Date: Fri, 04 Apr 2008 15:31:49 +0200 From: Kris Kennaway <kris@FreeBSD.org> To: Attila Nagy <bra@fsn.hu> Cc: "JINMEI Tatuya / 神明"@FreeBSD.ORG, =?UTF-8?B?6YGU5ZOJ?= <Jinmei_Tatuya@isc.org>, freebsd-performance@freebsd.org, bind-users@isc.org, Stefan Lambrev <stefan.lambrev@moneybookers.com> Subject: Re: Bad bind performance with FreeBSD 7 Message-ID: <47F62DC5.5010703@FreeBSD.org> In-Reply-To: <47F4E1A1.2020500@fsn.hu> References: <475B0F3E.5070100@fsn.hu> <m2lk6g71bc.wl%Jinmei_Tatuya@isc.org> <479DFE74.8030004@fsn.hu> <m2k5ltke09.wl%Jinmei_Tatuya@isc.org> <479F02A7.9020607@fsn.hu> <47F4D0DD.2040809@fsn.hu> <47F4D9F2.9070200@moneybookers.com> <47F4E1A1.2020500@fsn.hu>
next in thread | previous in thread | raw e-mail | index | archive | help
Attila Nagy wrote: > On 2008.04.03. 15:21, Stefan Lambrev wrote: >> Greetings, >> >> Attila Nagy wrote: >>> On 01/29/08 11:40, Attila Nagy wrote: >>>> ps: I have an other problem. I've recently switched from a last year >>>> 6-STABLE to 7-STABLE and got pretty bad results on the same machine >>>> with the same bind (9.4). >>>> The graphs are here: >>>> http://picasaweb.google.com/nagy.attila/20080129Fbsd6vs7Bind >>> The problem still persists and now I can provide some profiling info, >>> made by HWPMC. >>> >>> >> Sorry if you already answer this question, but at least I can find it >> in the thread. >> What scheduler are you using on RELENG_7 ? >> Did you check with both schedulers (ule/4bsd) to see which one works >> better for you? >> Also are you sure that you service the same number of requests - I see >> that the 6.x image shows CPU usage from >> Aug 2007 and 7.x image is from Jan 2008 ... is it possible, that you >> have more requests and that's why your CPU usage increased? > As for the pictures: GENERIC kernels, so 4BSD on both versions (6 and 7). > As for the profiling info: 4BSD on 6, ULE on 7 (because both were > upgraded yesterday, and ULE is now default in RELENG_7) > > The pictures are from the same timeframe (what aug 2007 refers to is the > time when the OS was compiled), the two machines were behind a per > packet load balancer, so yes: at least in pps, they've got exactly the > same traffic (of course it was possible be that one machine could serve > the answer directly from the cache, while the other had to go out, but > I've started them at the same time, so I think this effect was minimized). User time is much greater so named is doing much more work for some reason. It doesn't appear that this is a kernel problem. Verify that the config is identical, and you are not overloading it (bind doesn't scale beyond 4 threads). Kris
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?47F62DC5.5010703>