[1233] in Commercialization & Privatization of the Internet
Re: technical details
daemon@ATHENA.MIT.EDU (Vince Fuller)
Tue Aug 27 21:28:12 1991
Date: Tue, 27 Aug 91 18:27:14 PDT
From: Vince Fuller <vaf@Valinor.Stanford.EDU>
To: asp@uunet.uu.net (Andrew Partan)
Cc: wls@psi.com, ittai@shemesh.ans.net, com-priv@uu.psi.com,
In-Reply-To: Your message of Tue, 27 Aug 91 21:04:34 EDT
Excerpting from:
From: asp@uunet.uu.net (Andrew Partan)
Subject: Re: technical details
> From: Vince Fuller <vaf@Valinor.Stanford.EDU>
> Subject: Re: technical details
> To: wls@psi.com (William Schrader)
> Date: Tue, 27 Aug 91 14:58:14 PDT
>
> Date: Tue, 27 Aug 91 16:32:25 -0400
> From: wls@psi.com (William Schrader)
>
> 1. The CIX and the CIX Association are "scaleable", assuming Regionals
>
> This is an interesting statement. From a legal and political point of view,
> this may be true. From the technical standpoint, however, the existance of
> both CIX and ANS serves to exacerbate the scaling problems associated with
> IP routing - with multiple external attachments, a mid-level MAY need to
> carry the global routing database; with one attachement, it need only
> carry its own routes plus a default route to the external location.
> This is not true - I do not need a full global routing database. I
> have my own routes plus all of the CIX routes plus all of routes of
> other directly connected nets (like EUnet and NEARnet) and a default
> route of the NSFNET. It works just fine.
Note the use of the word "may" not "must". In any case, if we assume that all
mid-levels join the CIX and that the the majority of networks want to use the
unrestricted CIX, then the list of "my own routes plus all of the CIX routes
plus all of toutes of other directly connected nets" starts to approximate the
the global routing table. It will be a LOT of nets that may, in some form, need
to be carried by each participating midlevel. The reason that we can get away
with pushing only default around is that the NSFNet backbone serves as the
general-purpose "core" for the research community. An extensively connected
CIX with many networks used for both research and commercial traffic breaks
this model and requires more explicit routing information to be carried.
--Vince
