Geoff Rowell <geoff.rowell_at_gmail.com
<mailto:geoff.rowell_at_gmail.com?Subject=Re:%20SVN%20Performance%20Focus>>
wrote:
I think it's great that the development team has been looking into 1.7
performance, but I must admit I'm less interested in server
performance than that of the client. My recent experiences with huge
working copy folders has pointed out significant disparities in
performance under different operating and file systems. Meaningful SVN
performance testing should take into account the majority of popular
operating and file systems.
SVN uses the same libraries for basic services on the client
as it does on the server. Most of my optimization so far happen
focus on these parts:
* zlib (diff & un-diff, network data compression etc.)
* apr (windows file I/O)
* apr util (MD5 checksumming)
* svn-internal file I/O overhead
And, as Greg already said, the client-side is undergoing
a major rewrite. That is also why I concentrate on the
server-side: it is stable ground and I'm less likely to
optimize temporary artifacts.
My corporate experience indicates that hardware upgrades are usually
better tied to budgetary cycles and operating system upgrades.
But that is actually a perfect fit: SVN can be treated as a
"set-up-and-forget" server. New SW versions will often
at least suggest client updates etc. IOW, it requires a certain
amount of coordination and previous testing. As a result,
the server SW will be upgraded at most once during the
HW lifecycle.
So, if you are planning to roll out a new server version,
you can also take the opportunity to configure the HW
according to your needs. Some guideline can be found
in my recent post:
http://svn.haxx.se/dev/archive-2010-05/0247.shtml
-- Stefan^2.
