From owner-freebsd-net Fri May 28 4:15:26 1999 Delivered-To: freebsd-net@freebsd.org Received: from bayer2.bayer-ag.de (bayer2.bayer-ag.de [194.120.191.2]) by hub.freebsd.org (Postfix) with SMTP id 721B515A0F for ; Fri, 28 May 1999 04:15:21 -0700 (PDT) (envelope-from ANDREAS.KLEMM.AK@bayer-ag.de) Received: from BYE473.BAYER-AG.DE (bye473.bayer-ag.com) by bayer2.bayer-ag.de with SMTP id NAA28510 (SMTP Gateway 4.2 for ); Fri, 28 May 1999 13:13:12 +0200 Received: by BYE473.BAYER-AG.DE (Soft-Switch LMS 3.2) with snapi via MT0044 id 0006800011799081; Fri, 28 May 1999 13:15:05 +0200 From: ANDREAS.KLEMM.AK@bayer-ag.de To: " - *amb@gxn.net" Cc: " - *andreas@klemm.gtn.com" , " - *freebsd-net@freebsd.org" , " - *zebra@zebra.org" , " - *kunihiro@zebra.org" Subject: Antwort: Re: [zebra 553] OSPF eequal-cost paths, which algor Message-Id: <0006800011799081000002L012*@MHS> Date: Fri, 28 May 1999 13:15:05 +0200 Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable Content-Disposition: inline Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org =0AHi Alex, first of all: sorry, if you get some dirty characters, Lotus Notes cann= ot be configured, to be pretty 7Bit ASCII compliant and readable :-/ Thanks for your nice answer. Could you perhaps point out, what actually= happens. Do I have 2 x 2 x 2 Mbit =3D 8 Mbit throughput for IP traffic ? Or does the router decide to choose one route over one router pair ? If the latter is the case, what criterium the router chooses, which pat= h to use ? The roundrobin functionaliy you mention I know from having _one_ router= and two interfaces then turning on the fast cache, so doing processor switching= . But here the question is, _how_ do the packets or sessions flow when having to r= outer _pairs_ ?! Does OSPF really do something like roundrobin packet for packet (1st pa= cket uses route over R1-R2, 2nd packet uses the other equal-cost path). Another question is, when routing IPX, Appletalk and also turning on br= idging, it might happen, that one path is more loaded then the other (spanning = tree). Then the paths have the same OSPF cost, but are differently loaded, wha= t happens then ? Is one path preferred and after what algorithm ? Thanks Andreas /// amb@gxn.net on 28.05.99 11:00:07 An: kunihiro@zebra.org @ INTERNET, zebra@zebra.org @ INTERNET, freebsd-net@freebsd.org @ INTERNET, Andreas Klemm@BAYERNOTES, andreas@klemm.gtn.com @ INTERNET Kopie: Thema: Re: [zebra 553] OSPF eequal-cost paths, which algorithm, how Andreas, > I'm looking for the algorithm how OSPF does a routing decision and > what kind of load balancing is been done, between the two 4 MBit > leased lines, when OSPF has equal-cost paths like this. > Cisco IOS 11.2 or 11.3. Roughly: When Cisco has (by any meothod) two equal cost routes installed in the RIB (i.e. routing table), both get installed in the FIB (i.e. forwarding table). When a packet comes to be forwarded, generally some form of route cache= is used, which is normally just a hash of the destination IP being routed to (+/- hardware accellerated routing on higher end boxes, CEF etc. etc.). However, occasionally (normally the first packet *to* a given host) this caching algorithm will miss, and a lookup will be made in the FIB. Where there is more than one entry, these algorithm round-robbins between each of them (well actually I think it may chose randomly between them, which is the same thing in practice). So if you enter "no ip route-cache" on the interface, you will find packets (whereever they are to) round robin between the equal cost routes, thus load sharing. But when you have route-caching switched on (normally), you will find all the traffic to a given destination goes the same way, but providing= you have traffic to a large number of destinations, in practice you get good load sharing (don't try putting 2 newsfeeds down 2 2Mb lines this way though). The situation is more complex when you have protocols like EIGRP which will on later IOS versions do *non* equal load sharing. I believe what they do then is tag the relevant load sharing weight onto the RIB entry which propagates to the FIB and modifies the round robin algorith= m. However I have not tested this in practice. -- Alex Bligh GX Networks (formerly Xara Networks) = To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message