Dropping old revisions creates larger repository

[Erik Hemdal <er...@comprehensivepower.com>]

Dropping old revisions creates larger repository

I was looking to test what happens when one uses a dump/load cycle to remove old
revisions in a repository.

I moved my dump to a test machine so all my results are on a test box instead of
the production repository.  This is using v 1.4.2 on Fedora Core 6.

1.  I restored my production dump to a new repo.  The total size of the
repository is 1340664 bytes (results from du command).

2.  I looked at the size of the original dump file, it's 3168367372 bytes long. 
So far so good.

Then I dumped the test repository and left off the first 99 revisions

	svnadmin dump ~/test-repo -r 100:580 > dump-smaller

I reloaded the dump into a new repository.

The new repository looks just as I want it; the early revisions are gone, and
the revisions are renumbered.  I took a dump of the new repository. I wanted to
see how much smaller things would be after losing the early history.

The new repo takes 1378148 byes.  The new dumpfile is 3230834074 bytes long. 
The new repo with fewer revs is larger, so is the dumpfile.

Has anyone else seen this, and can anyone explain why this is happening?  I
could believe that the repository is a little bigger perhaps, but not the
dumpfile.  What am I missing?

Thanks for any ideas.    Erik

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

RE: (SOLVED) Dropping old revisions creates larger repository

[Erik Hemdal <er...@comprehensivepower.com>]

> If you had branches or tags or other copies of the revisions 
> that you had deleted, the fresh repository would now 
> replicate all of those files and thus take a lot of extra 
> space.  I would assume that there are a number of files that 
> exist both in trunk and various tags/branches.  Each of these 
> gets bloated to its full size assuming that the original file 
> was in the set of revisions you filtered out.

That makes sense.  The early revisions in my sample included a lot of
branching, and I can see how that could bloat the repository.  My cheap
copies got more expensive because I dropped the revisions that allowed them
to be cheap.  

Now I feel better about this.  Thanks very much, Steve.  Erik


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org

Re: Dropping old revisions creates larger repository

[Steve Bakke <st...@amd.com>]

If you had branches or tags or other copies of the revisions that you had
deleted, the fresh repository would now replicate all of those files and
thus take a lot of extra space.  I would assume that there are a number of
files that exist both in trunk and various tags/branches.  Each of these
gets bloated to its full size assuming that the original file was in the set
of revisions you filtered out.

Could dumpfilter be made more intelligent such that it would recreate the
deltas between trunk, tag, and branch versions of the same file?  (does that
even make sense?)

-steve

Perhaps if the dumpfilter was smart enough


On 2/2/07 2:22 PM, "Erik Hemdal" <er...@comprehensivepower.com> wrote:

> Dropping old revisions creates larger repository
> 
> I was looking to test what happens when one uses a dump/load cycle to remove
> old
> revisions in a repository.
> 
> I moved my dump to a test machine so all my results are on a test box instead
> of
> the production repository.  This is using v 1.4.2 on Fedora Core 6.
> 
> 1.  I restored my production dump to a new repo.  The total size of the
> repository is 1340664 bytes (results from du command).
> 
> 2.  I looked at the size of the original dump file, it's 3168367372 bytes
> long. 
> So far so good.
> 
> Then I dumped the test repository and left off the first 99 revisions
> 
> svnadmin dump ~/test-repo -r 100:580 > dump-smaller
> 
> I reloaded the dump into a new repository.
> 
> The new repository looks just as I want it; the early revisions are gone, and
> the revisions are renumbered.  I took a dump of the new repository. I wanted
> to
> see how much smaller things would be after losing the early history.
> 
> The new repo takes 1378148 byes.  The new dumpfile is 3230834074 bytes long.
> The new repo with fewer revs is larger, so is the dumpfile.
> 
> Has anyone else seen this, and can anyone explain why this is happening?  I
> could believe that the repository is a little bigger perhaps, but not the
> dumpfile.  What am I missing?
> 
> Thanks for any ideas.    Erik
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: users-help@subversion.tigris.org
> 
> 
> 



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org