Re: CVS migration to subversion?

[David NEARY <dneary wanadoo fr>]

James Henstridge wrote:
> Peter Williams wrote:
> >Another potential problem is that their metadata seems pretty huge. I
> >checked out the source to svn and the working directory was 48 MB; after
> >"find -name .svn |xargs rm -Rf", it was 15. The gnome CVS mirrors seem
> >to keep on running out of space as it is, and svn would probably not
> >help in that regard.
>
> Of course, the size of working copies is not very relative to the anon 
> cvs servers.  The size of the repository is much more relevant.  I don't 
> know what the relative sizes of SVN and CVS repos are.  I would guess 
> that in some cases SVN would come out on top due to being able to store 
> binary diffs.  It would be interesting to know what the difference is 
> for a source code base like Gnome CVS.

You might think that, but usually the overhead of Berkeley DB log files pushes the size up considerably. These can be cleaned out once in a while (if you're as paranoid as the subversion guys, you only clean them out when they're backed up to tape - the Subversion repository is entirely rebuildable from tape backupped Berkeley DB logs), but basically every transaction generates a log entry which is roughly proportional to the size of the change being made, in addition to the actual change taking up about the same amount of space in the database. Cleaning up logs after a big import can save you a huge amount of space, but the DB overhead makes it unattractive if serverside storage space is at a premium. 

That said, for projects with lots of binary data, binary diffs would have to save space at some stage. In general, though, I would guess that a CVS repository would not be bigger than its equivalent CVS repository.

Cheers,
Dave.

--
Dave Neary
Marseille, France