[DCP-o-matic] Benchmarking DCP-o-matic on Raspberry Pi / Raspian

Hi guys, It's time for a first feedback on DCP-o-matic encoding on a slave Raspberry Pi 2 server. My setup: - cheap PC as master (Pentium G3430 - 2 cores @3.30Ghz - 4GB RAM), Passmark 3437, Ubuntu 14.10 + DCP-o-matic - Mikrotik RB750up router, using only 100M ports and closed LAN. - RPi2 with Raspbian Wheezy + custom DCP-o-matic ARM server nogui homebrew (thanks Carl). - Monitoring tools: Htop on the Pi & Ubuntu system monitor on PC BB Bunny asset bench: - PC alone = 5mn15 => 2.62 fr/s - PC + Pi (4 cores) = 4mn28 => 3.08 fr/s - PC + Pi (4 cores) + logging on Pi = 4mn45 - PC + Pi (4 cores) + monitoring = 4mn37 - PC + Pi (3 cores) + monitoring = 4mn44 I did the last test because I saw less than 15Mb/s peaks on network, which made me think that Pi didn't fully handled networking load (like when compiling...). In fact, Bunny frames are small and the bottleneck is on Pi CPU, not networking. Comments: it seems that this setup is performing better than most 4 cores listed systems... http://dcpomatic.com/benchmarks/input.php?id=2 Sintel asset bench: - PC alone = 3h18mn30 => 1.87 fr/s - PC + Pi (4 cores) = 2h51mn29 => 2.17 fr/s On these calculations, I exclude intro and final checksums which last longer than with Bunny asset. Networking never went more than 50Mb/s. So no need for gigabit LAN. RPi2 RAM never top up, staying behind 900M. No need for hudge RAM too on the PC. It stayed below 1G. RPi2 CPU did get really hot while running at 400% during hours, but no burnout, so no need for a heatsink. Power stayed behind 4W. Comments: RPi2 handles 0.30 frame per second, roughly 1 frame every 3 seconds and 4 frames in 12 seconds. This is quite the pattern that I noticed on networking graphes. It looks like adding roughly 1.5 more core on my PC (?). See attached two screen captures of RPi2 htop monitoring and PC monitoring while encoding (sorry, my computers mainly speaks French). I'll send all logs to Carl tomorrow. Regards, Fred