Date: Wed, 19 Apr 2000 10:32:26 +0930 From: Greg Lehey <grog@lemis.com> To: Mark Costlow <cheeks@swcp.com> Cc: freebsd-scsi@FreeBSD.ORG Subject: Re: vinum write performance Message-ID: <20000419103225.B92373@freebie.lemis.com> In-Reply-To: <200004171846.MAA13065@shimi.swcp.com> References: <200004171846.MAA13065@shimi.swcp.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Monday, 17 April 2000 at 12:46:55 -0600, Mark Costlow wrote: > > I'm setting up a freeBSD 3.4 system on a Micron Netframe (PIII-533, 256M) with > an Adaptec SCSI controller on the motherboard (model number escapes me right > now). There are two SCSI disks connected, both are IBM DNES-318350Y's (18G, > 7200 RPM). The system boots off an IDE disk. > > My intention is to configure the SCSI disks as a RAID-1 mirror with vinum. I > did this successfully, however the write performance was really bad, much > worse than I expected from the papers I've read on vinum performance, so I > think I must have something configured badly. > > The simple test I've been running to guage performance is 'iozone 500' (old > version of iozone that just does a sequential write followed by a sequential > read). I understand why RAID-1 writes will be slower than raw single-drive > writes, but the difference here is a factor of 7, not 2 (and with only two > drives on the bus I wouldn't expect much of a slowdown). You really want to use rawio, not iozone. > When I first noticed that writing to the vinum mirror volume was > slow, I thought the problem might be one or both of the disks. So I > reconfigured vinum to use just one disk (each in turn) and saw the > same performance on each one. I also put regular file system on > each disk to test them and saw more normal write performance. Here > are the results of those tests: > > Throuput numbers are in MB/s > iozone 500: FS Type Read Write > ------------------------------------- > da0: vinum 19.3 2.2 > da0: BSD 19.4 15.2 > da1: vinum 18.6 2.2 > da1: BSD 17.4 14.0 This doesn't look good. > The newfs parameters in all cases are the system defaults. The vinum config > being used is: > > drive d0 device /dev/da1s1e > volume vtest > plex name vtest.p0 org concat vol vtest > sd name vtest.p0.s0 drive d0 plex vtest.p0 len 35843405s driveoffset 265s plexoffset 0s > > The write speed I see on a vinum volume with a single plex and sd is > the same as I saw with the drives in a mirror config: 2.2MB/s. As > the numbers above show, the problem isn't specific to one drive. If > the writes go through vinum, they're slow. This puzzles me. I can't think of any good reason for this. > FWIW, regarding the recent discussion about the drive's write cache: > the write cache is enabled on these drives. > > If anyone has any ideas what I might be doing wrong, or if I've left > out any useful configuration information, please let me know. I don't know where to start with this one, but how about trying rawio (/usr/ports/benchmarks/rawio) and telling me the results of that. Also, for the sake of reference, the info I ask for at http://www.lemis.com/vinum/how-to-debug.html. 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-scsi" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000419103225.B92373>