On Mon, May 29, 2006 at 08:40:00AM +0000, abrasive@axdf.net wrote:
> >Number:         33585
> >Category:       port-i386
> >Synopsis:       Heavy SATA disk load causes huge interrupt load (piixide)
> >Confidential:   no
> >Severity:       serious
> >Priority:       medium
> >Responsible:    port-i386-maintainer
> >State:          open
> >Class:          support
> >Submitter-Id:   net
> >Arrival-Date:   Mon May 29 08:40:00 +0000 2006
> >Originator:     James Laird
> >Release:        NetBSD 3.0
> >Organization:
> 	AXDF
> >Environment:
> System: NetBSD what-are-birds 3.0 NetBSD 3.0 (BIRDS.PROF) #0: Tue May 30 02:29:48 EST 2006 root@what-are-birds:/usr/src/sys/arch/i386/compile/BIRDS.PROF i386
> Architecture: i386
> Machine: i386
> >Description:
> 	Whenever SATA disk is heavily loaded (e.g. making filesystems) the CPU interrupt time climbs to 60-80% (Intel P4 2.8Ghz).
> 	vmstat -i shows an interrupt rate on the appropriate ioapic pin in the region of 1-2 thousand per second with disk load, and less than 100 with no load.

How many disk transfers per second is there ? With the current driver you
have one interrupt per transfers, and with modern disk with large write-back
cache we can certainly get that much I/O per second.

'systat vm' will show you these informations (both interrupt rate and I/O rate)

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--