Date: Mon, 13 Feb 2006 10:42:04 -0500 (EST) From: Andrew Gallatin <gallatin@cs.duke.edu> To: Robert Watson <rwatson@FreeBSD.org> Cc: Poul-Henning Kamp <phk@phk.freebsd.dk>, current@FreeBSD.org Subject: Re: [TEST/REVIEW] cpu time accounting patch, step 2 Message-ID: <17392.43212.367624.807423@grasshopper.cs.duke.edu> In-Reply-To: <20060211005438.E90460@fledge.watson.org> References: <17388.44976.250463.383429@grasshopper.cs.duke.edu> <5383.1139586916@critter.freebsd.dk> <17388.60202.862312.337026@grasshopper.cs.duke.edu> <20060211005438.E90460@fledge.watson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Robert Watson writes: > > On Fri, 10 Feb 2006, Andrew Gallatin wrote: > > > Unfortunately, even after your patch, we are still about 38% slower than > > linux x86_64 on the same box for loopback ping-pong, and 32% slower for > > ping-pong over 10GbE. (bandwidth is lower for streaming tests, and CPU > > utilization is much, much much higher in FreeBSD as well). > > > > I think you nailed the biggest source of overhead, but there is apparently a > > lot more performance that we can get out of the hardware. I'd love to see > > you commit this. > > I can't remember if I pointed you at this before, but I remember us talking > about it by e-mail. What happens to your loopback performance if you compile > PREEMPTION out of the kernel? As long as I have machdep.cpu_idle_hlt=0 and am using the BSD scheduler, disabling preemption does not help. If I do disable preemption, then I can also enable machdep.cpu_idle_hlt without a penalty. Drew
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?17392.43212.367624.807423>