From owner-freebsd-hackers Wed Oct 4 12:27:21 1995 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.6.12/8.6.6) id MAA00947 for hackers-outgoing; Wed, 4 Oct 1995 12:27:21 -0700 Received: from epsilon.qmw.ac.uk (epsilon.qmw.ac.uk [138.37.6.3]) by freefall.freebsd.org (8.6.12/8.6.6) with SMTP id MAA00920 for ; Wed, 4 Oct 1995 12:26:30 -0700 Received: from canary.dcs.qmw.ac.uk by epsilon.qmw.ac.uk with SMTP-DNS (PP) id <06351-0@epsilon.qmw.ac.uk>; Wed, 4 Oct 1995 19:31:48 +0100 Received: from ruby.dcs.qmw.ac.uk [192.135.231.243] by canary.dcs.qmw.ac.uk (8.6.12/QMW-server-2.4s) with SMTP; poster "Mark Dawson "; id TAA28993; Wed, 4 Oct 1995 19:27:17 +0100 Received: locally by ruby (4.1/QMW-client-3.2b); for "md@dcs.qmw.ac.uk"; poster "md"; id AA15138; Wed, 4 Oct 95 19:31:34 BST Received: from Messages.8.5.N.CUILIB.3.45.SNAP.NOT.LINKED.ruby.cs.qmw.ac.uk.sun4.41 via MS.5.6.ruby.cs.qmw.ac.uk.sun4_41; Wed, 4 Oct 1995 19:31:34 +0100 (BST) Message-Id: Date: Wed, 4 Oct 1995 19:31:34 +0100 (BST) From: Mark Dawson To: hackers@FreeBSD.org Subject: A RAID solution for FreeBSD Sender: owner-hackers@FreeBSD.org Precedence: bulk Now I have a few spare minutes I thought I should say "thank you" to everyone on the list who has helped me out over the last couple of months - I wouldn't have got this thing flying without your help! Special thanks must go to Stefan for making changes to the PCI probe code that made the following possible... At QMW we have a large lab. of over 100 (legacy) Macintoshes for our students. Over the summer we've given this a major overhawl. The Macs are now connected on 18 segments to a 3com switched-hub with a 100Mbps link to a FastEthernet network comprising 3 compute servers (memory-rich Sparc-20s) and a "student" file server (Compaq Proliant 1500/120 running FreeBSD). >From my point-of-view the fun part is the Compaq. The Compaq Smart Array Controller (EISA) does the RAID completely transparently to FreeBSD. It's configured with five 4.3Gb disks in a single RAID-5 "logical disk" of 16GB and a sixth disk as a "hot-spare" (and plenty of empty slots for future expansion.) The controller's battery-backed 4MB cache memory means disk accesses complete once data has been moved into the controller. This gives alarming "rm -rf *" performance :-). Better, using SMCs PCI FastEthernet cards I get the following iozone results from a compute server via NFS: IOZONE: Performance Test of Sequential File I/O -- V1.16 (10/28/92) By Bill Norcott Operating System: SunOS -- using fsync() IOZONE: auto-test mode MB reclen bytes/sec written bytes/sec read 1 512 1263974 10025968 1 1024 1321070 17049731 1 2048 1269425 25934936 1 4096 1155822 29564016 1 8192 1273568 42032068 2 512 1128729 9577919 2 1024 1255028 15582247 2 2048 1221635 22480418 2 4096 1273994 29275908 2 8192 1150043 34849744 4 512 1219658 9621994 4 1024 1225791 15527906 4 2048 1222215 22804554 4 4096 1227209 16358889 4 8192 1270302 34678311 8 512 1161705 9623274 8 1024 1247379 15725818 8 2048 1199133 23001643 8 4096 1210558 29912095 8 8192 1191661 35262549 16 512 1177531 5421519 16 1024 1208808 15163076 16 2048 1204631 23017771 16 4096 1191427 29986625 16 8192 1248278 19627731 Completed series of tests and these locally on the Compaq: IOZONE: Performance Test of Sequential File I/O -- V1.16 (10/28/92) By Bill Norcott Operating System: POSIX 1003.1-1988 IOZONE: auto-test mode MB reclen bytes/sec written bytes/sec read 1 512 5368709 14913080 1 1024 7456540 19173961 1 2048 8947848 22369621 1 4096 10324440 16777216 1 8192 13421772 11184810 2 512 5064819 14128181 2 1024 7456540 19173961 2 2048 9942053 7064090 2 4096 9586980 4709393 2 8192 14913080 3397917 4 512 5368709 5965232 4 1024 7780737 5113056 4 2048 9099506 4511520 4 4096 10324440 4098251 4 8192 12485370 4436949 8 512 4994148 13256071 8 1024 7724761 10424677 8 2048 9850842 5187158 8 4096 10845877 4628197 8 8192 14708792 4006499 16 512 5450466 12413200 16 1024 7724761 17459216 16 2048 4051855 20259279 16 4096 4329604 20259279 16 8192 4782814 22369621 Completed series of tests (not sure how to interpret these!) The de driver does a good job of shipping packets - I've seen ~3000 per second at peak - although the NFS test shown above generated about 900. Netatalk on the Compaq also makes an amazingly good AFP server for the MacOS Macintoshes. I'm talking to people at Compaq to get the go-ahead to contribute my FreeBSD driver for their Smart controller. So far no response. Anyone from Compaq out there? Mark