[cvsnt] Re: cvsnt Vs cvs : Performance!

[Tony Hoyle]

nick.minutello at uk.bnpparibas.com wrote:
> Checkout pef was about the same - 2.5 mins for a large module. cvs hardly
> showed up on top - peaking at 6or 7% cpu... cvsnt hit around 20%.
> However, update perf was *wildly* different.
> cvs took 6 sec, cvsnt took 15 sec. However on the cpu stakes, cvs used
> about 4-8% whereas cvsnt hit 75% at some point - and a lot of time at 50%!!

cvsnt is doing more work - cvs does not have access control, or 
directory/file rename mapping, or file-level locking, for example, so 
you'd expect some differences... there's a *lot* of work going on now. 
I get a peak of about 15% for a full update - as long as there's unused 
CPU that's fine... it's there to be used.  cvslockd barely registers 
more than 2-3% most of the time.

Direct comparison isn't easy... I couldn't run the old cvs on any of my 
repositories because it simply couldn't handle them.  Basing it around 
remote access (sourceforge vs cvs.cvsnt.org, to a US client) cvsnt comes 
out just slightly ahead on operations involving large files - because it 
does more in memory - and behind with many small files, as it does a lot 
more work per file.

2.0.58d is faster than the old 2.0.51d, btw. and 2.0.6x is faster still 
in some cases (maybe slower in others).

Tony