[8755] in Commercialization & Privatization of the Internet
Re: NAP philosophy & IP technology
daemon@ATHENA.MIT.EDU (Peter S. Ford)
Fri Dec 3 10:47:00 1993
From: "Peter S. Ford" <peter@goshawk.lanl.gov>
To: "Roger Fajman" <RAF@cu.nih.gov>
Cc: fair@apple.com, com-priv@psi.com
In-Reply-To: Your message of Thu, 02 Dec 93 21:34:50 -0500.
Date: Fri, 03 Dec 93 08:45:49 MST
> But in the short term the trend with CIDR is to tie the customer closer
> to the provider. If you have to number from the provider's address
> space and if, as some were proposing at the last IETF, exceptions are
> not allowed, how would one be able to connect to two providers at all?
>
Roger,
CIDR does not *require* that you be tied to a particular provider.
What the CIDR plan calls for is that the default case be for an "end
site" to take its address from its "immediate provider". The CIDR RFCs
document the exception cases and how they can be handled. The IETF's
CIDR deployment working group does not propose that exceptions not be
allowed.
What was discussed is that people who change providers will
need to change their addresses over time to reduce the total amount of
global routing information as is possible with the CIDR routing
hierarchy. Much of the discussion in the working group was on how
providers can help their customers to make these changes in a manner
which does not require flag days and disruption of end site services.
DHCP will help considerably in this process. Changing addresses sounds
onerous to many. Fortunately, most people do not use addresses
directly because most use of the Internet is based on names. I don't
have a clue as to the IP addresses used to deliver mail to
com-priv@psi.com, or what the address used for ``ftp ftp.uu.net" will
be. The target of the current work in the IETF is to make the
management of the name to address binding even more transparent thereby
making renumbering a simple, non-intrusive matter for Internet users.
A dual end site is supported with CIDR, at the expense of additional
routing information floating around in the Internet. The site can take
an address from one of its immediate providers and have its 2nd
provider also propogate that address. Presumably the customer for dual
homing will be paying the marginal cost of every provider carrying this
extra information. There is actually a layer in the current CIDR
hierarchy which can capture this extra routing entropy, but to gain
maximal benefit of the dual homing this information will be carried by
most major transits on a continent. CIDR addressing plans, coupled
with sane topological interconnection (NAPs, FIXs, GIXs, CIXs, etc.)
will make dual homing more tractable in terms of total entropy of the
routing system. The entropy of extra routing information propogation
can be traded off amongst providers against optimality of routing.
I think the best way to get the reliability you are searching for is to
get your immediate provider to build redundant infrastructure and make it
available to you as a service. IP supports this sort of redundancy/reliability
within a provider very well since this was one of the main drivers in
the design of IP, and I suspect it is the most cost effective way for a
site to get "non-stop" operation. Being compliant to the default case of
the CIDR plan is an additional benefit of this mode of building higher
reliability.
cheers, peter
