runtime, s latency, us median -1sigma +1sigma -3sigma -1sigma +1sigma +3sigma bs400 performance: make -j1 64.96 0.16 0.11 47 8 5 23 make -j3 23.95 0.12 0.28 41 10 7 35 make -j6 14.01 0.03 0.04 25 9 29 7100 make -j9 14.27 0.10 0.02 25 8 25 7200 make -j12 14.33 0.05 0.13 28 8 43 11000 x264 medium 13.88 0.02 0.01 10 4 6 70 x264 ultrafast 1.459 0.003 0.004 44 12 42 66 cfs performance: make -j1 67.39 0.47 0.32 40 13 9 230 make -j3 24.76 0.15 0.25 33 11 12 1400 make -j6 14.16 0.07 0.05 9900 9000 2300 7900 make -j9 14.24 0.13 0.05 10000 9400 3900 9500 make -j12 14.35 0.10 0.14 9600 9000 4500 14000 x264 medium 13.87 0.03 0.03 840 340 1900 8000 x264 ultrafast 1.649 0.014 0.013 130 58 870 4200 bfs400 ondemand: make -j1 72.50 0.55 0.39 90 14 7 43 make -j3 26.63 0.09 0.17 85 53 12 54 make -j6 14.96 0.07 0.11 24 8 64 6600 make -j9 15.15 0.08 0.09 22 8 65 7200 make -j12 15.46 0.19 0.08 27 9 69 11000 x264 medium 14.19 0.03 0.08 11 5 8 130 x264 ultrafast 1.585 0.035 0.011 30 14 40 84 cfs ondemand: make j1 89.22 0.58 0.35 94 19 14 480 make j3 30.22 0.25 0.20 66 32 46 1600 make j6 15.87 0.11 0.33 7900 6500 5100 11000 make j9 15.89 0.08 0.12 12000 9400 3000 14000 make j12 16.08 0.17 0.31 11000 8100 4800 18000 x264 medium 14.06 0.07 0.04 960 561 1900 8300 x264 ultrafast 1.719 0.034 0.030 200 66 900 3700 [1] runtime results were collected with 'perf stat' program over 13 runs. +/- 1sigma values are are 2nd and 12th best result respectively. For normal distribution, these quantiles would be roughly equal to plus and minus one sigma, and about 70% of samples would fall into this interval. [2] latency results were computed with modified wakup-latency program which recorded latencies between expected and actual wakeup time. Resulting data were normalized to their median and then data over all 13 runs were combined. Reported latency values are absolute values of p=0.5 * (1 +/- 0.683) and p=0.5 * (1 +/- 0.997) quantiles of the combined normalized distributions. [3] hardware configuration: CPU: 6-core 3.3GHz AMD CPU with 6MB of L3, turbo mode enabled. RAM: 8GB ECC DDR2 800MHz PC2-6400 CL5 2x64bit channel unganged. I/O: none, all tests were done in tmpfs. [4] software configuration: Base kernel version for patches is 2.6.35.12, options PREEMPT=y HZ=1000 were used. System is 64-bit debian sid as of 2011-04-08 with special init scripts for the benchmark run, resulting in a completely idle system. [5] benchmark configuration: make -jN was used to build vanilla allnoconfig 2.6.35.12 kernel with gcc-3.4 x264 0.114.1924M 08d04a4 was used with standard input file park_joy_420_720p.y4m.