From owner-freebsd-scsi Tue Sep 28 4:49:57 1999 Delivered-To: freebsd-scsi@freebsd.org Received: from mail.rdc2.on.home.com (ha1.rdc2.on.home.com [24.9.0.15]) by hub.freebsd.org (Postfix) with ESMTP id 2656A1547D for ; Tue, 28 Sep 1999 04:49:42 -0700 (PDT) (envelope-from street@iname.com) Received: from mired.eh.local ([24.64.136.188]) by mail.rdc2.on.home.com (InterMail v4.01.01.07 201-229-111-110) with ESMTP id <19990928114942.ECCN5795.mail.rdc2.on.home.com@mired.eh.local> for ; Tue, 28 Sep 1999 04:49:42 -0700 Received: (from kws@localhost) by mired.eh.local (8.9.3/8.9.3) id XAA01197; Mon, 27 Sep 1999 23:40:37 -0400 (EDT) (envelope-from kws) To: "Kenneth D. Merry" Cc: groudier@club-internet.fr (Gerard Roudier), scsi@FreeBSD.ORG Subject: Re: sym driver 0.3.0 References: <199909271956.NAA38933@panzer.kdm.org> From: Kevin Street Date: 27 Sep 1999 23:40:37 -0400 In-Reply-To: "Kenneth D. Merry"'s message of "Mon, 27 Sep 1999 13:56:09 -0600 (MDT)" Message-ID: <874sgfog96.fsf@mired.eh.local> Lines: 88 X-Mailer: Gnus v5.6.45/XEmacs 21.1 - "20 Minutes to Nikko" Sender: owner-freebsd-scsi@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org "Kenneth D. Merry" writes: > Gerard Roudier wrote... > > You test is not serious. If you only can provide 1.5MB/second throughput > > for a benchmark of SCSI driver, please go away or make comparison with an > > AHA 1542. I consider your posting to be a bad joke and hope that it is not > > intentionnaly just bad taste. This original hasn't made it to me yet so I don't have all the context. If this comment by Gerard was aimed at me, then my response is: I'm assuming that you intended the sym driver to be used for real applications on real machines. If its only purpose is to run benchmarks, then let me know that and I will go away. > I think what you may want to do is ask Kevin to give you some comparisons > of iostat output from dumps with the ncr driver and dumps with the sym > driver, ... I've done a few more tests. I tried to see the effect that the slow Syquest drive is having. When I dump to an IDE drive then the sym driver is faster than the ncr by 15-20%. But when I dump to the Syquest, then the interference slows down the sym driver more than the ncr. Here's what I got from 8 test runs of dump. dump from da1 to ad0 (IBM SCSI to WD IDE) time KB/sec ncr 246 1764 ncr 250 1736 sym 211 2056 sym 209 2076 dump from da1 to da0 (IBM SCSI to Syjet SCSI) time KB/sec ncr 275 1578 ncr 277 1566 sym 335 1295 sym 336 1291 Looking at iostat shows dump pushing 64KB blocks at the Syquest which is almost keeping up. It gets behind occasionally and the da1 transfers stop until da0 catches up. With the ncr driver it looks like this: tty da0 da1 cpu tin tout KB/t tps MB/s KB/t tps MB/s us ni sy in id 0 30 64.00 24 1.49 4.75 265 1.23 1 0 5 2 92 0 30 64.00 25 1.55 4.71 219 1.01 1 0 2 1 97 0 30 64.00 23 1.42 0.00 0 0.00 0 0 0 0100 0 30 64.00 25 1.55 0.00 0 0.00 0 0 0 0 99 0 30 53.92 25 1.30 3.56 275 0.96 1 0 6 1 92 0 30 64.00 25 1.55 3.60 564 1.98 0 0 15 2 83 On the sym driver, the pauses from da1 are more frequent and longer. On the Syquest at da0, there's something strange going on. The transaction rate on da0 drops way off while it's trying to catch up, which means that da1 is forced to wait longer before it can start blasting data at it again. tty da0 da1 cpu tin tout KB/t tps MB/s KB/t tps MB/s us ni sy in id 0 30 64.00 22 1.36 2.91 397 1.13 0 0 21 18 60 0 30 64.00 21 1.30 3.41 292 0.97 1 0 14 21 64 0 30 64.00 20 1.24 3.75 82 0.30 0 0 5 21 74 0 30 64.00 21 1.30 0.00 0 0.00 0 0 0 27 72 0 30 64.00 21 1.30 0.00 0 0.00 0 0 0 26 73 0 30 64.00 19 1.18 0.00 0 0.00 0 0 0 24 75 0 89 64.00 17 1.05 0.00 0 0.00 2 0 0 17 81 0 30 64.00 15 0.93 0.00 0 0.00 1 0 0 14 85 0 30 64.00 8 0.50 0.00 0 0.00 0 0 0 8 92 0 30 64.00 8 0.50 0.00 0 0.00 0 0 0 1 99 0 30 55.27 22 1.18 3.69 702 2.53 2 0 11 6 82 0 30 64.00 22 1.36 3.78 670 2.47 1 0 21 16 62 0 30 64.00 23 1.42 3.19 478 1.49 1 0 21 17 61 0 30 64.00 20 1.24 0.00 0 0.00 0 0 0 22 77 0 30 64.00 17 1.05 0.00 0 0.00 1 0 0 16 83 0 30 64.00 8 0.50 0.00 0 0.00 0 0 0 9 91 0 30 64.00 4 0.25 0.00 0 0.00 0 0 0 4 96 0 30 48.95 21 0.99 3.43 520 1.74 2 0 8 2 88 So this would appear to be the cause of the slower dumps. Perhaps this is error recovery activity on da0? Whatever it is, the sym driver isn't handling it as gracefully as the ncr. -- Kevin Street street@iname.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-scsi" in the body of the message