[vlc-devel] vlc x246 benchmark
pegasus at nerv.eu.org
Sun Mar 21 19:33:25 CET 2010
After reading http://x264dev.multimedia.cx/?p=185 I deceided to do a simple benchmark.
I have a vlc 1.0.1 running on ubuntu 8.10. I'm timing a transcode of a 20h long dump of a random tv channel here in h264 from tv resolution to 640x480 again in h264, comparing different ubuntu kernels on same machine.
2.6.24-24: 1057min user + 11m11s sys time
2.6.27-17: 823min51s user + 15min20s sys time
2.6.28-18: 1063m3s user + 15m0s sys time
2.6.31-20: 872m5s user + 94m55s sys time
2.6.32-16 kernel from not yet released ubuntu 10.04: 875min35s user + 93min43s sys time.
Based on the above reading I would expect much better result on newer kernel, but what I see does not fit the expectations. Is time(1) a proper way to evaluate efficiency of a transcode? Why do I see such drastic differences bot in user and in system time? Environment, other running processes and hw configuration are the same for all tests, the only thing changing is the kernel.
More information about the vlc-devel