From owner-freebsd-fs Tue Feb 3 22:21:11 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id WAA24577 for freebsd-fs-outgoing; Tue, 3 Feb 1998 22:21:11 -0800 (PST) (envelope-from owner-freebsd-fs@FreeBSD.ORG) Received: from master.inter-linc.net ([12.10.101.2]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id WAA24570 for ; Tue, 3 Feb 1998 22:21:06 -0800 (PST) (envelope-from cdillon@inter-linc.net) Received: from cheetah.inter-linc.net (207.3.81.148) by master.inter-linc.net (Worldmail 1.3.167) for freebsd-fs@freebsd.org; 4 Feb 1998 00:19:37 -0600 Message-ID: X-Mailer: XFMail 1.1 [p0] on FreeBSD Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 8bit MIME-Version: 1.0 Date: Wed, 04 Feb 1998 00:02:01 -0600 (CST) From: Chris Dillon To: freebsd-fs@FreeBSD.ORG Subject: May have pounced on something weird with ccd and newfs (rather old) Sender: owner-freebsd-fs@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org X-To-Unsubscribe: mail to majordomo@FreeBSD.org "unsubscribe freebsd-fs" A short time ago (late November or early December of 1997, about when these maillists started bouncing mail for people that didn't have valid reverse lookups) I tried to post a bit about the tests I I did to figure out which ccd stripe size would probably work best with my equipment. Running 2.2.5-stable, I set up two new IBM DCAS-34330W's (4.3GB UW-SCSI) on a Tekram DC-390F and tried setting various stripe sizes with ccd and running bonnie on the resulting filesystems. I did something like this: ccdconfig -c ccd0 $BLOCKSIZE 0 /dev/sd0s1g /dev/sd1s1g newfs -b 8192 -f 1024 ccd0c mount /dev/ccd0c /mnt bonnie -d /mnt -s 100 -m "I=$BLOCKSIZE" | grep "I=" >> ~/ccdtest.txt umount -f /dev/ccd0c (only did -f since weirdly enough, it would fail sometimes) ccdconfig -u ccd0 BLOCKSIZE varied from 0 to 8192, in powers of two (0, 16, 32, 64, etc.). Here is the resulting output of the above. -------Sequential Output-------- ---Sequential Input-- --Random-- -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--- Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU std 100 3462 97.9 7636 40.8 2976 28.6 3732 98.6 7942 32.3 153.9 7.2 I=0 100 3422 98.0 7311 37.1 2905 28.0 3707 98.5 7404 29.8 156.8 7.5 I=16 100 3319 96.1 12043 81.3 4515 45.4 3626 97.3 12683 66.4 167.3 8.2 I=32 100 3418 97.2 13433 87.8 4747 46.1 3671 98.3 14597 77.5 164.5 8.1 I=64 100 3413 97.0 13986 89.9 4844 47.2 3700 98.9 14703 74.8 166.9 8.2 I=128 newfs -b 8192 -f 1024 fails with "write error: 2047871 wtfs: invalid argument". I=256 100 3433 97.3 12782 92.2 4912 52.7 3718 98.9 13539 88.9 165.9 9.3 I=512 100 3474 98.1 12391 86.7 6107 66.6 3746 98.8 12306 75.7 166.6 9.3 I=1024 100 3473 98.0 12403 91.1 4449 49.5 3735 98.5 9870 57.7 170.4 9.6 I=2048 100 3453 98.0 11876 88.2 3493 39.1 3752 98.8 8397 47.0 168.9 9.5 I=4096 100 3449 98.1 11851 84.0 3217 35.6 3708 98.2 7922 44.2 169.3 9.5 I=8192 100 3464 97.8 12011 87.9 3116 32.8 3745 99.0 7652 42.4 174.8 9.9 "std" is the output when NOT using ccd (i.e., a regular disk). With a ccd stripe size of 128, newfs refused to create a filesystem with the error you see above. (I decided on using the 64 block stripe ultimately, by the way.) Is this a known anomaly, or did I stumble onto some underlying problem? If this is common knowledge, please enlighten me. :-) This also seemed like the appropriate maillist to post this to (I'm on just about ALL of them), but if not, I'll move it elsewhere. --- Chris Dillon --- cdillon@inter-linc.net --- Powered by FreeBSD, the best operating system on the planet for Intel x86 based computers (and soon Sparcs). ---- (http://www.freebsd.org)