Date: Thu, 10 Jan 2019 09:45:24 +0100 From: Kristof Provost <kp@freebsd.org> To: Gleb Smirnoff <glebius@freebsd.org> Cc: "Andrey V. Elsukov" <ae@freebsd.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r342908 - head/sys/netpfil/ipfw Message-ID: <20190110084524.GH49515@vega.codepro.be> In-Reply-To: <20190110073827.GE18452@FreeBSD.org> References: <201901100147.x0A1lvQX035635@repo.freebsd.org> <20190110073827.GE18452@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2019-01-09 23:38:27 (-0800), Gleb Smirnoff <glebius@freebsd.org> wrote: > On Thu, Jan 10, 2019 at 01:47:57AM +0000, Andrey V. Elsukov wrote: > A> glebius@ has reported that they at Netflix discovered, that initialization > A> of this variable produces significant overhead on packet processing. > A> After patching I managed to increase performance of packet processing on > A> simple routing with ipfw(4) firewalling to about 11% from 9.8Mpps up to > A> 11Mpps (Xeon E5-2660 v4@ + Mellanox 100G card). > > And with ipfw being processed at driver level, rather than ip_input() > the boost is even more dramatic than 11%. > > https://github.com/glebius/FreeBSD/commits/pfil > > Soon to reach FreeBSD head :) > Nice! That pfil epoch-ification change should help pf too. Kristof
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20190110084524.GH49515>