[comp.protocols.time.ntp] Cornell time server

Mills@udel.edu (05/22/91)

Folks,

You probably have not noticed that erstwhile NTP stratum-2 server
lilben.tn.cornell.edu (192.35.82.50) has been retired and replaced
by a Unix box running xntp at the same address. While the accuracy
may have suffered a factor of ten, most of you won't care much
about that. However, the xntp configuration does not include
crypto-authentication, which has been a required feature of all the
fuzzball primary and secondary servers, as well as some Unix/xntp
installations that cared to participate in the top-level system.

It might well be that the Cornfolk may elect to bring up the crypto
and I encourage them to do that; however, this brings up the wider
issue. As evident from the care I put in the spec and in my own
implementations, I feel that any server which can be wholly trusted
must maintain an solidly authenticated path to the primary server(s)
and, at all costs, the primary servers must maintain fallback paths
using their own secret keys. As excuse for my paranoia, consider the
case where somebody manages to warp a bunch of primary or secondary
servers with a massive attack of fake stratum-1 falsetickers. You
may kiss your archives goodbye. 

Therefore, I would like to encourage those of you who care and who
have the necessary oomph to craft your xntp configuration tables
accordingly to bring up the authentication option. If you manage
to do that and are willing to burn the modest CPU cycles to munge
the crypto-checksum, I would be glad to supply the requisite key(s).
Sure, this informal key-distribution scheme will be a laugh to my
more learned friends, but it may also be pragmatically prudent.

Dave