Date: Fri, 9 Apr 2004 10:55:03 -0700 From: Luigi Rizzo <luigi@freebsd.org> To: Ruslan Ermilov <ru@freebsd.org> Cc: net@freebsd.org Subject: Re: polling(4) and rl(4) Message-ID: <20040409105503.A35357@xorpc.icir.org> In-Reply-To: <20040409164724.GD2461@ip.net.ua>; from ru@freebsd.org on Fri, Apr 09, 2004 at 07:47:24PM %2B0300 References: <20040409164724.GD2461@ip.net.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Apr 09, 2004 at 07:47:24PM +0300, Ruslan Ermilov wrote: > Hey Luigi, > > Have you actually measured the performance of rl(4) with polling(4) > enabled? With 8139 anomaly of four (register based) TX descriptors no, nor i did expect any improvement -- the code was only there to help when the 8139C+ was supported. But now that happens in a different driver. Re. the removal, I still think it is beneficial in receiving, (not performancewise, just to avoid livelock), so as a temporary measure why don't you just short-circuit the logic that enables polling in the driver rather than ripping it out completely ? cheers luigi (rushing out for dinner) > it's becoming a royal pain in the ass with polling(4) enabled -- > the TX performance just sucks -- I could only get the comparable > results with HZ=5000, which is overheating my CPU. My suggestion > is to drop polling(4) support from the rl(4) driver completely. > > Are there any objections? Has anybody got different results with > rl(4) and polling(4) enabled? > > Having it in re(4) is of course a good thing. ;) > > As an aside, I've started working on the ``[-]polling'' option for > ifconfig(8) that, when done, will allow changing the polling status > of individual interfaces in run-time, e.g., the following command > will disable polling on nge0: > > ifconfig nge0 -polling > > > Cheers, > -- > Ruslan Ermilov > ru@FreeBSD.org > FreeBSD committer
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040409105503.A35357>