Hi, I have'nt resolved the wait I/O question.
Someone suggested that the wait I/O percentage
is negligeable when going through ufs or advfs
because of the read-ahead mecanism. I've tried
reducing the advfs read-ahead to it's minimum
(32*512 bytes) for a device and it made no change,
wait I/O still does not increase (with cpu idle time).
My guess is that it's a bug in the way the kernel
accumulates cpu wait i/o time.
Original question was :
I'm collecting cpu statistics with the table() interface.
In the tbl_sysinfo structure, there is a variable accumulating
wait time. My understanding is that it's cpu idle time
with I/O pending. Is that correct ?
I've noticed that this wait time seems to reflect disk I/O
through the raw devices and not through the file system(ufs
or advfs). I've not observed that on other OS (Solaris or AIX).
Why would that be the case?
Thanks,
Daniel Gagne
Montreal, Canada
dgagne_at_globale.net
Received on Thu Sep 30 1999 - 00:56:29 NZST