Steven Bakke wrote:
>>> However through profiling we found a performance bottleneck in that
>>> it starts a new repository session (network connection) once for each
>>> target of the command. In our case the problem was with 'svn
>>> update', but I think it applies to almost anything. The performance
>>> became dominated by ~2-3 seconds per target overhead each time it
>>> made a new svn+ssh connection to the server.
>>
>> What can possibly cause a several-second delay? Does your reverse DNS
>> lookup resolve quickly? Are IDENT queries dropped and timing out?
>>
>
> I don't know for sure. First, there is the initiation of an SSH
> connection. That is followed by creation of an ra_session. I'm
> wondering if we could do a comparision by using the file:// access
> method. I'm not the person who did the code profiling.
Some overhead in starting a new ssh session would be normal, but several
seconds is an enormous number of CPU operations these days so I'd look
elsewhere for this kind of delay. As you connect, sshd is probably doing
a reverse DNS lookup on the connecting IP address and an IDENT query to
the connecting client for the user owning the socket, and the time for
those reponses (which may have nothing to do with the svn server itself)
may dominate the elapsed time you measure. Getting a big improvement
might be as simple as turning on a local caching nameserver or changing
logging options.
--
Les Mikesell
lesmikesell_at_gmail.com
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: users-help_at_subversion.tigris.org
Received on 2008-01-04 17:37:02 CET