> > Run1 latches total versus runs -- difference and pct > > Run1 Run2 Diff Pct > > 7,067 2,301,590 2,294,523 .31% > > *** >=20 > Are you sure this is a reliable figure? No. Actually it were systemwide latches. And test was run on suffering from lack of resources instance. On idle system I have numbers like: Run1 ran in 1 hsecs Run2 ran in 17 hsecs run 1 ran in 5.88% of the time Run1 latches total versus runs -- difference and pct Run1 Run2 Diff Pct 3,410 30,898 27,488 11.04%
> > To me, v$access is only usable on idle instances. >=20 > Not necessarily. v$* views were put there for a reason, if I'm not talking about generic v$* views. I'm talking only about slow (very slow on real nstances) selects from v$access view.
> Don't forget you are not comparing apples with apples: > traces themselves have not been instrumented with waits, you simply > do NOT know if your trace isn't causing a lot more overhead in > places where waits are not counted! It might be true for systemstate dump, but not for hanganalyze dump. 'hanganalyze level 2' is very fast even on overloaded (in terms of CPUs and disks) instances.
