Date: Mon, 20 Mar 2000 12:23:26 -0800 From: Greg Lehey <grog@lemis.com> To: fbsd-dave <fbsddave@mrcaffeine.com> Cc: Mark Smith <msmith@beta.tricity.wsu.edu>, freebsd-questions@FreeBSD.ORG Subject: Re: vinum stats and question(addendum) Message-ID: <20000320122326.I2898@mojave.worldwide.lemis.com> In-Reply-To: <Pine.BSF.4.21.0003180634400.53717-100000@boris.netgate.net>; from fbsddave@mrcaffeine.com on Sat, Mar 18, 2000 at 06:54:48AM -0800 References: <200003180747.XAA29239@beta.tricity.wsu.edu> <Pine.BSF.4.21.0003180634400.53717-100000@boris.netgate.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Saturday, 18 March 2000 at 6:54:48 -0800, fbsd-dave wrote: > On Fri, 17 Mar 2000, Mark Smith wrote: > >> Here's the revised vinum benchmarks using both rawio (the first batch) >> and bonnie (the second). In each case, the first 2 entries are >> the seperate drives. Then comes the vinum volume, testc. I ran >> striping at 128b, 256b, 512b and 1024b. >> >> In all cases, vinum absolutly dies in the sequential write, at least in >> my case. >> >> Any comments? >> >> first is the conf file. >> ------------------------------------------------------------ >> drive drive1 device /dev/da1s2e >> drive drive2 device /dev/da2s2e >> >> volume testc >> plex org striped 1024b >> sd length 5169m drive drive1 >> sd length 5169m drive drive2 >> ------------------------------------------------------------ >> >> root$ rawio -av /dev/da1s2e >> Random read Sequential read Random write Sequential write >> ID K/sec /sec K/sec /sec K/sec /sec K/sec /sec >> anon 709.9 44 44065.0 2690 877.6 54 6575.6 401 >> root$ rawio -av /dev/da2s2e >> anon 707.9 44 49995.6 3051 931.5 58 6818.3 416 >> root$ rawio -av /dev/vinum/testc #128b >> anon 1082.3 67 40147.9 2450 1031.1 64 1944.5 119 >> root$ rawio -av /dev/vinum/testc #256b >> anon 1090.6 68 37885.9 2312 1006.4 62 1946.2 119 >> root$ rawio -av /dev/vinum/testc #512b >> anon 1112.6 69 38752.4 2365 1009.6 62 1948.3 119 >> root$ rawio -av /dev/vinum/testc #1024b >> anon 1129.3 69 38876.6 2373 1000.1 62 1942.4 119 >> >> Bonnie -s 200 > [omitted; the values are unintersting] > > "b"? I would guess all your writes are walking on each other. I'd guess you assume 'b' means 'byte'. In fact, it's a deprecated abbreviation for 'sector', so the stripe size is in fact 512 kB, which is reasonable. > And with power-of-two stripe sizes all your superblocks will be on > one drive. Right, but that's not the problem here. rawio doesn't do any file I/O, though I should probably add some options for it. I'm puzzled by the slow sequential write performance. Is it possible that you (Mark) are doing this with FreeBSD 3.x? In that case, you're running the disk tests against the block device, which caches. Use /dev/rda1s2e and /dev/rda2s2e and things will be more comparable. > That's a slow combination. Try some tests again but starting at 128k, and > try, for instance, 129k to see what it does. Greg is working on a program > to figure ideal sizes so one day that'll be easier. Also, I notice your > processor is mostly maxed. I think it's spending itself on the small > stripe transactions. A faster processor could help but that would be the > wrong fix IMO. I'd be surprised if the processor were the bottleneck. You can check that with the verbose output option; I'd guess that you'd be using less than 5% CPU in these tests. The maxing in the bonnie tests is because bonnie usually tests the processor, not the I/O subsystem. Greg -- When replying to this message, please copy the original recipients. For more information, see http://www.lemis.com/questions.html 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-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20000320122326.I2898>