Date: Fri, 17 Feb 2006 13:09:00 -0800 From: David Leimbach <leimy2k@gmail.com> To: "Brian K. White" <brian@aljex.com> Cc: freebsd-current@freebsd.org Subject: Re: Virtual memory consumption (both user and kernel) inmodern CURRENT Message-ID: <3e1162e60602171309w4e0acf87h40c92733d38aef66@mail.gmail.com> In-Reply-To: <008101c63402$de029d10$6c00000a@venti> References: <20060215024339.N22450@atlantis.atlantis.dp.ua> <43F29BF5.4060300@freebsd.org> <20060216123548.GA35910@uk.tiscali.com> <20060216135138.GA16669@flame.pc> <43F525A6.3080701@rogers.com> <20060217013039.GA31540@xor.obsecurity.org> <43F6174A.7060801@rogers.com> <008101c63402$de029d10$6c00000a@venti>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2/17/06, Brian K. White <brian@aljex.com> wrote: > > ----- Original Message ----- > From: "Mike Jakubik" <mikej@rogers.com> > To: "Kris Kennaway" <kris@obsecurity.org> > Cc: "Giorgos Keramidas" <keramida@ceid.upatras.gr>; "Dmitry Pryanishnikov= " > <dmitry@atlantis.dp.ua>; <freebsd-current@freebsd.org>; "David Xu" > <davidxu@freebsd.org>; "Brian Candler" <B.Candler@pobox.com> > Sent: Friday, February 17, 2006 1:34 PM > Subject: Re: Virtual memory consumption (both user and kernel) inmodern > CURRENT > > > > Kris Kennaway wrote: > >> On Thu, Feb 16, 2006 at 08:23:50PM -0500, Mike Jakubik wrote: > >> > >>> And what am i trading off here? I have "/etc/malloc.conf@ -> ajz" and= my > >>> memory usage has gone up the roof. My system used to be swap free, an= d > >>> now its swapping over 40 MB. Can someone explain to me why this new > >>> malloc is better? I don't see any speed improvements. > >>> > >> > >> It's a couple of orders of magnitude faster for threaded binaries. > >> See earlier posts by the author for extensive discussion. > >> > > > > Great, too bad only 2% of my applications are threaded. I just don't se= e > > this change very positively, using 40MB of swap, where before was none > > does not seem to me like a speed improvement. I'm all for better > > performance of threaded apps, but the trade off seems too high. > > Especially if: What's orders of magnitude faster, the main parts of the a= pp > or merely the act of spawning/destroying a new thread? Logically, since this is a malloc implementation one could infer that it's "malloc" and "free" that end up orders of magnitude faster due to less locking and dedicated per-thread memory arenas. At least that's my deduction. I hope I'm correct :). Dave
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3e1162e60602171309w4e0acf87h40c92733d38aef66>