[MEncoder-users] Optimizing Mencoder on Ubuntu vs FreeBSD
Corey Hickey
bugfood-ml at fatooh.org
Wed Apr 12 06:57:10 CEST 2006
Andrew Edmond wrote:
> All --
>
> I previously ran mencoder on several FreeBSD 5.4 systems, which ran
> wonderfully well. We were doing a lot of processing of large videos into
> XVID and DIVX, and 2 hour long files were taking around 10 hours to run a
> single pass plus two bit rate second passes (700Kbps and 1500Kbps).
>
> We recently changed to Ubuntu Linux to support the new On2 Flix Engine for
> Linux (which also runs mencoder), and though the Flix Engine works great,
> the speed on Linux for the XVID encodes works TERRIBLE. The encodes are
> taking now 24 hours instead of 10, a 140% time increase.
> I've upgraded to the Ubuntu Kernel for 686 and SMP (2.6.12), and that didn't
> speed anything up at all.
>
> The machines we are running on are 512MB Pentium IV (Compiled for x86 CPU
> with extensions: MMX MMX2 SSE SSE2), and 10K SATA hard drives on stripe for
> speed. These are the same machines that were running FreeBSD 5.4 running
> mencoder, so I know it's not a hardware issue, but some Linux optimization
> issue I'm missing.
>
> Anybody ever been through this problem before, or am I missing the "optimize
> mencoder for Linux" web page somewhere?
A few thoughts; maybe somebody else can help more:
1. If you still have the BSD available, can you try some other
benchmarks? Such as: (a) bzip2 on the same file (b) mencoder with lavc
and/or x264 instead of xvid.
2. Do you have CPU frequency scaling stuck on the "powersave" governor
or something? Is 'cat /proc/cpuinfo' what you expect?
3. Are you sure you're running exactly the same mencoder and xvid versions?
4. Do you have an assembler installed? There was a recent thread in
which someone pointed out that xvid doesn't use any assembly code if you
don't have an assembler. I haven't experienced that myself, but run
xvid's configure and check for this line:
checking for whether to use assembly code... yes
-Corey
More information about the MEncoder-users
mailing list