Re: IAB's proposals in response to the work of the ROAD group

COMMENTS ON IAB STATEMENT

The following is a set of initial comments regarding the recent statement
from the IAB suggesting the adoption of TUBA and CIDR as a practical way
forward in solving the impending scaling problems as the Internet grows.
The comments are a result of discussions in the Joint Network Team's
IP Technical Advisory Group (which advises the JNT on technical issues
affecting the operation of the IP service over JANET, amongst other
things), but the overall conclusions are mine, made in the context of
managing the JANET IP service.

Of course, comments at this stage can only be preliminary, as most people
here have not had sufficient exposure to the technical arguments as yet. I'm
sure this situation will change as we study the pros and cons more deeply.

Bob Day
Manager, JANET IP Service
Joint Network Team, UK

1. We believe that it is a sensible policy to try to set the technical
   work in a defined direction; if it is necessary to act in the short
   term then this should be done using technology that is sufficiently
   mature that it will be reliable in practice, and relatively
   inexpensive to deploy.

2. We believe that if ISO CLNP is used as a replacement for IP to
   alleviate the scaling problem then it is important that attempts are
   not made to "improve" it, and hence to depart from the Base Standard
   (and preferably not to introduce a new profile, if an existing one
   can be used instead). Given that we see the use of ISO protocols as
   a desirable strategic goal, it would be counter-productive to
   introduce non-standard variants at this stage in the life of CLNP.

3. We have serious concerns, however, about the technical suitability
   of CLNP to replace IP in the longer term. These revolve around the
   suitability of CLNP for mapping onto high-performance networks (ATM
   in particular), and whether CLNP has associated with it a
   sufficiently functional routing capability to really solve the
   routing problems of the global Internet in the longer term.
   Therefore we would not like to see CLNP as a long term goal at the
   expense of a better network protocol (whatever this might turn out
   to be). This implies that the resource put into deploying CLNP
   should be kept as low as possible -- ie by using existing products
   without expensive developments and "enhancements", and concentrating
   the bulk of resources on a technically more far-reaching solution.

4. We have some concern that the details of the TUBA proposal are not
   sufficiently worked out, even at the top level. In particular it is
   not clear at this stage that the simple model described of hosts
   determining the capability of other hosts via the DNS takes into
   account the problems of intervening regional networks that do not
   support TUBA. Without some strategy at this level, as well as at
   host level there seems to be a danger of either unreachability or
   excessive traffic due to failures using CLNP followed by retries
   using IP.

5. We will support the use of classless addressing on the JANET IP
   Network as soon as we can do so. However, we note that there may be
   cases where an attached site network cannot use classless addressing
   itself due to limitations of its equipment (and of money to do
   anything about it), and suggest that the NICs should take this into
   account when assessing a request for a Class B network number -- ie
   it may not just be a matter of number of hosts that determines a
   site's requirement, but also one of the technology it has
   available.