Today I started my new second hand thin client (in mini ITX form), and followed the benchmarking procedure as described in OGN website to compare with modern single board computers.<\/p>\n
My thin client: Fujitsu Futro S550 (CPU AMD Sempron 2100+ 1GHz single core, RAM 1GB), benchmarked with\u00a0Ubuntu 14.04.04 LTS running on\u00a0USB flash drive:<\/p>\n
$ sysbench –num-threads=1<\/b> –test=cpu –cpu-max-prime=2000 run<\/p>\n
Test execution summary:
\ntotal time: 8.9729s<\/b>
\ntotal number of events: 10000
\ntotal time taken by event execution: 8.9659
\nper-request statistics:
\nmin: 0.87ms
\navg: 0.90ms
\nmax: 8.91ms
\napprox. 95 percentile: 0.88ms<\/p>\n
Threads fairness:
\nevents (avg\/stddev): 10000.0000\/0.00
\nexecution time (avg\/stddev): 8.9659\/0.00<\/p>\n
$ sysbench –num-threads=2<\/b> –test=cpu –cpu-max-prime=2000 run<\/p>\n
Test execution summary:
\ntotal time: 8.9830s<\/b>
\ntotal number of events: 10000
\ntotal time taken by event execution: 17.9520
\nper-request statistics:
\nmin: 0.87ms
\navg: 1.80ms
\nmax: 17.78ms
\napprox. 95 percentile: 4.88ms<\/p>\n
Threads fairness:
\nevents (avg\/stddev): 5000.0000\/2.00
\nexecution time (avg\/stddev): 8.9760\/0.00<\/p>\n
$ sysbench –num-threads=4<\/b> –test=cpu –cpu-max-prime=2000 run<\/p>\n
Test execution summary:
\ntotal time: 8.9726s
\ntotal number of events: 10000
\ntotal time taken by event execution: 35.8414
\nper-request statistics:
\nmin: 0.87ms
\navg: 3.58ms
\nmax: 35.99ms
\napprox. 95 percentile: 12.88ms<\/p>\n
Threads fairness:
\nevents (avg\/stddev): 2500.0000\/1.22
\nexecution time (avg\/stddev): 8.9604\/0.00<\/p>\n
$ sysbench –test=memory –memory-block-size=1M<\/b> –memory-total-size=256M<\/b> run<\/p>\n
256.00 MB transferred (771.98 MB\/sec<\/b>)<\/p>\n
Test execution summary:
\ntotal time: 0.3316s
\ntotal number of events: 256
\ntotal time taken by event execution: 0.3303
\nper-request statistics:
\nmin: 1.10ms
\navg: 1.29ms
\nmax: 9.15ms
\napprox. 95 percentile: 1.74ms<\/p>\n
Threads fairness:
\nevents (avg\/stddev): 256.0000\/0.00
\nexecution time (avg\/stddev): 0.3303\/0.00<\/p>\n
$ sysbench –test=memory –memory-block-size=512K<\/b> –memory-total-size=512M<\/b> run<\/p>\n
512.00 MB transferred (799.95 MB\/sec<\/b>)<\/p>\n
Test execution summary:
\ntotal time: 0.6400s
\ntotal number of events: 1024
\ntotal time taken by event execution: 0.6375
\nper-request statistics:
\nmin: 0.53ms
\navg: 0.62ms
\nmax: 8.86ms
\napprox. 95 percentile: 0.60ms<\/p>\n
Threads fairness:
\nevents (avg\/stddev): 1024.0000\/0.00
\nexecution time (avg\/stddev): 0.6375\/0.00<\/p>\n
—<\/p>\n
In comparison with embedded SBCs, this is equivalent to Raspberry Pi 2 running on 4 cores, or Raspberry Pi 3 running with 2 cores, and about 3 times faster than BeagleBone Black. If a software uses only single core, then Futro S550 is faster than RPi 3, however if a software maximize the use of 4 cores, RPi 3 is faster than Futro S550. The SBCs’ performance was reported in:<\/p>\n