rsync vs. rcp

[Donovan Baarda]

On Thu, 2003-02-20 at 08:55, James Knowles wrote:
> > RSYNC DOES NOT WORK WITH 1GB+ FILES... unless you have a sufficiently
> > large block size. 
> 
> According to the archives, block size doesn't fix anything. At any rate, I'm 
> highly disappointed that rsync is relying on statistical good fortune.

increasing block size does increase the probability that the rsync
algorithm will succeed.
 
> We've used rsync extensively in our company for moving information around. 
> With tens of gigabytes of data at stake, it's very disappointing. Now we 
> have to dump rsync, redo all of our backups, and possibly write our own 
> tool. I don't want to futz around with something that I can no longer trust.

If you read the thread, rsync the tool is reliable, just the rsync
algorithm does not work (as currently implemented). The rsync tool will
fall back to a whole file transfer if the rsync algorithm fails, as
indicated by a full md4sum of the whole file. This does mean rsync the
tool will end up transferring the file twice when this occurs; once
using the rsync algorithm, and again sending the whole file.

As for rsync relying on statistical good fortune, that should have been
clear from the beginning. However, because of the whole file checksum
used to verify the final transfer, there is a 1/2^128 chance of it
getting it wrong per file. That is less than 1/10^36 chance, or 1 in
so-many-billions-of-billions-of-billions-that-its-not-worth-worrying-about.

So if you spend from now until eternity transferring every file that
ever existed again and again... you might get one file wrong, but not
very likely.

I think there is a higher probability of getting corruption doing a 'cp'
directly onto the same HDD (cosmic rays etc).

-- 
----------------------------------------------------------------------
ABO: finger abo at minkirri.apana.org.au for more info, including pgp key
----------------------------------------------------------------------