ISPAC-1 - A Scalable Proposal?

[Larry Vaden]

First, a distinction:  This proposal uses the terms "ISP" and "NSP", the
rough distinction being that NSPs have national backbones and ISPs do not.

ISPAC-1 would be an ISP Address Coalition.   George Herbert
<gherbert at crl.com> has done some work in this area, as have Tony Li and
Yakov Rekhter.  For George's contribution, see the NANOG and COM-PRIV
archives;  for the work of Tony and Yakov, see:

http://www.globecom.net/(nobg,sv)/ietf/draft/draft-li-ispac00.shtml and
ftp://ftpeng.cisco.com/ftp/yakov/spa-vs-pac.ppt.

It would differ somewhat from the ISPAC model in the two above URLs;  the
primary difference is influenced by the "bottom up" approach, which is to
say it is more oriented towards ISPs (which need /20s thru /16s, e.g.) and
firms like digitalink.com, which publishes the Washington Post online
(which might need a /24 or /25) which need multihomed CIDR blocks. 

The objectives are to provide ISPAC-1 members with provider independent
multihomed CIDR blocks, as opposed to provider dependent (AKA PA) space.
Further, the objective attempts to pay particular attention to requirements
of good stewardship of net resources.

ISPAC-1 would apply for a suitably sized CIDR block from NSI/InterNIC or
directly from the IANA.

ISPAC-1 would issue a RFP for a "group buy" and then, using the normal
commercial negotiations process,  select 3-5 NSPs to carry their traffic.
Because ISPAC-1 members would be allocated PI space out of the same address
block, aggregation would be possible and likely.

Selected NSPs would advertise a single prefix (if initial sizing were
correct) for ISPAC-1 members.

Due mainly to the concern of Tony Li with whether this is scalable and any
NSP would want the business because the NSP's routers would have to contain
explicit routes to each ISPAC member, a very tentative and preliminary
proposal for prefixes would be:

xxxxxxxx.yyy.zzzzzzzz, with the length (in bits) of xxxxxxxx, yyy and
zzzzzzzz to be determined.  xxxxxxxx would be the most significant part of
the prefix.  yyy would be geographical (following George's thoughts) in
nature, but might be of length 0 if Tony's concerns are not of concern to
the selected NSPs.  If non-zero in length, yyy would serve to decrease the
number of specific routes in any "regional" router in the contracting NSP,
IMHO.  zzzzzzzz would be ISPAC member specific.

If yyy is of any use in reducing Tony's concerns, the credit goes to George
Herbert, gherbert at crl.com, for his discussions in NANOG and elsewhere
regarding address allocation policy in the greater San Franciso area.

Sincerely,

Larry Vaden
<standard disclaimers>

original distribution to: Kearney Connolly <Kearney at UU.NET>,
Shane_Hampton at corp.acsi.net, Chad Hutchings <chutch at digex.net>, "Robert L.
Shearing" <robert at priori.net>, with cc:'s to George Herbert
<gherbert at crl.com>, Tony Li <tli at juniper.net>, Yakov Rekhter
<yakov at cisco.com>, nair-founders at texoma.net, Stan Barber <sob at academ.com>