Re: Latch spin - OS timeslice

Here's one puzzle to me. Take my Solaris 2.6 on sun4u 8 CPUs as an example. mpstat tells me the number of spins on mutexes (smtx column) per second on one processor is way below 100. Maybe this is at night. But I can't see a number higher than 100 no matter what I do in the database or through my Portal site which is notorious for library cache latch contention (my _spin_count is 2000). ps -cL tells me the foregroup and background oracle processes are at TS class priority ranging from 8 to 46. dispadmin -c TS -g tells me that their time quantum ranges between 40 to 200 ms. I would imagine that time slice is more than enough for 2000 spins. Why does mpstat show such a low number?

Since this is an Oracle group, can anyone tell me how to trigger oracle process spins? I know when I access an Oracle Portal page, usually more than 1 session will be waiting for "latch free" (latch#106) for several seconds! Maybe I'll wait till 10am and possibly set _spin_count to 10000 or higher.

Yong Huang

alexbrds_at_yahoo.com (Alex Bardos) wrote in message news:<f5e30250.0203251807.92e1747_at_posting.google.com>...
> time slice. I would expect the whole SPIN cycle to be comparable to
> the time slice, still smaller though to increase chances of not being
> preempted by the OS after the time slice is up.
Received on Wed Mar 27 2002 - 23:59:18 CST