Re: [rdiff-backup-users] incremental or differential backup

[Andrew Ferguson]

On Jan 16, 2009, at 9:02 AM, Ryan How wrote:
> I've also wondered the same thing with restore times. Maybe we  
> should give it a test?. I know backing up my virtual machine images  
> (around 30GB single file) takes a LONG time to make the diff. a  
> couple of hours. But it isn't running on a super machine. That  
> essentially is the same as a database - big file with many small  
> changes. But I would assume that applying a diff would be much much  
> quicker than generating one because we aren't comparing a whole  
> file, we are just adding bits and removing bits in certain sections.


Without testing, I would draw the opposite conclusion. When generating  
a diff, we are merely asking the filesystem to read the entire file.

When applying a diff, you need to read the entire file, and change it,  
while writing it back to a new file. Most OS's do not store files in  
any way such that "adding bits and removing bits in certain sections"  
is an in-place operation. If you edit a file in OpenOffice, it reads  
the whole file into memory, you make your one change, and then it  
writes the whole file back out to disk as a temp file, delete the old  
one, and rename. (Or just rename over existing one for atomicity, but  
that doesn't work on every OS.)


Andrew