2007/8/21, David Liontooth <[EMAIL PROTECTED]>:
> What was the effect when you ran transcode in cluster mode on nodes of
> different speeds and loads?
My "distribute" program keeps one process active on each one. 100%cpu
load on each. The only exception is at the end of the first pass and
the end of the second pass if the last block happens to be running on
one of the slower nodes, but I can live with that.
> Is compiling transcode under gcc 3.2 worth doing, or should I install a
> local version in my user?
>
> I'd like to use it for x264, let me know if you've used this codec
> within transcode or separately.
I used xvid, and found a bug because it was outputting N-1 frames per
block, so the picture very gradually got ahead of the sound, up to
about 2.5 seconds by the end of the 60th block. That has been fixed in
the 1.0.4rc5 and in the 1.1 CVS head, and it works fine now.
I haven't used x264 but if you want to buy a few hours of my time I'd
be glad to run the same cluster test piece for you with a different
encoder. :)
I wouldn't have thought the compiler version would make *that* much
difference to the speed, given all the FP math that is required.
That said, a friend's (mostly integer math) language interpreter
doubled in speed from one release to the next, and the only reason he
could think of was that gcc had changed version in between. I'd be
interested to have hard figures on that.
M