Comments on: Another use case for WaitProf – diagnosing “events in waitclass Other”

By: lscheng

lscheng — Sat, 26 Sep 2009 16:33:52 +0000

Aha, I thought Oracle count as CPU usage when it is waiting in the runqueue (for the time model)

So in ASH it can actually get wrong CPU usage

Great, thanks for the tips!

By: Tanel Poder

Tanel Poder — Sat, 26 Sep 2009 14:37:38 +0000

So, things like ASH & v$session_wait are not good for determining CPU usage. In fact, if you want to get the correct picture, use vmstat or an OS tool.

By: Tanel Poder

Tanel Poder — Sat, 26 Sep 2009 14:36:59 +0000

Yes as if an Oracle session is not waiting for anything then it ought to be on CPU (or at least try to be on it).

The exceptions are instrumentation bugs where Oracle actually is waiting for some syscall to complete without reporting it in wait event and also suspended processes.

Regarding your other question – Oracle does not report more CPU usage than actually used if you query the right views (like V$SYS_TIME_MODEL DB CPU or even V$SYSSTAT session cpu usage). The data you see in these views comes straight from OS, Oracle asks CPU usage info via times() syscalls on Unixes.

Now if we talk about v$session_wait, ASH etc, yes they can account more CPU usage than there’s CPU time available, because Oracle thinks even CPU scheduling latency is CPU usage.

By: lscheng

lscheng — Thu, 24 Sep 2009 05:24:05 +0000

Hi Tanel

I have observed in your script that when process is waiting you use CPU / runqueue to show as wait event.

Do you know if in a CPU bounded system Oracle can account more DB CPU time than it actually uses? i.e Can Oracle report more CPU usage (in Oracle statistics such as DB CPU or CPU Used by this session) than it really uses?

By: Log Buffer #103: a Carnival of the Vanities for DBAs

Log Buffer #103: a Carnival of the Vanities for DBAs — Fri, 27 Jun 2008 16:34:52 +0000

[...] his Core IT for geeks and pros, Tanel Poder expounds another use case for WaitProf. He begins, “I recently diagnosed a performance issue where the ‘events in waitclass [...]

By: tanelp

tanelp — Fri, 20 Jun 2008 18:04:49 +0000

Thanks! Yep, normally you shouldn’t see any significant waits from the Other group (that’s why they all are consolidated there), but there are always some corner cases…

By: Bradd Piontek

Bradd Piontek — Fri, 20 Jun 2008 17:59:16 +0000

Awesome post. I have been seeing just about every 10g database we have with a warning for the ‘OTHER’ wait group in Grid control. I knew there were a ton of waits aggregated under it, but at least now I have some ammunition into diving into the true issues, if they exist.