North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Bill Manning considered fatal

  • From: Nicolas Williams
  • Date: Fri Jun 30 14:48:30 1995

Tony Li previously wrote:

>   two questions (and a task) for the list(s)

>	   2. 	What will it take to get router vendors to modify default
>		   setup to select a classless IGP instead of (RIP,IGRP)?

>The power-off of the last IPv4 network.

IPv4 will here for at least a while longer... :)

>You should understand that there are many reasons for this:

>- All of the classful protocols today are incredibly hard to work with
>for a beginning user.  OSPF has about 3 orders of magnitude more knobs
>than it needs.  EIGRP (which we tried REALLY hard to keep simple) is
>about two orders of magnitude too hard.  Both are resource intensive
>and require careful hand-holding.  RIP2 is less piggish, but still is
>two orders of magnitude too messy.

OSPF is unnecessarily hard to configure (even BGP4 is easier!). EIGRP is
very nice, very simple; I have no complaints agaisnt it, other than the
fact that it is proprietary. IS-IS is also nice to config, but far less
flexible (no area #s).

>- Please recall that the technical savvy of the beginner customer base
>has dropped two orders of magnitude in just the last three years.

Their problem.

>- You don't need classful in a stub network for CIDR purposes.  Yes,
>it helps for address efficiency IF the user can figure out VLSM.

No, you don't. Stub nets are what most small ISPs and what practically
all small private Internet customers will be for some time. Those ISPs
that can't grow very much past the stage of providing local dialup and
some leased lines will be the ones to go away the fastest. I don't feel
sorry for them.

>- The defaults need to be for the most common case: rank beginner in a stub.

And their ISPs should route these customers via _static_ routing only.
Makes life a lot simpler; besides, if an ISP is going to run a dynamic
routing protocol with a singly-conneted client, then they'll have to
filter the routing info they get from them (and maybe what they send),
at which point they might as well be doing static routing.

>- There are many networks which are NOT the Internet.  CIDR only
>applies to the Internet.

No, why? CIDR is _clean_, it is _nice_. I'd use it in a private IP net
that was not meant to connect to the Internet if it was large enough and
available from all vendors.

>I sympathize, but this is a case where the intelligent actually must
>lose to the tyranny of the majority.

We'll see who are the ones to go under over the next few years.