Re: agenda for dnsind 99.3.16

[Robert Elz <kre@MUNNARI.OZ.AU>]

    Date:        Thu, 18 Feb 1999 13:16:48 -0400
    From:        Jeff Schreiber <schreiber@PROCESS.COM>
    Message-ID:  <009D3EC9.676FCEE6.95@process.com>

  | Randy Bush <randy@psg.com> writes:
  |> rfc 1982 - serial number arithmetic
  |>   o to ds, but needs interop report, not obvious
  | 
  |  We tested in LA... I'm not sure if there is more nessessary than
  |  testing that the secondaries recognize the wraparound?


We know, the basic interop report on DNS server use of 1982 is easy
(there are many interoperable implementatiosn, and they have been tested).

The issue with resppect to 1982 is the requirement that all features of the
RFC have been implemented (more than once).   For better or worse, 1982
specifies serial number artitmetic for more than just 32 bit serial numbers.

I have an implementation of 1982 that works on (almost) any word length
(from 2 up to BIG), but I am not aware of any others.   This is the "not
obvious" part of writing the interop report for the IESG.

Now perhaps we can get away with just its uses for the DNS, and that will
satisfy everyone, and most likely this is what we should try - but to
be technically correct, the IESG ought require that all those parts that
have not had multiple implementations be stripped out of the doc before it
advances.   For 1982 that would mean an almost total rewrite, and would
most likely be so significant as to require it start again at PS.

Of course, if someone else would like to do some implementations of 1982
for other than 32 bits (I don't think it need be continuously variable,
as long as there is testing of at least 2, or better 3, word lengths, it
should be OK) that would certainly make life easier...

kre