Date: Tue, 27 Jan 2015 16:44:30 -0600 From: Jim Thompson <jim@netgate.com> To: =?utf-8?Q?Olivier_Cochard-Labb=C3=A9?= <olivier@cochard.me> Cc: "freebsd-net@freebsd.org" <freebsd-net@freebsd.org>, Michael Sierchio <kudzu@tenebras.com>, =?utf-8?Q?Antoine_Beaupr=C3=A9?= <anarcat@koumbit.org> Subject: Re: is polling still a thing? Message-ID: <B5108C20-6E48-44EA-B803-9CBC804F1755@netgate.com> In-Reply-To: <CA%2Bq%2BTcoanjpNX91vi5EX4-oprm0PiVUYi-EOPRfi6zzXXmwsow@mail.gmail.com> References: <871tmgceup.fsf@marcos.anarc.at> <A32D80F3-9D34-4136-A870-B28582F6EAA0@netgate.com> <87vbjsaxxy.fsf@marcos.anarc.at> <CA%2Bq%2BTcrkp3U%2Bb5816es3jsvhNfxpW_v%2BRQOa8a9BbLwNTm3RuQ@mail.gmail.com> <CAHu1Y71XDO1fuCJM_Cw1wm4jt%2BigKisaTWRb4gBX3nsaepwMsQ@mail.gmail.com> <CA%2Bq%2BTcoanjpNX91vi5EX4-oprm0PiVUYi-EOPRfi6zzXXmwsow@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
> On Jan 27, 2015, at 2:28 PM, Olivier Cochard-Labb=E9 = <olivier@cochard.me> wrote: >=20 > On Tue, Jan 27, 2015 at 9:15 PM, Michael Sierchio <kudzu@tenebras.com = <mailto:kudzu@tenebras.com>> wrote: >=20 >=20 > On small, embedded computers running ipfw w/kernel nat and device = polling enabled (on em ether adapters), I observed the *reported* system = load grow very high. When disabling polling on the interfaces, it went = back to something normal. >=20 > My impression is that the consensus among the core developers = concerned with networking is that device polling is an ancient hack and = is deprecated. In the case of a DDoS attack, there may be many other = things to try - at the infrastructure level - traffic diversion = techniques like BGP flowspec, use anycast, etc. On the individual = server level, use stateful rules with GRED enabled, dropping most new = tcp or udp traffic based on load. >=20 >=20 >=20 > If I remember well, Luigi had a surprise regarding the advantage of = using polling inside a VM: > https://lists.freebsd.org/pipermail/freebsd-net/2013-May/035626.html = <https://lists.freebsd.org/pipermail/freebsd-net/2013-May/035626.html> >=20 > But on real hardware, since the introduction of interrupt moderation = on NIC, polling is not more useful. The DPDK guys disagree.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?B5108C20-6E48-44EA-B803-9CBC804F1755>