Re: effectiveness of rsync and apt
On 5/1/06, Brian Eaton <eaton.lists@gmail.com> wrote:
Hello all -
Regarding the ideas discussed here:
http://rsync.samba.org/rsync-and-debian/rsync-and-debian.html
A few comments:
3.2 rsync is too hard on servers
That document claims it's an avoidable cost. It's not really, because
the client presents hashes which the server has to find. The hashes
can't be cached on the server because there would need to be 8 bytes
of hash per file offset in the file. i.e. the size of the hashes would
be 8 times the size of the file. Ergo, the server needs to scan the
entirety of the file anyway. Disk i/o *is* expensive. Now, if someone
can come up with a clever algorithm to compress the hashes by > 100
times, we might have something useful.
The reverse rsync algorithm described further down avoids this by
having the client do the search. AIUI tridge avoided it because it
appears to be patented.
The rest of the page is a pretty good summary. The compression problem
is the real stumbling block...
--
Martijn van Oosterhout <kleptog@gmail.com> http://svana.org/kleptog/
Reply to: