performance of 2.4 under Linux
Bert Freudenberg
bert at isgnw.CS.Uni-Magdeburg.De
Fri Jul 16 15:29:15 UTC 1999
On Fri, 16 Jul 1999, Ian Piumarta wrote:
> > when running with a remote X server, BitBlt operations appear to
> > be very slow.
>
> Ah, now I understand. The bottleneck is your network.
That's only half of the truth. Squeak's display performance seems to be
very much dependend on the X server used. Here's a little benchmark that
tests, in different depths, the performance of blitting into a form vs.
blitting to Display:
| form saveDepth result |
saveDepth _ Display depth.
form _ Form dotOfSize: 100.
result _ #(8 16 32) collect:
[:depth | Display newDepth: depth.
World notNil ifTrue: [World doOneCycleNow].
depth -> ({Form extent: 300 at 300 depth: depth. Display} collect:
[:dest | Time millisecondsToRun:
[150 to: 200 do:
[:y | 150 to: 200 do:
[:x | form displayOn: dest at: x at y]]]])].
Display newDepth: saveDepth.
^result
On my machine this gives
(8->(2560 5287 ) 16->(3588 6298 ) 32->(5770 9257 ) )
which means that X blitting is roughly as fast as Squeak internal blit.
[The first number is the time for a SqForm-SqForm blit, the second for
SqForm-SqDisplay-Xwindow]
On my friend's machine the result is
(8->(3615 13374 ) 16->(4975 15597 ) 32->(7692 15778 ) )
but under NT with defered update disabled
(8->(3245 5648 ) 16->(4566 7812 ) 32->(7280 18967 ) )
which is a huge difference.
I'm not yet sure what this all means, but perhaps some of you could send
me (privately) your results, and I could compile a list of "Recommended
Linux Squeak Graphics Cards". You may want to experiment with different
color depths (startx -- -bpp <n>) and include the output of xdpyinfo.
/bert
More information about the Squeak-dev
mailing list
|