Currently the nightly testing results are collected on krbdev.mit.edu
through my running a script with my Kerberos tickets and my SSH key. 
Some of the results (mostly the ones not on my personal machines) could
be collected by other people on the Kerberos core team, but it would
still be a manual process.

This should be automated, so we can view all the results without having
to fire off a script to fetch them and then wait for it to complete.

It should also be reversed, so that (properly authorized) nightly
testing machines can submit results when they're ready.  (Things to
consider: Don't let a machine fill the server's disk, or write arbitrary
files.  Intermediate results -- "I'm building now"; "I've been compiling
this same source file for three hours" -- might be helpful too.  If
intermediate results are submitted, try to avoid submitting any log
files twice, using timestamps, rsync, whatever.  If a build tree is
blown away and the build restarted, don't mix up the two sets of
results; probably throwing away the first set is best.)

If the results are submitted when they're ready, we probably don't need
to keep the build trees around -- at least in the successful cases. 
Deleting the build trees means the space can be reused for other builds,
so we could test more combinations of versions, build options, etc.

Authorizing build machines to submit results means machines I don't have
access to could be used for testing, e.g., newer versions of AIX or IRIX
than we have at MIT right now.