Mark Bober <bober_at_wustl.edu> writes:
> What's the current wisdom on doing checkouts to NFS and CIFS?
>
> I'm seeing general checkout times of a given repository in the 2
> second range for local disk, and that jumps all the way up to a minute
> and a half or more going over CIFS or NFS (Windows 2008R2 fileserver,
> running OpenText NFS for the NFS side), in either 1.6 or 1.7.
>
> An 'svn export' of the same repository to the same mounts hovers
> around 20 seconds, revealing that 'checkout' is doing a lot more work
> that's not going well over the remote mount.
>
> This subject doesn't seem to come up very much, and I may have missed
> the key post that would answer this for me. I'm aware from my research
> that Subversion may had been attempting certain locking operations,
> and also tended towards a lot of file renaming/duplication that made
> it less than efficient over NFS/CIFS.
http://subversion.tigris.org/issues/show_bug.cgi?id=4176
If you rebuild 1.7 with the patch in the issue then the working copy
operations are much faster. The problem with the patch is that it
changes the behaviour of the library so that some applications may not
work. The Subversion command line client works in most cases and if you
also apply r1336442 then it works in all cases covered by the Subversion
regression tests. I don't know whether other applications such as
TortoiseSVN, Eclipse, etc. work.
--
Philip
Received on 2012-08-15 10:56:05 CEST