Date: Thu, 29 Nov 2007 23:46:31 +0100 From: Kris Kennaway <kris@FreeBSD.org> To: Matt Reimer <mattjreimer@gmail.com> Cc: freebsd-stable@freebsd.org, Pete French <petefrench@ticketswitch.com> Subject: Re: Also seeing 2 x quad-core system slower that 2 x dual core Message-ID: <474F4147.8020904@FreeBSD.org> In-Reply-To: <f383264b0711291422h1d185f06n4834908f525d8dcf@mail.gmail.com> References: <E1IxklH-000ElU-3w@dilbert.ticketswitch.com> <474F0BDF.8070605@FreeBSD.org> <f383264b0711291106x4b431cbaj56967bb4b8762408@mail.gmail.com> <474F1105.5020708@FreeBSD.org> <f383264b0711291422h1d185f06n4834908f525d8dcf@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Matt Reimer wrote: > On Nov 29, 2007 11:20 AM, Kris Kennaway <kris@freebsd.org> wrote: >> Matt Reimer wrote: >>> On Nov 29, 2007 10:58 AM, Kris Kennaway <kris@freebsd.org> wrote: >>>> Pete French wrote: >>>>> On the dual core processors this takes about 20 seconds. On the quad >>>>> cores it takes about 3 minutes! This is true for both the 32 and 64 bit >>>>> versions of FreeBSD :-( >>>> That almost certainly has nothing to do with how many CPUs your system >>>> has, since rm -rf is a single process running on a single core. >>> I wonder if I'm seeing this too. Running super-smack on a 2 x quad >>> core 1.6GHz Dell 1950 I get about 40000 qps, whereas on a 2 x dual >>> core 3.0GHz box I've seen 80000 qps. >> Please, let's try to stay focused :) rm -rf has nothing to do with >> super-smack and vice versa. > > It's relevant to $subject. Yes but saying "I wonder if I am seeing this too" in reply to a message about a clearly unrelated issue muddies the waters. I'm not trying to be pedantic here, this kind of thing can cause considerable confusion when people aren't clear to distinguish between problems that are clearly different. >> > Is this expected? >> >> It is not very surprising. super-smack is not a good SMP benchmark, it >> does stupid things like 1-byte I/O, so it is not very scalable nor a >> good model of real-world database activity. Accounting for your CPUs >> being twice as fast on the dual core, it roughly says that the benchmark >> is not scaling beyond 4 CPUs, which is in line with my own observations. > > Is sysbench a better benchmark? It gives me 2362.99 on the 2 x > dual-core box vs 1327.26 on the 2 x quad-core box. It is, but then the issue becomes tuning of your database (mysql also has significant scaling problems in its default configuration). See http://people.freebsd.org/~kris/scaling/mysql.html for information on how to tune mysql to reduce these performance problems. I'm assuming of course that you are already using the ULE scheduler. Kris
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?474F4147.8020904>