Date: Thu, 7 Sep 2000 10:27:40 +0930 From: Greg Lehey <grog@lemis.com> To: Mike Smith <msmith@FreeBSD.ORG> Cc: Mike Tancsa <mike@sentex.net>, Tim Tsai <tim@futuresouth.com>, stable@FreeBSD.ORG Subject: Re: Vinum (was: RAID) Message-ID: <20000907102740.B7718@wantadilla.lemis.com> In-Reply-To: <200009052339.QAA03602@mass.osd.bsdi.com>; from msmith@FreeBSD.ORG on Tue, Sep 05, 2000 at 04:39:48PM -0700 References: <20000906084835.B21113@wantadilla.lemis.com> <200009052339.QAA03602@mass.osd.bsdi.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tuesday, 5 September 2000 at 16:39:48 -0700, Mike Smith wrote: >>> Just about to test and compare the same Atlas IV drives (QUANTUM ATLAS IV 9 >>> WLS 0B0B) usings a 2940U2 (aic7890/91) with vinum. I will post the results >>> with the same three SCSI drives in the various adaptors in RAID5 (3 drive >>> config) when done. >> >> Please make sure to use rawio in the tests. Other things, such as >> bonnie and dd, give better looking results, but they're not very >> related to what happens in practice. > > Rawio is largely a seek-time tester, Correct. That's the biggest issue in disk I/O. > and not particularly representative of real-world I/O loads either. It's reasonably close. It could always be better, of course, and as discussed I'm planning to write code to snapshot real-life disk activity. In the meantime, I think the current version is relatively accurate. What specific problems do you see? > To get a feel for how a controller interacts with the buffer cache, > you need to present the load via the filesystem. No, you just need to present the same kind of load. > I typically use a large number of bonnie processes (20 or so) > working with large datasets (1-10GB) and then ignore their output > and instead monitor the controller's throughput using devstat. > Other filesystem load generators are probably worth considering as > well, since bonnie's I/O patterns are not very imaginiative. Nor are they particularly representative of real-world I/O loads :-) In addition, monitoring devstat output requires a lot of interpretation. Greg -- Finger grog@lemis.com for PGP public key See complete headers for address and phone numbers To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000907102740.B7718>