Date: Sat, 4 Aug 2001 17:27:46 -0700 (PDT) From: Matt Dillon <dillon@earth.backplane.com> To: Leo Bicknell <bicknell@ufp.org> Cc: Bernd Walter <ticso@mail.cicely.de>, sthaug@nethelp.no, oppermann@telehouse.ch, freebsd-hackers@FreeBSD.ORG, freebsd-net@FreeBSD.ORG Subject: Re: 303,000 routes in kernel Message-ID: <200108050027.f750RkG77073@earth.backplane.com> References: <20010804215529.C7176@cicely20.cicely.de> <32301.996956619@verdi.nethelp.no> <20010805002233.A7991@cicely20.cicely.de> <20010804184045.A87444@ussenterprise.ufp.org>
next in thread | previous in thread | raw e-mail | index | archive | help
:We're rapidly getting off topic here, but for the record...
:
:Route caching has not delivered acceptable performance for "core"
:routers for some time. As the internet got larger and used more,
:the percentage of the total routing table that was in use (and
:hence cached) grew larger and larger, exhausing the smaller, faster
:cache memories.
:
:All of the current designs used in the core, and many of the edge
:designs as well keep the "full table" (distilled to the minimum
:amount of information to forward a packet) available to the hardware
:forwarding engine. This includes Cisco's GSR line, and Junipers
:M-series routers. While working differently, Cisco's 7200's and
:3600's also do the "full table thing".
I agree re: the route cache concept. It's dead.
As far as I know routers on the edge tend to have full route
tables, but I was under the impression that modern core designs - that
is, the routers in the middle of a network, tended to use more of a
layer-2 tagged switching model. The incoming edge routers with the full
route tables tag the packet, then intermediate routers switch the packet
using the tag (and ignore the IP address), and then when the packet
gets to the other edge of the network it gets run through a normal routing
table again. This allows routers in the middle of the network to use
simple array lookups / layer-2 switched designs which are much faster
then full-route-table designs.
-Matt
:Looking at next generation designs, all vendors agree the only
:designs that will work in the core are designs that will work.
:So, unless the packet needs local processing (eg ICMP pings)
:a packet is a packet is a packet to today's routers.
:
:--
:Leo Bicknell - bicknell@ufp.org
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-net" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200108050027.f750RkG77073>