Subject: Re: siop(4) and tagged queuing
To: Manuel Bouyer <email@example.com>
From: Matthew Jacob <firstname.lastname@example.org>
Date: 10/23/2000 10:16:31
On Mon, 23 Oct 2000, Manuel Bouyer wrote:
> On Mon, Oct 23, 2000 at 09:58:39AM -0700, Matthew Jacob wrote:
> > Almost all of your drives will do this- what you need to do is bump
> > SDOUTSTANDING up above 100 or so in sd.c.
> Well, OK, I'll try this on older drives, with a small cache.
> BTW, is SDOUTSTANDING really used ? It seems that all drivers doing tagged
> queueing override this by the number of tags they can do (siop does this) ?
if (sc_link->openings > SDOUTSTANDING)
sc_link->openings = SDOUTSTANDING;
which throttles things to SDOUTSTANDING.
Be careful about removing this. OpenBSD did this recently with amusing
The midlayer needs to be a lot smarter about job scheduling so that it
respects an HBA's openings limit and also w/o letting one spindle saturate a
system but also w/o artificially limiting things.
> > You won't see the benefits of tagged queiing on bonnie- you need a
> > multithreaded test that tests multiple parts of the drive.
> Ok, I though the seekers would do this.
> Any benchmark to recommend ?
Urr,.... I've been looking for one other than my buffer cache thrasher
myself... The back end of SPECnfs... Something such that a sufficiently long
queue of disk operations down at the drive level benefit from drive sorting.
You should note that this is an area of some controversy. There are a lot of
conflicting data points, mostly based on experiments done years ago, as to
whether or not TQ'ing really helps or not.