CURRENT MEETING REPORT
Reported by Bill Manning,
Information Sciences Institute and Roger Fajman, National Institutes
Minutes of the Procedures
for Internet/Enterprise Renumbering Working Group (pier)
Solicitation Number Two
General comments were made as to how Solicitation Number Two should fit into the PIER effort. The focus is to send it specifically to vendors. They expect very specific information.
The working group was encouraged
to discuss this issue. Ralph Droms wants to make it a parallel
and complimentary effort to the first solicitation. It is possible
that individual copies will be sent to the vendors that are identified.
Howard Berkowitz stressed
avoiding duplicating efforts. Paul Ferguson wanted to define a
framework for the final documents. Bill Manning discussed using
the final output to help people. Brian Carpenter talked about
the IPng efforts to coordinate responses to their solicitations.
Elliot Lear stressed
his desire to get actual hard specific information, not warm fuzzy
B. pointed out that the people we are soliciting are more operations
people, while the IPng efforts were more focused on strategic
planners. Will Leland pointed out we should rename it "Vendor
Drainage of the Swamp in
192/8 (presentation by Suzanne Woolf, ISI)
Some people are encouraged,
by the numbers, others are somewhat surprised. People discussed
the level of educational efforts requested back. Bill asked what
we do with the 189 people who responded about renumbering. The
5980 who responded are not the entire list because of a variety
of reasons. Peter
Lothberg pointed out that we shouldn't care about non-routable
takes up about 20% of the current routing tables (http://www.isi.edu/div7/pier/whose-routes).
Bill Simpson asked about the 23 who aren't using their networks,
but wouldn't give it back. He also
suggested a BCP about
keeping contact information up-to-date. Steve Bellovin pointed
out that people are scared about giving addresses back.
Bill pointed out that one
area PIER needs to concentrate on is education. There is a feeling
that 192/8 are valuable since it is such a swamp that it will
always be routable.
Current Experiences (Elliot Lear, SGI)
Going from 1 B and numerous
C's to one/16, emphasis was to stablize their routing.
SGI is somewhat like
an ISP since everything is centralized. Most sites didn't want
or have any incentive to renumber. Most sites could not do variable
length subnets (ie. CIDR). The subnet masks at those sites had
to be fixed. It
must be insured that all routers could run OSPF and get rid of
unnumbered WAN interfaces (using 255.255.255.252 masks). In addition,
all routers must be brought up to the same software revision level.
Scripts have been developed that automatically renumber SGI's
machine. SGI has a policy to use FQDN's at all sites. Paul Vixie's
bind implementation serves as a commendable effort. Lear pointed
out that NIS has been a real problem for their large site. NIS
Slave Servers are tricky to get the order right, in terms of getting
new maps to them from the master, for SGI doesn't have a good
DHCP until next release. The next kernel will have support for
classless protocols. There was more discussion about getting OSPF
areas to match across unnumbered interfaces. It was noted that
there are also problems with secondary interfaces. Five different
sites of a variety of sizes have been piloted, most under 50 hosts.
They will soon be doing some 2-3 hundred sites soon. Seven million
dollars has been spent so far, most of the expenses going towards
personnel costs. This is estimated and includes more than renumbering.
It was stressed that instructions are very important.
Ed Kern, Digex
Heavily-used options on CISCO's
to help a transition plan. Many people were moved from /24's to
/27's from their CIDR block. DNS moves caused the most problems.
In addition, there
were performance concerns about translators.
Yakov has data which
he will provide via email. He will also provide informal data.
Yakov does not plan on writing a formal document.
8000 node Bridged Network (Brian Carpenter, CERN)
They are recabling every building
from coax to cat-5 UTP. Everyone must renumber since they are
moving off of the old bridged /16. They are renumbering people
in small flag days. People get 4 weeks notice, then two weeks
later another paper mail telling them again, and then an email
the day before. A team of 5 people is working for 18 months, as
well as routers running EIGRP.
Steve Bellovin, AT&T
They are just starting because
of the breakup.Their addresses are mostly geographic so moves
have a big impact. The ability to assign multiple IP addresses
is very important. Solaris, BSDI, SGI (IRIX 5.3 with patches),
and a module for SunOS NIS hates renumbering.
Paul Traina, CISCO
They moved from several dozen
C's to several B's. This was done in coordination with a physical
move. Sweep teams came by later to make sure people renumbered
okay. Each day about 200 were completed.
They are doing about
6000 hosts at 120-200 sites. Paul Ferguson is the coordinator
of this section.
IP Addresses in Applications
Phil described plans to develop
a cookbook for specific IP applications. He cited known problems
with PROM-coded IP addresses in GE Medical patient scanners; the
patient is made transparent by the application, but the application
is not transparent to the Internet.
HP Openview and Cabletron
Spectrum have been described as problems, but the people who have
mentioned these firms have been very responsive to renumbering
requests. Elliot commented that SGI had primarily encountered
problems with network management products
There was consensus that vendors
should be given guidance on how to write applications without
hard-coding IP addresses. A short RFC was suggested on guidelines
for creating license numbers that did not depend on hard-coded
Steve Bellovin reminded PIER
that ipsec will require public/private keys for every machine,
and there was a logical choice for license seeds. It was observed
that key management for this approach has not been resolved.
Cadence and Mentor Graphics
were also mentioned as having hard-coded license keys. Netscape
1.1's SOCKS variable needs to be an IP address, but this may have
been fixed in Netscape 2.0.
Router Renumbering (Howard Berkowitz, PSC International)
RFC 1916 has come out.
Dennis O'Leary will talk to Tim about setting up a Web page. In addition, we need to talk to Alison about http protocol problem with multiple IP addresses for virtual servers. (Bill Manning will do this.)
Roger would like to see a document about not using IP addresses in applications. (Look at RFC 1900)