Re: Re: GPE syncing

Hi,

Owen Cliffe <occ@cs.bath.ac.uk> writes:

> > > 2. keep trace of modifications, with a modified flag.
> I'm not sure this is sufficient for modifications. it doesn't allow you
> to apply sensible policies for collisions (where client and host are
> modified).
>
> a better approach would be a timestamp associated with each field.
> unfortunately this is quite expensive in terms of space +at least 4
> bytes for every individual entry you want to sync.
>
> the even better alternative is to use a modification log, which
> unfortunately adds a bit more complexity to either the client program or
> to the DB handler.
>

With SyncML the modified flag would be enough. After a sanity check,
the client send all the data that changed. Each modified entry is
completely sent. All the conflict resolution is done on the server
side to simplify as much as possible the client.

So to make an efficient "conduit", the only thing needed is a
"modified" flag, because if such a flag doesn't exist the conduit has
to keep for each entry a signature (MD5 for example) and compare it
with the current entry, this is far less efficient.

Best regards,

-- 
Christian Gillot <cgillot@neo-rousseaux.org>
Développeur GNU/Linux
http://www.neo-rousseaux.org/cgillot