On Mon, Jan 11, 2010 at 6:45 AM, Chris Webb <[email protected]> wrote: > >> Using a standardized format may be better such as VCDIFF. > > >From the metadata, all you know are which blocks are identical and which > blocks have been written to and are potentially different. You're going to > have to read the changed blocks to send them anyway, but wouldn't doing > VCDIFF require reading the whole image including unchanged data blocks as > well to do the pattern matching, so losing all the disk IO advantages?
I mean how about using an existing format as the output of `shepherd diff' instead of creating a new format. I think the VCDIFF (RFC3284) format can be one of candidates. http://tools.ietf.org/html/rfc3284 Xdelta and open-vcdiff use this format. Of course, we don't need to read the whole image. Regards, Kazutaka Morita -- sheepdog mailing list [email protected] http://lists.wpkg.org/mailman/listinfo/sheepdog
