Date: Sun, 21 May 2006 10:52:10 -0700 From: Jason Evans <jasone@FreeBSD.org> To: Attila Nagy <bra@fsn.hu> Cc: freebsd-current@freebsd.org Subject: Re: jemalloc paper/slides available Message-ID: <4470A8CA.7010203@FreeBSD.org> In-Reply-To: <4470A58E.1050101@fsn.hu> References: <8E2A65C9-1CEA-452F-B8BB-CAEB573F60DD@freebsd.org> <4470A58E.1050101@fsn.hu>
next in thread | previous in thread | raw e-mail | index | archive | help
Attila Nagy wrote: > On 2006. 05. 18. 19:19, Jason Evans wrote: > >> The paper and slides for jemalloc that I presented at BSDcan 2006 are >> available at: >> http://people.freebsd.org/~jasone/jemalloc/bsdcan2006/ >> If you read only one or the other, choose the paper. > > Thank you. Here is a quick comparison (the article is in hungarian) on a > Sun T2000 with Solaris and Linux and an Intel dual core low voltage Xeon > (Sossaman, Core Duo as a server processor): > http://hup.hu/node/25322 > > I've lost the results for FreeBSD 6 for different blocksizes, so only > 1kB is on the picture, which contains all OSs. With 1MB, you can see > only Solaris and Linux on the same machine, besides FreeBSD -CURRENT. > > Do you have any ideas about the sudden drop after 16 threads? (the > machine had four cores) I doubt this is an issue for the benchmark, but it's possible that the default of 16 arenas (on a 4-CPU system) is causing the dropoff. You can test this by setting MALLOC_OPTIONS=N in the environment (each N doubles the number of arenas). Jason
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4470A8CA.7010203>