Serf vs. Neon Performance tests

I have not seen much info on whether serf is really better than neon
for subversion in the real world. Everything I've read is hearsay or
theoretical, so we ran some informal tests. I thought I'd pass this on
for anyone interested. I do not have any actual recorded data; this is
what the developers have reported to me.

All of our tests used TortoiseSVN on Windows as the client. The server
is running SVN 1.6.3 over https: We only timed checkouts. However, I'd
expect similar results with updates and commits.

On a local network, we saw no significant differences between serf and
neon. However, over a WAN with long latencies (typically 100ms or
greater) the results were dramatically different. Checkouts using serf
were anywhere from 1x to 4x times faster than neon. The most
improvement happened when checking out a large number of small files
(typical large source code project). There was little or no difference
checking out a few large files (a library with headers, for example).

So my conclusion is that serf has no real benefit on a local LAN, but
can offer significant performance improvements over a WAN, especially
when working with many files.

-- John

------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2385236

To unsubscribe from this discussion, e-mail: [users-unsubscribe_at_subversion.tigris.org].
Received on 2009-08-19 16:13:04 CEST