Re: what is the fundamental difference between OSPF and IS-IS?

Radia Perlman - Boston Center for Networking <Radia.Perlman@SUN.COM> Sun, 01 September 2002 02:12 UTC

Received: from cherry.ease.lsoft.com (cherry.ease.lsoft.com [209.119.0.109]) by ietf.org (8.9.1a/8.9.1a) with ESMTP id WAA17119 for <ospf-archive@LISTS.IETF.ORG>; Sat, 31 Aug 2002 22:12:55 -0400 (EDT)
Received: from walnut (209.119.0.61) by cherry.ease.lsoft.com (LSMTP for Digital Unix v1.1b) with SMTP id <12.006F9D72@cherry.ease.lsoft.com>; 1 Sep 2002 22:14:19 +2000
Received: from DISCUSS.MICROSOFT.COM by DISCUSS.MICROSOFT.COM (LISTSERV-TCP/IP release 1.8e) with spool id 126253 for OSPF@DISCUSS.MICROSOFT.COM; Sat, 31 Aug 2002 22:14:18 -0400
Received: from 192.18.42.14 by WALNUT.EASE.LSOFT.COM (SMTPL release 1.0f) with TCP; Sat, 31 Aug 2002 22:04:18 -0400
Received: from sydney.East.Sun.COM ([129.148.9.16]) by nwkea-mail-2.sun.com (8.9.3+Sun/8.9.3) with ESMTP id TAA10594 for <ospf@discuss.microsoft.com>; Sat, 31 Aug 2002 19:04:17 -0700 (PDT)
Received: from labeast (labeast [129.148.75.22]) by sydney.East.Sun.COM (8.11.6+Sun/8.11.6/ENSMAIL,v2.2) with SMTP id g8124GL14223 for <ospf@discuss.microsoft.com>; Sat, 31 Aug 2002 22:04:16 -0400 (EDT)
MIME-Version: 1.0
Content-Type: TEXT/plain; charset="us-ascii"
Content-MD5: mO/dF8mLUGRz/jnAeDHmGQ==
X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.4.6_06 SunOS 5.8 sun4u sparc
Message-ID: <200209010204.g8124GL14223@sydney.East.Sun.COM>
Date: Sat, 31 Aug 2002 22:04:15 -0400
Reply-To: Mailing List <OSPF@DISCUSS.MICROSOFT.COM>
Sender: Mailing List <OSPF@DISCUSS.MICROSOFT.COM>
From: Radia Perlman - Boston Center for Networking <Radia.Perlman@SUN.COM>
Subject: Re: what is the fundamental difference between OSPF and IS-IS?
To: OSPF@DISCUSS.MICROSOFT.COM
Precedence: list

        From: "Liu B." <binl@EEE-FS7.BHAM.AC.UK>
>>what is the fundamental difference (or
>>improvement?) between OSPF and IS-IS?

>>Bin Liu

Several people mentioned my paper from 1991, and since it will
probably be hard to get (I don't have a copy, and haven't seen
it in years), I thought I'd mention it's probably not worth
worrying about. I'd be surprised if it would
be too helpful today.  Both protocols have changed since then, so I'd think
the paper would be mostly interesting, if at all, for historical reasons.

So back to "what is the difference between OSPF and IS-IS?"...

This sort of question should be asked more often, especially where there
are "foo" vs "bar" debates. As I said in my "Miss Manners meets the IETF"
talk, unfortunately such questions often lead to err..nontechnical
responses. I'm glad that all the responses to this note have been
thoughtful and technical.

And since people will periodically wonder about the real technical
differences between foo and bar, it is useful to write the stuff down.

This particular question (differences between IS-IS and OSPF)
is a topic that is asked about sufficiently often that it would
probably be worth writing an updated paper, based on the most recent
versions of the two protocols. My book (Interconnections) discusses
some more differences, and Dave Katz's presentation is also
helpful. But I think it would be nice to have some set of people
carefully compare both specs, and
calmly write down the differences and the pros and cons of the differences.
And given that it seems like both protocols will persist, if there
are cases where one scheme is significantly
better, it ought to be folded into the
other protocol if possible.

As people on the list have pointed out, there is a lot of similarity between
OSPF and IS-IS. To understand why there are two protocols, it helps
to know some of the history.

Basically, the first link state protocol was for the ARPANET. It
introduced the idea of link state protocols, with the major
ideas being:
   . flood link state information to everyone
   . use Dijkstra's algorithm on the link state database to
     compute paths
   . an algorithm for incrementally updating the Dijkstra tree
     when there are just a few link changes.

The improvements (that I can think of off the top of my head)
introduced by IS-IS were:
  a) making link state distribution robust (the ARPANET had an amusing
     failure mode)
  b) efficient incorporation of LANs, including the concept of Designated
     Routers, LANs as pseudonodes, and link state syncronization over
     a LAN using CSNPs. (ARPANET was just point-to-point links)
  c) exchange of parameter information (such as how long to wait before
     declaring neighbor down), so that parameters can be set independently
     and differently, and still interwork

IS-IS was originally designed for CLNP, which had two forms of routing:
       . exact match of bottom 6 bytes, or
       . shortest prefix of top part
That's where the "two levels" came about, though the "area routing"
stuff could in theory be multilevel. IP only has one type of routing, which
is like the shortest prefix, area routing in CLNP.
IS-IS would have looked a little different if it had originally
been designed for IP.

OSPF was designed specifically for IP and was loosely based
on IS-IS. At the time OSPF was beginning
to be designed, it didn't occur to anyone that IS-IS could be easily
adapted to route
IP. Ross Callon noticed that once a routing algorithm existed, it was
a mere detail to add reachability information for a different data
packet format, and he wrote RFC 1195. (Interestingly, the concept
of integrated routing wasn't a new idea. I realized years later, when
looking at RIP's packet format, that RIP, deployed years before
integrated IS-IS was conceived, was designed for routing multiple
address families).

Perhaps if the idea of using IS-IS for IP was thought of before
work on OSPF had started, there wouldn't be two protocols. But once
a group starts on something it's hard to stop. So there wound up
being two protocols.

NLSP was a version of IS-IS for IPX, and introduced some improvements,
my favorite being DR election (see below).

Differences I can remember (off the top of my head).

a) Designated Router election: it's "deterministic" in IS-IS, which means,
   given the same set of routers, the same router will be elected. It's
   "sticky" in OSPF, meaning that once you get to be DR, you stay DR.
   This makes things more stable...if the highest priority router is
   flaky in IS-IS, every time it goes up it takes over, only to crash again.
   But I was told when designing IS-IS that determinism was important, which
   requires the behavior in IS-IS.

   For DR election, if I had to choose between OSPF's way and IS-IS's
   way, I'd choose OSPF's way, but NLSP (IS-IS for IPX plus
   some improvements) had a method
   that gave the best of both worlds.
   It has nodes change their priority after becoming DR,
   which allows you through astute priority settings, to
   choose deterministic or sticky behavior, or anything
   in between.

b) LSP distribution on a LAN: IS-IS does it with CSNP's. OSPF with
   explicit acks. I believe both ways are just fine.

c) Originally IS-IS passed no upper layer information into areas, and
   you just exited via the nearest level 2 router, whereas OSPF always
   fed all information into the area so you could choose the optimal
   exit point. Both have now been modified so you can choose any
   point on the tradeoff between extra routing info and optimal routing.

d) parameter synchronization: IS-IS allows neighbors to have different
   values for things like Hello Timer, and still interwork.


So anyway, there are a bunch of little nerdy differences, some of
which might matter and some of which are just different because different
groups did them. For the ones that matter, mostly the protocols have
evolved to take advantages of features in the other protocol.

I assume if there was an updated IS-IS vs OSPF document, someone would
have mentioned it in response to Bin Liu's post. So assuming there
isn't such, if someone wanted to try to do it, I (and I'm sure lots
of other people) would be happy to help.

Radia