Date: Mon, 27 Jan 2020 12:20:59 -0800 From: Gleb Smirnoff <glebius@freebsd.org> To: Hans Petter Selasky <hps@selasky.org> Cc: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r357004 - in head/sys: kern sys Message-ID: <20200127202059.GV1268@FreeBSD.org> In-Reply-To: <7d7db96d-26b1-1d2b-9f8d-a3f8fbe8c33c@selasky.org> References: <202001230124.00N1OlXi029506@repo.freebsd.org> <fb6e84b6-85a9-3252-c27a-bb9ee7ca203f@selasky.org> <23f272a4-c997-a454-19d6-10392713e71f@selasky.org> <20200124170532.GO1268@FreeBSD.org> <7d7db96d-26b1-1d2b-9f8d-a3f8fbe8c33c@selasky.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Hans, On Fri, Jan 24, 2020 at 08:21:54PM +0100, Hans Petter Selasky wrote: H> > If a driver has interrupt moderation than epoch batching counter H> > basically won't ever grow over 1. It kicks in only of driver doesn't H> > have it, or receives interrupts at a very high rate. H> H> Depending on the load an interrupt imposes, this batching counter may H> cause epochs to last for a long time. Have you considered using ticks H> for this instead? I.E. if the congestion lasts more than two ticks, then H> re-acquire the EPOCH? H> H> For example if the network controller spends more time processing H> packets then there is between interrupts, then this unconditional 1000x H> thing can be quite dangerous. I didn't try using ticks instead of execution counter. It could be ticks would be better. Needs more experiments. H> > H> 2) You need to make a new request function for interrupts which take a H> > H> pointer to an EPOCH and replace that IH_NET in hflags! H> > H> > Initially I did that way, but then pondered over this approach and have H> > abandoned it. Most likely we will have just few globally recognized H> > epochs in the kernel. And even less might be associated with interrupt H> > handlers. Complexity and performance impact of using a pointer are H> > noted by Drew in D23347. H> H> Let not the network epoch become the new Giant of EPOCH's. There might H> be realtime constraints for EPOCH's aswell. Epoch isn't a lock, so it can't become the new Giant. -- Gleb Smirnoff
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20200127202059.GV1268>