On Fri, Oct 9, 2009 at 9:19 AM, Roman Korchagin
> We have started using merge (because it supports merge tracking now) to work
> more in the agile style, but we are finding that we cannot actually.
> Checkout and switch are working very well. The data is flowing at a good
> speed, let’s say 200Kbps over the channel that we have.
> But we are seeing 6 minutes to merge small changes in 20 files or so (30
> seconds per file?). There are no large directories, just a normal project
> approx 1500 source code files, no more than 50 per directory. In the packer
> analyzer I’ve got only 1.2MB of traffic in 2,316 packets. Could not look
> into them because they are HTTPS. The network graph is around 20Kbps. The
> server appears to be idle in everything. The client is not stressed either.
> Ping is 551ms if that helps. Both are waiting for what?
> With a bit more changes between revisions we are easily hitting 1 hour or
> more to do a merge or integration...
> Any help is greatly appreciated.
Make sure you run svn up before you merge. Having your working copy
at a single revision, and perhaps even at HEAD, allows the merge
process to take some performance optimizations.
Other than that, check your working copy for subtree mergeinfo. If
you have a lot of nodes in your tree with the svn:mergeinfo property
set, then there are a lot of extra server roundtrips to perform. If
all of your mergeinfo is at the tree root, the merge performance
should be on-par with pre merge tracking versions of Subversion.
To unsubscribe from this discussion, e-mail: [users-unsubscribe_at_subversion.tigris.org].
Received on 2009-10-09 19:44:01 CEST