Re: IS-IS reference

[Vadim Antonov <avg () kotovnik com>]

Alex P. Rudnev <alex () Relcom EU net> wrote:

> > The real answer - do static routing whereever you have only a single
> > path for packets to go thru.  To eliminate mistakes, generate
> > configuration automatically from master maps kept at network engineering
> > computers.

> This case OSPF do the same thing - generate routes when it started up;
> and no doubt it do less mistales then the human person.

No, it doesn't do the same thing - the difference is about the same
as the difference between compile-time and run-time type checking.

OSPF (and any other dynamic routing protocol) introduces and removes
routes every time a link or device goes up or down - or perceived
to go up and down.  A typical scenario can be like: connecting a PC
with a broken NIC card to an Ethernet segment can easily cause massive
packet loss, causing rapid route flap which basically infects the
entire backbone.  If that backbone carries even a fraction of full
exterior routing table, this would cause loss of coherence between
routing tables and persistent routing loops.

Additionally _no_ exising IGP has anything resembling protection
from malfunctioning routing software _or_ malicious or negligient
operators of host-based rotuing software.  I had to track down people
who enable gated on their linux boxes just for the fun of it, and
screw the entire network up in the process, more times than i care
to count.

In other words, dynamic routing is very brittle, and requires quite
a lot of care to make sure it works right, and that a single-point
failure won't affect the significant portion of the network.  If
you would claim that you ever did an analysis like that for any real
network i'd have to ask you to enlighten all of us about the obviously
break-through novel network design technique you're using.

> No, the comparation between OSPF and STATIC looks like the comparation
> between the old (from 1950 year) and modern (Mersedess-600) cars - the
> first is very simple implemented and difficult to drive; the second is
> very complex implemented but very simple to drive (but if you are to be
> starving on the unhabitant ireland with the good roads, you'll choose the
> first car; but it seemd for me you just choose something more complex in
> the real life).

Did you notice that it takes a highly trained specialist with appropriate
(and rather expensive) equipment to diagnose and fix a problem in a Mercedes?
A hammer and few expletitives usually suffice for a Packard.

That's the real difference.  People who understand routing protocols
and how they interact with level-2 transport are still extremely rare
and rather expensive.  Even the major ISPs here in US have serious staffing
problems.  An average corporate MIS department is best characterized as
clueless (what other explanation is here for the Microsoft dominance? :)

In other words, you're advising kids who don't yet know how to hold a
hammer to start using chain saw.  In a situation like that i would
expect a lot of cut-off bodily parts.

> PS. And if someone use STATIC widely, a few years ago some other person
> should be sitting for a few days and flame the first one digging through
> a heaps of the static routes /it's real example from my life/.

At least he'll be able to understand what's broke when the network goes down.

BTW, dealing with heaps of statics is very simple: do a numbering plan first,
so the routes are aggregable.  That is helpful for dynamic routing, too.

--vadim