Date: Wed, 22 Nov 2006 09:44:05 -0600 From: Dave <kreios@gmail.com> To: O. Hartmann <ohartman@mail.zedat.fu-berlin.de> Cc: freebsd-performance@freebsd.org, Marcelo Gardini do Amaral <marcelo@registro.br>, Robert Watson <rwatson@FreeBSD.org> Subject: Re: DNS Performance Numbers Message-ID: <77EFE6C7-96D4-4509-AC03-63F1AAE9BBA4@gmail.com> In-Reply-To: <455F1021.6040004@mail.zedat.fu-berlin.de> References: <20061030192702.GG76994@registro.br> <20061111091844.I63959@fledge.watson.org> <20061116164053.GR57732@registro.br> <455F1021.6040004@mail.zedat.fu-berlin.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On Nov 18, 2006, at 7:52 AM, O. Hartmann wrote: > Marcelo Gardini do Amaral wrote: >>> FYI: In response to feedback from ISC, there are UDP transmit >>> optimizations >>> in FreeBSD 7.x. These have a relatively minor performance impact >>> for >>> single-threaded applications, but in the special case of BIND >>> accessing a >>> single UDP socket from many different threads, it significantly >>> improves >>> performance. I'll look at MFC'ing these to 6.x after 6.2-RELEASE >>> (especially if reminded in a month or so :-). >>> >>> With regard to the possible bge issue -- I would encourage you to >>> test >>> using a 7.x kernel, ideally with all the debugging disabled, and >>> see if >>> there's been any improvement (or regression). There has been a >>> lot of >>> change in these areas, and it would be helpful to know what, if >>> any, impact >>> this has had. >>> >> >> >> I made some tests using 7.x with all the debugging disabled: >> >> queries / s >> >> Int bind (d_t) bind (e_t) nsd (1_s) nsd (2_s) >> --- ---------- ---------- --------- --------- >> >> bge 15439 14733 12910 10946 >> em 37655 34092 42411 41974 >> >> >> d_t: disable threads >> e_t: enable threads (libpthread) >> 1_s: 1 server forked >> 2_s: 2 server forked >> >> Bind: 9.2.3 >> NSD: 3.0.2 >> em: Dell 1950, Intel NIC, SMP kernel >> bge: HP Blade BL35p, Broadcom NIC, SMP kernel >> Client: Dell 1750, Intel NIC, FreeBSD 4.11 UP, running queryperf >> >> >> >> The results are very good for em NIC, better than my numbers [1] with >> FreeBSD 6.1 some months ago. So I guess that we had an >> improvement :-) >> >> But I got the same poor performance with the bge interface. The >> problem remains. >> >> [1] http://lists.freebsd.org/pipermail/freebsd-net/2006-September/ >> 011767.html >> >> Cheers, >> Marcelo >> >> > > > These results looks very puzzling to me. > As far as I know, multithreading and/or multiprocessors should perform > better anyway than a single threaded application within other > applications on an UP box. Strange results ...And more strange than > this > is the result taken from the FBSD 4.11 box! Is there an explanation > why > FreeBSD performs so bad beyond 4.X and on SMP boxes? Please show me > threads ... > > Thanks and regards, > Oliver The FreeBSD pthread library and BIND don't work well together. If you use the libthr library, performance goes up. -- Dave
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?77EFE6C7-96D4-4509-AC03-63F1AAE9BBA4>