The data setup here is a bunch of Win2K workstations and a bunch of Win2K fileservers, with data liberally distributed among the lot. I've got ~10 GB of files I'm trying to synch up from a fileserver to my desktop.

Since I've got Cygwin installed, I say to myself: "Self, sounds like a job for rsync".

Well, it more-or-less gets the job done, but it takes its time doing it....

So I think about what I'm doing, and how rsync likely tackles the problem.

Normally, rsync is used in one of six modes, the main ones being local-local, local-remote, remote-local, and file listings (variants include use of rsync server vs. remote shell/ssh sessions). What makes rsync useful is that it only copies updated files, and it only transfers that part of the data which has been modified.

Potential problem here is that rsync may think it's got a local-local situation going on where it's actually got a networked connection, and is being profligate with its data. Anyone here familiar with internals and how / what rsync's tossing over the local network? It does seem to improve on the situation somewhat, but I'm seeing runtimes of several minutes to negotiate what's actually a pretty small bit of effective data transfer.

Sample session summary:

\r\n

\r\nwrote 1247944957 bytes  read 468 bytes  4665216.54 bytes/sec\r\ntotal size is 10577091343  speedup is 8.48\r\n\r\nreal    4m27.345s\r\nuser    0m27.608s\r\nsys     1m8.358s\r\n

\r\n

So: is this the best that can be done, or could the situation be improved on further?

...though it's a bit hard to see how this would be done without\r\nforcing some work on the part of rsync.

One athought is that I'm actually doing more work than if I'd simply\r\ncopied the files. If rsync has to compute a hash or checksum from the\r\nsource files, then I have to read the remote files anyway, and\r\nI'd be better off simply copying them, rather than reading them,\r\ncomputing a checksum, then re-reading them for the copy.

OTOH, if rsync can look at a file and say "the size and modification\r\nitmestamps are consistent with the existing copy I'm supposed to be\r\ncomparing this to, let's punt and say the files are the same". In the\r\nlatter case, I'd be seeing some significant benefits.

Moving 10 GiB across a network in four minutes isn't bad for a day's\r\nwork.... Hell, it took ~15 minutes just to get linecounts of the data\r\n(granted, gunzipping it to a pipeline), so I guess I'm getting my\r\nmoney's worth.