On Fri, Oct 31, 2008 at 4:18 PM, Mark Eichin <eichin_at_gmail.com> wrote:
<snip>
> A friend just pointed out that the reason it isn't needed anymore is
> the ControlMaster/ControlPath feature of modern ssh, which takes care
> of it directly; looks like you can do something like
>
> ssh -o ControlMaster=yes -o ControlPath=/tmp/mysocket svnserver
> and leave that around, then
> export SVN_SSH="ssh -o ControlMaster=no -o ControlPath=/tmp/mysocket"
> and use svn+ssh as you already do... I haven't tried this myself yet
> to see how much of a performance difference it makes, just confirmed
> that the commands work...
>
Thanks Mark,
This does seem to speed things up. Something that was
taking 64 seconds before is taking 34 seconds with this
ssh ControlMaster trick. But it takes about 2.5 seconds when I
point directly to the repository using file://. Maybe I should
just always use file://... the repository is accessible via NFS
export and the faq seems to imply that NFS access is ok
for fsfs back end. I can't remember why we're using ssh
and not just file://. Anyway, thanks for all your responses.
Cheers,
David
PS - I forgot to mention... I saw a bunch of these errors
in the original "ControlMaster" ssh session as I was
accessing the repository via svn+ssh:
channel 1: chan_read_failed for istate 1
They didn't seem to hurt anything as far as I could tell
although I guess it's possible that they're a clue as
to why svn+ssh still takes >10x as long as file access
to the repository via NSF.
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: users-help_at_subversion.tigris.org
Received on 2008-11-02 19:01:33 CET