Date: Mon, 14 Sep 2009 01:58:39 -0700 (PDT) From: Barney Cordoba <barney_cordoba@yahoo.com> To: Bruce Evans <brde@optusnet.com.au>, rihad <rihad@mail.ru> Cc: freebsd-net@FreeBSD.org Subject: Re: [POLLING] strange interrupt/system load Message-ID: <676338.40771.qm@web63905.mail.re1.yahoo.com> In-Reply-To: <4AADF2D8.5050505@mail.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
=0A=0A--- On Mon, 9/14/09, rihad <rihad@mail.ru> wrote:=0A=0A> From: rihad = <rihad@mail.ru>=0A> Subject: Re: [POLLING] strange interrupt/system load=0A= > To: "Bruce Evans" <brde@optusnet.com.au>=0A> Cc: freebsd-net@FreeBSD.org= =0A> Date: Monday, September 14, 2009, 3:38 AM=0A> Bruce Evans wrote:=0A> >= On Sat, 12 Sep 2009, rihad wrote:=0A> > =0A> >> The box experiences ~230 m= bit/s traffic flow=0A> through it. I've doubled some sysctls after reading= =0A> polling(4):=0A> >> kern.polling.each_burst=3D10 # was: 5=0A> >> kern.p= olling.burst_max=3D350 # was: 150=0A> >> =0A> >> FreeBSD 7.2-RELEASE-p3 amd= 64=0A> >> HZ=3D1000=0A> > =0A> > How much better does it work without POLLI= NG?=0A> > =0A> Without polling (current load around 190-200 mbit/s, around= =0A> 24-26 kpps):=0A> =0A> top:=0A> CPU:=A0 0.0% user,=A0 0.0% nice,=A0 8.4= %=0A> system,=A0 0.0% interrupt, 91.6% idle=0A> =0A> Interrupts/s: 18322 to= tal=0A> 28 mpt0 irq16=0A> 1999 cpu0: time=0A> 6906 em0 irq256=0A> 3392 em1 = irq257=0A> 1999 cpu1: time=0A> 1999 cpu2: time=0A> 1999 cpu3: time=0A=0AYou= really need to look at the taskq usage as averaging on a 4 core=0Asystem m= uddies things up. em will generally run on 1 core per NIC, and=0Ainterrupts= are filtered so you won't see any interrupt usage. On a =0A4 core system y= ou could exhaust a core and still be at 25% overall, so =0Ayou need to watc= h the max usage per core.=0A=0AThings aren't measured properly in polling m= ode so its difficult to=0Acompare them one to one. You really don't need to= ; intuitively it makes=0Azero sense to use polling with em.=0A=0AYou'll do = a lot better setting your ITR to 2000 or so. You really don't=0Aneed an int= errupt every 4 packets at those traffic levels.=0A=0ABarney=0A=0A=0A
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?676338.40771.qm>