STOP_TRACE;
>
> to stop it. You may need to install dbms_support which you can do by
running
> ?/rdbms/admin/dbmssupp WHEN CONNECTED AS SYSDBA
>

 THIS is what I am going to do next !
There must be some waitings that even slow down a "desc" sometimes ...

> >
> > I could not verify what may cause this behavior. It happens on normal
> times
> > or on heavy duty times (with much redo),
> > with 300 sessions online, but only 10 active simultaniously at the peak.
I
> > (think that I) have appropriate hit ratios and so forth.
>
> I can't resist commenting on your remark that 'the instance is tuned' ...

well, as far as I could tell, and obviously there is still something to do about ...

---
Thank You very much !

Jan=)