[darcs-users] Benchmark results.

[Petr Rockai]

Eric Kow <kowey at darcs.net> writes:
> === big-zoo/darcs-repo ===
>
>           || darcs-2.0.2      | darcs-2.1.0      | darcs-2.2.0pre1
> ==========++==================+==================+=====================
>  annotate ||       14.3s 104M |       14.0s 103M |           14.2s 107M
>     check ||        9.0s  14M |        9.3s  13M |            6.4s  95M
>  get full ||        1.7s   4M |        1.1s   4M |            1.1s   7M
>  get lazy ||        0.2s   0M |        0.2s   0M |            0.2s   0M
>  pull 100 ||        2.5s  15M |        2.4s  14M |            2.3s  14M
> pull 1000 ||       19.0s  27M |       19.0s  27M |            9.8s  30M
>    repair ||       14.9s  77M |       23.2s 110M |            6.4s  95M

=== big-zoo/darcs-repo ===

          || ../branch-2.2/darcs     | ../release-2.0.2/darcs    
==========++=========================+===========================
 annotate ||               16.1s 64M |                  16.3s 63M
    check ||               16.0s  8M |                  25.4s  4M
 get full ||               11.3s  2M |                  13.1s  2M
 get lazy ||                2.0s  0M |                   2.2s  0M
 pull 100 ||                9.4s  7M |                  10.3s  7M
pull 1000 ||               74.9s 18M |                  87.8s 16M
   repair ||               16.6s 10M |                 184.7s 44M

A few remarks:

- This is a comparable system (2x2 core2 xeons, 16GiB RAM), but with NFS
  home. The NFS makes a huge difference in cost of IO. The difference between
  check and repair in 2.0.2 is due to check running in /tmp, while repair runs
  in t he repo itself. (Note that tmpfs on /tmp would likely further speed up
  check. Kowey, do you have one?)

- The "branch-2.2" version of darcs has my repair leak fix applied. You can
  notice how memory use in repair went down from 44M to 10M. Check went from 25
  seconds down to 16, repair from 184 seconds down to 16.

- I wish I knew why 2.0.2 took 77M in Kowey's case and only 44M in mine, doing
  repair.

Anyway, I'm re-running all of the benchmark suite on a local filesystem on the
same machine. I'm especially interested in ghc-hashed and frugalware-hashed
results for check and repair. I'll report when the benchmarks finish.

Yours,
   Petr.

PS: Have a nice Christmas! : - )

-- 
Peter Rockai | me()mornfall!net | prockai()redhat!com
 http://blog.mornfall.net | http://web.mornfall.net

"In My Egotistical Opinion, most people's C programs should be
 indented six feet downward and covered with dirt."
     -- Blair P. Houghton on the subject of C program indentation