Date: Fri, 14 Jan 2011 15:54:37 +1100 (EST) From: Bruce Evans <brde@optusnet.com.au> To: Chuck Swiger <cswiger@mac.com> Cc: Charles Owens <cowens@greatbaysoftware.com>, freebsd-net <freebsd-net@freebsd.org> Subject: Re: igb watchdog timeouts Message-ID: <20110114154326.E27511@besplex.bde.org> In-Reply-To: <0B45B324-A819-4230-BBE3-F8468F2DA88F@mac.com> References: <20100729215649.GB2615@icir.org> <20110103210209.GA13091@icir.org> <4D2E66C4.5090607@greatbaysoftware.com> <AANLkTinxDryptLu%2B7NRnLPLE7716BHw=CZ==jYOb_Q%2BY@mail.gmail.com> <4D2F20BB.5080204@greatbaysoftware.com> <AANLkTimK8VEQLd-m-zPsw8-%2BoBi-oJ5pc5eScmFXmujy@mail.gmail.com> <4D2F71BE.2080801@greatbaysoftware.com> <0B45B324-A819-4230-BBE3-F8468F2DA88F@mac.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 13 Jan 2011, Chuck Swiger wrote: > On Jan 13, 2011, at 1:42 PM, Charles Owens wrote: >> This is very good news overall, in that we can certainly disable polling for igb. This begs the question, though, as to whether polling is recommended these days at all for em/igb NICs... or even in general. From other conversations we've seen there seems to be some general debate about this. In testing we've done in the past (circa 7.0) there certainly seemed to be benefit to using this feature. What are your thoughts about this? > > To quote an earlier post: > > "Polling mode operation generally performs better when using older 100Mbs ethernet NICs which do not support interrupt mitigation and various capabilities like TSO4; gigabit ethernet NICs are smarter hardware and can generally outperform polling mode." I think "older 100Mbs" means "low-end 100Mbps". Mega-bit-seconds are strange units, and 100Mbps NICs with enough buffers don't benefit much from polling mode. They even avoid dropping the *nix newline character. > Polling is well-suited for dedicated routers, firewalls, and other boxes which have a constant flow of traffic and for which you are looking for well-bounded latency. End-user machines, servers, and the like which have bursty traffic tend to do better using normal NIC operation, especially if you have decent gigabit NICs which support interrupt mitigation and have larger buffers than the old 100Mbs NICs had. I never saw any problem with interrupt mode fxp 100 Mbps NICs. They have enough buffers (128 for each of tx and rx IIRC). The only thing polling mode gave for them was lower latency, but this cost enabling polling in the idle loop, which wastes 100% of at least 1 CPU and some power. Without polling in idle, polling gives very high latency (even worse than low-quality interrupt moderation does). Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20110114154326.E27511>