Date: Fri, 3 Jul 2009 04:31:44 -0700 (PDT) From: Barney Cordoba <barney_cordoba@yahoo.com> To: Larry Baird <lab@gta.com> Cc: freebsd-net@freebsd.org, Paul Thornton <prt@prt.org> Subject: Re: Polling and kern.polling.idle_poll Message-ID: <117993.46390.qm@web63907.mail.re1.yahoo.com>
next in thread | raw e-mail | index | archive | help
=0A=0A--- On Thu, 7/2/09, Larry Baird <lab@gta.com> wrote:=0A=0A> From: Lar= ry Baird <lab@gta.com>=0A> Subject: Re: Polling and kern.polling.idle_poll= =0A> To: "Barney Cordoba" <barney_cordoba@yahoo.com>=0A> Cc: freebsd-net@fr= eebsd.org, "Paul Thornton" <prt@prt.org>=0A> Date: Thursday, July 2, 2009, = 11:31 AM=0A> > Curious as to why you don't=0A> simply lay out the $40 for a= modern ethernet card that=0A> doesn't need polling to work well?=0A> These= are low end embedded boxs.=A0 No way to=0A> upgrade.=A0 But in actuallity= =0A> they have modern (fxp) NICS.=A0 From our testing, live=0A> lock is a r= eal issue=0A> for gateways receiving lots of network traffic.=A0=0A> Pollin= g prevents live=0A> lock.=0A=0AThere are few applications where dropping bu= ckets of packets is preferable to losing the keyboard. If "livelock is a pr= oblem", then your systems are too slow for the task.=0A=0ALivelock can easi= ty be tuned with the "work" parameter. fxp drivers are hard coded to (I thi= nk) 6000 ints per second, so its not rocket science to set your tuning para= meters to whatever the system can handle before entering a livelock scenari= o. In reality, interrupt moderation is the equivalent of polling 6000 times= per second, in the case of fxp. You wouldn't want to set hz to 6000 as you= substantially increase system overhead. Moderation is a much more efficien= t way of managing your ethernet loads, because the "polls" are only directe= d at the driver, and if you dont have traffic then you don't need to get p= olled.=0A=0ABarney=0A=0A=0A
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?117993.46390.qm>