Re: Latch spin - OS timeslice

Jonathan Lewis doodled thusly:

>
>Bottom line - there are two possible performance problems -
>excessive CPU usage with minimal time lost on sleeps
>(latch free waits), or excessive sleeps on latches with
>plenty of spare CPU. If all else failed, the former might
>suggest reducing the spin_count cautiously, the latter might
>suggest increasing the spin_count cautiously.

Thing that gets me is: why not make the sleep time a parameter? After all, that's much more related to the CPU type and system load than the spin count. One might say: 6 of one, half a dozen of the other. Not quite. An adjustable sleep time is a completely different proposition than a variable spin count followed by a fixed sleep time. IMHO, adjusting sleep time wouldn't affect CPU use counts as much as changing the spin count.

>
>Interesting point - Oracle 9 introduces shared read latches
>to reduce the potential impact of latch collision; and also
>may reduce the sleep times to microsecond levels in
>response to increased CPU speeds in 9.2.
>

Nice! Sounds like this whole area is being looked at in the near future.

Cheers
Nuno Souto
nsouto_at_optushome.com.au.nospam Received on Fri Mar 29 2002 - 06:57:34 CST