statcvs-users

[Statcvs-users] Latest CVS-Checkout of statcvs terribly slow

[Pedot, W. <wp...@ha...>]

Hi,

I am using statcvs on the ever growing repository in my department
(SW-Development). Until yesterday I used a slightly modified version I checked
out last year (my build date shows 12. December), now I decided to get the
newest available development version and checked that out yesterday. The build
and a first test with cvsweb-integration went fine but when generating the whole
stat I noticed that the new build is awfully slow.

My older Version took ~40min to create the statistic (~7.5Million LOC, one
overall-stat and 6 sub-stats so everything is done twice), the new Version did
not manage to complete the overall statistic in 2h (had to interrupt it). 

I excluded the two biggest parts now (2 modified Linux-Kernels, gonna be 3 in
the future) and the runtime dropped to 15min on ~1.1Million LOC.

Is there any explanation for this slowdown? One thing is the cvsweb-integration
but even without it its by far slower.

greetings
Wolfgang Pedot

Re: [Statcvs-users] Latest CVS-Checkout of statcvs terribly slow

[Richard C. <ri...@cy...>]

Hi Wolfgang,

From: "Pedot, Wolfgang" <wp...@ha...>
> [...] I noticed that the new build is awfully slow.

For small repositories, the current build is *a bit* slower because of
additional charts, and requires *much* more memory because we use a
different collection class (tree-based instead of array-based).

The larger memory footprint turns into a heavy performance hit when the JVM
has to garbace collect all the time or starts swapping. I guess that's
what's happening in your case.

So here's what you can do (not very useful, I know ...):

- buy more RAM

- fiddle with the -mx setting to set the max amount of RAM for the JVM:
  "java -mx512m -jar statcvs.jar ..."

- use the -verbose switch of StatCvs to see if it's still running and
  how much memory it needs (shown after finishing)

Richard