From owner-freebsd-questions Sat Jun 1 06:44:43 1996 Return-Path: owner-questions Received: (from root@localhost) by freefall.freebsd.org (8.7.5/8.7.3) id GAA15497 for questions-outgoing; Sat, 1 Jun 1996 06:44:43 -0700 (PDT) Received: from eros.britain.eu.net (eros.Britain.EU.net [192.91.199.2]) by freefall.freebsd.org (8.7.5/8.7.3) with SMTP id GAA15479 for ; Sat, 1 Jun 1996 06:44:39 -0700 (PDT) Received: from nadt.org.uk by eros.britain.eu.net with UUCP id ; Sat, 1 Jun 1996 14:44:24 +0100 Received: from infodev (infodev.nadt.org.uk [194.155.224.205]) by charlie.nadt.org.uk (8.6.12/8.6.12) with SMTP id OAA03130 for ; Sat, 1 Jun 1996 14:34:02 +0100 Date: Sat, 1 Jun 1996 14:34:02 +0100 Posted-Date: Sat, 1 Jun 1996 14:34:02 +0100 Message-Id: <199606011334.OAA03130@charlie.nadt.org.uk> X-Sender: robmel@mailhost X-Mailer: Windows Eudora Light Version 1.5.2 Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" To: questions@freebsd.org From: Robin Melville Subject: Agonisingly slow SCSI (was: excessive interrupts) Sender: owner-questions@freebsd.org X-Loop: FreeBSD.org Precedence: bulk Dear FreeBSD Folks, I have installed the CD FreeBSD 2.1 release on an HP NetServer LM. This is a P90 with 16Mb and a 2033Mb HD driven by an HP-hacked on-board Adaptec 7770. Although it works, disk accesses are agonisingly slow -- eg. kernel recompiles taking 10x as long as on a 486/33 with WD drives. My previous posting on this issue (re excessive interrupts) was a red herring caused by a quirk in xperfmon++. Running "bonnie" on this host and on a 16Mb PCI-based P90 with non-SCSI drive shows the following comparison: **** BENCHMARK RESULTS (bonnie) **** ** Hewlett Packard NetServer LM ** -------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 100 136 0.0 1824 0.0 582 0.0 149 0.0 1411 0.0 40.7 0.0 ** PCI Host ** -------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU 100 1452 40.5 2122 12.6 824 7.5 1329 36.3 1896 10.0 59.9 3.7 These results seem to show that the per-char throughput on read and write on the SCSI machine is only 10% of that on the PCI. I've tried these benchmarks with both a GENERIC and host-specific kernel with comparable results. The benchmarks were carried out with other SCSI devices disconnected and the network card removed. Clearly, since this machine is intended as an NFS fileserver, this kind of performance is hopeless. Hope you can help me! Many thanks, Robin. ---------------------------------------------------------------------------- Robin Melville Addiction Information Service Nottingham Alcohol & Drug Team Vox +44 (0)115 952 9478 Fax +44 (0)115 952 9421 Email robmel@nadt.org.uk WWW http://www.innotts.co.uk/nadt/ ----------------------------------------------------------------------------