efficient file appends

Dave Dykstra dwd at bell-labs.com
Fri Dec 14 01:18:15 EST 2001

Welcome to the rsync list, Phil.  Martin and David provided excellent
responses, but there's one more small point I want to make:

On Wed, Dec 12, 2001 at 03:42:59PM -0800, rsync at ka9q.net wrote:
> The main drawback I can see to exchanging hashes by
> default is the extra receiver CPU time consumed in producing them.

Another drawback to your original idea is that rsync would either need to
make an extra pass through the file, meaning there's extra disk access, or
rsync would need to accumulate all the per-block checksums in memory.  The
rsync algorithm already reads through the original file twice on the
receiver side, once to generate checksums and once to pull out the old
blocks that are still good.  Depending on the comparitive speed of "disk"
access (which may actually be a remote file system) and the speed of the
network between the two rsync halves, it may or may not pay.

This really is a minor point because as Martin pointed out, the fact that
rsync's pipelining would be defeated means that the idea would have a
drastic effect on throughput.

- Dave Dykstra

More information about the rsync mailing list