Don Lewis wrote...
> On Jun 3,  8:36am, Jason Thorpe wrote:
> } Subject: Re: Support for Adaptec AIC-7895?
> } On Wed, 03 Jun 1998 08:36:02 -0700 
> }  Matthew Jacob <mjacob@feral.com> wrote:
> } 
> }  > is in that. I've begun really liking the CAM stuff he's doing and
> }  > am mulling over whether NetBSD ought to have it too- but this is
> }  > not near term at all.
> } 
> } What's so great about it?
> 
> I've heard a number of reports of significantly improved I/O performance.
> The error recovery is also supposed to be better than the previous SCSI
> implementation.

	The real performance improvement comes with random I/O.  Performance
with sequential I/O will probably be similar to the old SCSI layer.

	I think the error recovery is a bit better, in general.  (okay, I
may be a bit biased..:) )  Most of the error recovery actions are
table-driven, so you can more easily control how the system reacts to
various errors.  We may run through the error recovery code yet again, but
I think it's in good shape now.

> } Note that NetBSD is going to have to basically
> } rewrite the SCSI code anyway, for the multi-threaded kernel.  If Justin's
> } CAM isn't up to par for that, well....
> 
> The CAM implementation is mostly functional in the FreeBSD SMP branch, though
> I've heard several reports of a bug that Justin is having trouble tracking
> down.

	I think Justin fixed the bug that was supposedly SMP-related,
although it really wasn't.  It just happened that the people who ran into
it had SMP boxes.  FWIW, I've been running CAM on my dual-P6 machine since
July or August of last year, and it's been very stable.


Ken
-- 
Kenneth Merry
ken@plutotech.com