ServerStats problem. "Database Time Per Sec" greater than 100 centisecs

Question

Hi,
I have set up a ServerStat to read the V$SYSMETRIC table.
OLT reports the value to be greater than 100 centisecs.
How is this possible?
Regards,
Axel
Edited by: ao on 2011-feb-22 13:17

Answer 1

As Kecskemethy points out, the CPU count matters but so does the number of active sessions.
Imagine a simple example where you have a single CPU core and you have 100 sessions. Session 1 does a SELECT ... FOR UPDATE on a single row in a table. Session 1 then does some work that requires 1 second of CPU time and does no I/O. Sessions 2-99 do a SELECT ... FOR UPDATE on that same single row just after session 1 gets the lock. The total database time that Oracle will count for that single second is 100. Session 1 will have 1 second of database time due to CPU. Sessions 2-99 will each have 1 second of database time due to lock contention.
If you're looking at generalized performance, total database time is often a superior metric to something like CPU time because it's measuring all the waits that all your users are actually experiencing. If session 1 is idle (doing no work) while it is holding the lock (say, because it is waiting on a user to press a button in the GUI), the CPU and I/O utilization of the server may be minimal but you may have dozens or hundreds of users racking up significant wait time.
Justin