[cvsnt] performence problem

[John Peacock]

Stéphane Nicoll wrote:

>> Stéphane, what's the pattern of CVS use at your organisation?
> Around 50 developers. 5-10 updates of the repository per day per
> developer. Around 3 commits per day per developer I would say. One
> cruise control, running every 24h. One Jira instance retrieving CVS
> logs every hours.


There's no excuse for such performance issues with such a [relatively] 
small number of hits on the server.  Can you completley turn off AV 
software for a day to see how that changes things?  The reason I ask is 
that most AV software hooks into the filesystem at a very low level 
(basically every read/write), and even if you disable active scanning 
for a given directory, the hooks still get called on every read/write to 
the disk.  Some AV software is worse about this than others.

The other thing to consider is whether the 8000 Java files you have are 
all in a small number of directories.  NTFS doesn't scale very well with 
large directories; the effect can even be felt higher in the tree (i.e. 
not just in the directory with lots of files).

If you have the option of reconfiguring the server, you can change the 
default block size to something larger than the default; the file 
contents will then be stored in the directory entry under NTFS.  But 
remember, just because the files are ~8k when checked out, the full 
history of all changes is stored in each file, so the repository files 
can be significantly larger than the checked out copies.

HTH

John