Hello,

I've an inquiry regarding a proposal listed on the NetBSD
projects page--a 'drop-in' replacement for rsync.

The proposal mentions that the replacement should handle
large files, and large sets of files, more efficiently.

It's somewhat unclear as to wether this refers to transmission
efficiency, or cpu/io load.  Furthermore, one would expect
a 'drop-in' replacement for rsync to use the same protocol,
a requirement somewhat at odds with performance improvements.

I've looked at Andrew Tridgell's rsync paper, and it seems to
me that there are several fairly obvious trade-offs possible in the
rsync algorithm that can reduce the amount of data sent for
large files, or reduce the load on the server, at the expense
of latency.

A final question is wether there's really any pressing need
for a BSD-licensed rsync replacement.  NetBSD has sup and
mtree in base, which provide for much of rsync's functionality,
if not its performance.  
Unless I'm mistaken, csup (http://www.mu.org/~mux/csup.html)
provides a BSD-licenced synchronisation tool that uses the 
rsync protocol.

The rsync protocol can be improved to decrease server load, but
unless there are sound reasons to do so, a project to write
a replacement seems like a gigantic waste of effort to me.
-- 
		-- Michiel