[809] in Commercialization & Privatization of the Internet
Re: BGP masks
daemon@ATHENA.MIT.EDU (Milo S. Medin" (NASA ARC NSI Proje)
Mon Jun 3 21:53:27 1991
To: brian@napa.Telebit.COM
Cc: yakov@watson.ibm.com, ietf@ISI.EDU, com-priv@psi.com, hwb@sdsc.edu,
In-Reply-To: Your message of "Mon, 03 Jun 91 16:35:46 PDT."
Date: Mon, 03 Jun 91 18:46:07 -0700
From: "Milo S. Medin" (NASA ARC NSI Project Office) <medin@nsipo.nasa.gov>
Also, while doing many novel things with assigning masks does require
new addressing mannerisms, doing some good things doesn't. Consider
a regional like BARRNET. Most of it's subscribers have no other network
connectivity (probably true for most regionals). One could take a
Class A net (36? :-)) and give out appropriately sized pieces to all
the BARRNET member sites with no other connectivity. OSPF can handle
this easily. If BGP supported netmasks, then BARRNET could get rid of
most of the routes it sends to the rest of the Internet, and send in only
36 and the nets for multiply connected sites, slashing the number of
routes others people have to pass around because of them.
This requires no magic at all. In fact, it probably doesn't require any
BGP interpretation of variable length masks, as the IGP can provide the
glue to support all this inside the BARRNET core OSPF system. You just
need a fairly stupid way of passing subnet info around. And speaking of
fairly stupid ways to pass subnet info around, RIP will work fine here
as an Inter-AS protocol. Having BGP support what RIP does can't be
*THAT* hard... :-)
Note that this scenario is very much like hierarchical based routing
in OSI. You assign addresses based on your primary carrier. If the
Inter-AS protocol could support variable length masks, thus eliminating
the restrictions on subnets being connected, you could even deal with
multiply homed sites using subnets, and support "carrier" changes without
any renumbering being necessary.
I'm not saying that this is the best thing to do in all cases,
but it certainly could cut down the number of nets used, AND the total
number of nets the regionals and the backbones have to pass around. And
it's very simple to do if you have OSPF (or dual IS-IS) in your backbones,
and RIP between your AS's. Wouldn't it be nice if BGP could support this
as well.
I understand the desire to get BGP out there and fielded. But if mask
support can't be added right now, it ought to be the first thing added
for the next version.
Thanks,
Milo
