Cygwin performance results linux-headers-2.6.21-2: bzr 0.18 vs hg 0.9.4
Robert Collins
robertc at robertcollins.net
Fri Aug 3 02:11:36 BST 2007
On Thu, 2007-08-02 at 20:36 +0300, Jari Aalto wrote:
> I ran tests under Cygwin as well for comparison. P II/400
> W2ksp4/latest Cygwin. Both programs are native Cygwin applications
> with Cygwin Python 2.5.
>
> The difference is huge. Let me know if I can help profiling.
Wow the difference certainly is significant. If you could get lsprof
output for some of this it would be useful. Also IIRC terminal output is
extremely slow for cygwin, so I'd like to know --quiet/--no-verbose
figures for these.
More below
> Jari
>
> bzr 0.18
>
> $ bzr init
> real 0m52.398s
> user 0m2.814s
> sys 0m8.802s
This is amazingly slow. My guess here is that it is the time to load bzr
from disk. hg has many fewer python modules, and I really should finish
my import simulator off given this info. For comparison:
$ time bzr init bar
real 0m0.464s
user 0m0.332s
sys 0m0.100s
Also, what plugins do you have installed?
> $ bzr add .
> real 2m56.609s
> user 0m16.854s
> sys 0m16.403s
Again a huge difference between user+sys and real. That suggests
blocking IO (e.g. terminal or waiting for disk) occuring.
> $ bzr ci -m "- commit"
> real 10m56.302s
> user 2m4.989s
> sys 1m1.748s
And again, about the same ratio.
I think it would be fantastic if you can spend some time profiling and
determining where the time is going. We're serious about making bzr fast
- and on linux we are comparable to hg for many operations today, so
this suggests something that cygwin is bad at, and that we make heavy
use of, interacting.
-Rob
--
GPG key available at: <http://www.robertcollins.net/keys.txt>.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20070803/758ee637/attachment.pgp
More information about the bazaar
mailing list