Hill all, I'm now testing some benchmarks on qemu.
I am not sure the time measured by "time" command in linux over qemu is precise, since that qemu is an emulator. So, alternatively, I have tried ssh to a native linux machine to get the correct time. However, ssh provide some overhead, and this overhead varies every time, the variance is a large amount in my testbenches. Is there anything else can help me get more precise execution time? Here are some of my experiment results: Remote(s) Local(s) Benchmark 3.86 3.55 automotive/basicmath 0.65 0.45 automotive/bitcount 0.68 0.48 automotive/qsort 0.73 0.40 automotive/susan 4.60 4.39 telecomm/CRC32 2.93 2.72 telecomm/FFT 3.20 2.96 telecomm/adpcm 2.03 1.84 telecomm/gsm 0.20 0.01 office/stringsearch 3.34 3.11 office/ispell 5.88 5.66 office/rsynth 1.37 0.96 office/ghostscript 0.51 0.30 network/dijkstra 1.27 0.85 network/patricia 0.47 0.17 consumer/jpeg 1.11 0.84 consumer/typeset 0.69 0.45 consumer/tiff2bw 1.32 1.10 consumer/tiff2rgba 1.12 0.93 consumer/tiffdither 1.15 0.95 consumer/tiffmedian 0.19 0.00 consumer/lame 0.19 0.00 consumer/mad Thanks. Zack