Performance issues versus CVS

 One of my coworkers wanted to perform a timing comparison between CVS
and Subversion. Specifically, he was interested in the time to import a
new repository, and do a fresh checkout. All operations were network
based.

 For the test, he wanted to use the latest GCC source distribution.
This equates to almost 16000 files, and over 170 M of source.

 The time differences were staggering. The comparisons were as follows:

CVS Import
 PServer, Remote: 1 minute, 35 seconds
 PServer, Remote, Z3: 2 minutes, 5 seconds

CVS Checkout
 PServer, Remote: 1 minute, 40 seconds
 PServer, Remote, Z3: 1 minute, 50 seconds

SVN Import
 HTTP: 15 minutes, 50 seconds
 SVN: 5 minutes, 30 seconds

SVN Checkout
 HTTP: 16 minutes, 30 seconds
 SVN: 5 minutes, 20 seconds

While I recognize that Import/Checkout are only 2 commands, and that
other commands may yield a better showing by SVN, I thought you [as a
SVN developer] would be interested in these times. I didn't see
anything in the FAQ, nor the book, which would suggest such a huge
discrepancy.

Additional Information:

CVS Server is on a Sun Ultra 10 [440 MHz], 1G memory, 100Mbs, Single
CPU, SCSI [10000 RPM] HD, Solaris 2.7
SVN is on a Dual Xeon 2.0 GHz, 2G memory, 100Mbs, IDE [7200] HD, Fedora
Core 1.0

The CVS server version is 1.10.7. Subversion v 1.0.1, from RPMs made
available on tigris. Apache conf is OOTB, except for SVN auth.

Has anyone seen performance issues such as these?

Thanks!

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Apr 15 18:33:06 2004