On Fri, Jun 25, 1999 at 11:07:07AM -0400, Paul Koning
<pkoning(_at_)xedia(_dot_)com> wrote:
>> (By the way, a timestamp may be better than a sequence number.)
Marcel> I don't see the advantages of a timestamp. The advantage of
Marcel> a sequence number is that you immediately know about holes,
Marcel> updates which you have missed. As long as entries are
Marcel> superseding each other, a timestamp is just fine, but if it
Marcel> is only additive (as with PGP), I feel that seqnos are more
Marcel> appropriate.
Good point, but make them bigger than 32 bits so you can avoid
worrying about wrapping (which is a problem, see Rosen). And you may
still want a timestamp, indicating when the counter was initialized,
so you can recover from loss of the counter state.
Paul,
thanks for the references. In our design, we have been using (compressable)
IPv6 addresses and 64 bit counters, although we believe that 32 bit is
more than enough. Even if 32 bits should be exhausted, by adding an additional
IP address (kludge alert), another 2^32 sequence numbers could be allocated.
I see the advantage of the time stamp, but our idea was to recover the
counter state from a query to the local DB at start-up. And if the DB
should have crashed, rebuilding it according to the information received
from the other keyservers will also re-establish counter state.
-Marcel