Date: Tue, 18 Apr 1995 07:34:02 -0400 (EDT) From: Peter Dufault <dufault@hda.com> To: vernick@cs.sunysb.edu (Michael Vernick) Cc: freebsd-scsi@FreeBSD.org Subject: Re: SCSI command processing Message-ID: <199504181134.HAA02991@hda.com> In-Reply-To: <199504172155.RAA29886@cs.sunysb.edu> from "Michael Vernick" at Apr 17, 95 05:55:58 pm
next in thread | previous in thread | raw e-mail | index | archive | help
Michael Vernick writes: > > I have also done some testing of SCSI overhead processing. On average > an 'Inquiry' command sent to the disk over an EISA bus takes 1.2 ms, > that is going in and out of user space from start to finish. To > retrieve 1 block from the disk cache also takes about the same amount > of time, 1.22ms on average. I test this by first doing a read of a > random block and then re-read it. The first read incurs the seek, > rotation, and transfer overhead, the second read is satisfied by from > the disk cache. So the 1.22 ms includes: going in and out of user > space, 2 each of: EISA bus transfer, SCSI controller overhead, SCSI > bus transfer and disk overhead. Thus, it would seem the disk overhead > is quite small. When I have some spare time I'm going to instrument the upper levels of the code. There is a fair amount of overhead in that software, but I suspect it is drowned by the overhead of the host adapter and the target firmware. The fact that Test Unit Ready takes forever makes me most suspicious of the host adapter firmware, and it would be nice to have a standard way of measuring the host adapter overhead across the different supported adapters. -- Peter Dufault Real Time Machine Control and Simulation HD Associates, Inc. Voice: 508 433 6936 dufault@hda.com Fax: 508 433 5267
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504181134.HAA02991>