I just thought we should feel a little proud today: we lost a server due
to an unknown bug.  All volumes on it were read-write replicated volumes
and Coda's replication/resolution features can be used to "repair" such a
lost server - this happens through resolution which kick in automatically
by doing "ls -lR". 

It is an absolute first (I believe) that we got 1.1 GB of data resolved
without any server crashes or hangs -- and in sharp contrast with
experiences only 10 months ago.

Puneet Kumar was responsible for the design and implementation of
resolution, and much to his credit we achieved the current state of
affairs by me only fixing between half a dozen and a dozen bugs -- I must
say that I found some of them really hard to find, involving distributed
locks, subtle errors in the directory code and referencing some
unallocated memory here and there. 

Last year we felt that fixing the resolution bugs was a priority we will
now be migrating our attention to the area of low bandwidth connectivity,
hoarding, repairing and reintegration.  

For more information about Coda, see http://www.coda.cs.cmu.edu/.

- Peter Braam - 
Carnegie Mellon University