Date: Sat, 9 Sep 2000 11:33:19 +0930 From: Greg Lehey <grog@lemis.com> To: Esko Petteri Matinsola <esko@kapi.net>, Clark Shishido <clark@desktop.com> Cc: hackers@FreeBSD.ORG Subject: Re: Vinum RAID-5 performance problem Message-ID: <20000909113319.E5876@wantadilla.lemis.com> In-Reply-To: <Pine.BSF.4.10.10009082147290.46483-100000@sputnik>; from esko@kapi.net on Fri, Sep 08, 2000 at 10:24:51PM %2B0300 References: <Pine.BSF.4.10.10009082147290.46483-100000@sputnik> <v0422080bb5def254254a@[192.168.1.42]> <Pine.BSF.4.10.10009082147290.46483-100000@sputnik>
next in thread | previous in thread | raw e-mail | index | archive | help
On Friday, 8 September 2000 at 22:24:51 +0300, Esko Petteri Matinsola wrote: > Thank you for your last reply about the Promise 100 controller, I got it > working with that patch. > > But now, I have Asus P2L97 mobo with 64MB RAM, 266MHz P2 and 4 Maxtor > 54098H8's, two per channel. OS is FreeBSD 4.1-STABLE. > > When I use vinum with the following config: > > drive drive0 device /dev/ad0e > drive drive1 device /dev/ad1e > drive drive2 device /dev/ad2e > drive drive3 device /dev/ad3e > > > volume raid0 > plex org striped 256k > sd length 512m drive drive0 > sd length 512m drive drive1 > > > so I have RAID-0 with the master and slave drive from controller one. > bonnie -s 512 on that gives me: > > > -------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 > 512 13287 97.3 20237 41.0 8023 25.6 8572 97.8 21442 29.4 128.1 2.2 Please don't use bonnie. It's not measuring what you think it is. > not bad, even when two drives on the same controller. Next, I tried vinum > with the following config: > > drive drive0 device /dev/ad0e > drive drive1 device /dev/ad1e > drive drive2 device /dev/ad2e > drive drive3 device /dev/ad3e > > volume raid0 > plex org striped 256k > sd length 512m drive drive0 > sd length 512m drive drive2 > > so I have RAID-0 with the masters from both controllers. bonnie -s 512 > gives me: > > -------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 > 512 13263 97.4 21340 43.7 8169 26.1 8572 98.0 25858 36.6 213.4 3.9 > > only a little better, except the seeks. Of course. The seeks are the only thing which relate to real-life performance. Look at the sequential character output, for example. It's showing clearly that the bottleneck is the CPU, not the storage subsystem. In block writes, the real-world performance of the first configuration is in fact worse than what you see there, but since bonnie is only writing one file at a time, you don't run into any contention problems. You should be using rawio, which will show you what the storage system is doing. > Now next to my problem. When I use vinum with the following config: > > drive drive0 device /dev/ad0e > drive drive1 device /dev/ad1e > drive drive2 device /dev/ad2e > drive drive3 device /dev/ad3e > > volume raid5 > plex org raid5 256k > sd length 512m drive drive0 > sd length 512m drive drive1 > sd length 512m drive drive2 > sd length 512m drive drive3 > > > so I have RAID-5 with all the drives from both controllers. bonnie -s 512 > gives me: > > > -------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 > 512 2368 17.7 2358 4.8 2016 6.6 8218 94.4 20068 29.1 232.5 4.3 > > read performance and seeks seems OK, but block writes only about 1/9th ! > As I read from www.vinumvm.org I should get something like 5MBps block > writes. I don't know how you'd find anything at vinumvm.org which tells you what to expect from bonnie. > Have I misunderstood something, configured someting improperly, it's > because of only two IDE-controllers or why I get so lousy write > performance ? Well, firstly I don't know what your write performance is. It really depends on what you're trying to do. But yes, having only two controllers will halve your RAID-5 write performance. Note also that with a 256 kB stripe, you'll run into drive contention problems because all your superblocks will be on the same subdisk. Take a size like 273 kB, for example. > I could live with 5MBps, if I only could get that ! :) I'd expect the performance to increase by about 50% or 60% if you use four controllers. On Friday, 8 September 2000 at 12:55:51 -0700, Clark Shishido wrote: > At 22:24 +0300 2000.09.08, Esko Petteri Matinsola wrote: >> >> read performance and seeks seems OK, but block writes only about 1/9th ! >> As I read from www.vinumvm.org I should get something like 5MBps block >> writes. >> >> Have I misunderstood something, configured someting improperly, it's >> because of only two IDE-controllers or why I get so lousy write >> performance ? > > Your configuration looks fine. Well, using master and slave together isn't fine in my book. > I just recently tried a similar setup using Promise Ultra66 controllers > with some Quantum and IBM drives (tried both). I got lousy performance > doing sustained writes to a RAID5 volume, using 4 UDMA66 drives each a > master on a controller. just doing a cat /dev/zero > blah > > I just decided to stick with striping and make regular backups. > I may try vinum with RAID5 on a SCSI array later. > > FWIW, > FreeBSD 4.1-STABLE > Abit BP6 dual 466 Celerons > (not using the HPT controller) > Promise Ultra66 > Quantum KA 13.6 gig and IBM Deskstar 30 gig drives Again, it would be nice to see some real figures. Why aren't you using the onboard controller on the BP6? 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-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000909113319.E5876>