Beowulf Aug2010/Installation/Test Applicationer
From Teknologisk videncenter
<accesscontrol>teacher</accesscontrol>
Brug 100% CPU
Skal måske startes flere gange for at bruge begge kerne.
#!/bin/bash
while : ; do
true
done
Teste MPI båndbrede
Testen er lavet på vores Dell PowerEdge 1750 med onboard netkort i en 3550 100mbit/s switch
Hello from 0 of 2
Timer accuracy of ~0.953674 usecs
Hello from 1 of 2
8 bytes took 331 usec ( 0.048 MB/sec)
16 bytes took 206 usec ( 0.155 MB/sec)
32 bytes took 213 usec ( 0.301 MB/sec)
64 bytes took 229 usec ( 0.559 MB/sec)
128 bytes took 266 usec ( 0.962 MB/sec)
256 bytes took 334 usec ( 1.533 MB/sec)
512 bytes took 472 usec ( 2.170 MB/sec)
1024 bytes took 758 usec ( 2.701 MB/sec)
2048 bytes took 1112 usec ( 3.683 MB/sec)
4096 bytes took 1402 usec ( 5.843 MB/sec)
8192 bytes took 2138 usec ( 7.664 MB/sec)
16384 bytes took 3555 usec ( 9.218 MB/sec)
32768 bytes took 6403 usec ( 10.235 MB/sec)
65536 bytes took 12508 usec ( 10.479 MB/sec)
131072 bytes took 23635 usec ( 11.091 MB/sec)
262144 bytes took 46029 usec ( 11.390 MB/sec)
524288 bytes took 90590 usec ( 11.575 MB/sec)
1048576 bytes took 179719 usec ( 11.669 MB/sec)
Asynchronous ping-pong
8 bytes took 212 usec ( 0.075 MB/sec)
16 bytes took 205 usec ( 0.156 MB/sec)
32 bytes took 211 usec ( 0.303 MB/sec)
64 bytes took 237 usec ( 0.540 MB/sec)
128 bytes took 264 usec ( 0.969 MB/sec)
256 bytes took 336 usec ( 1.524 MB/sec)
512 bytes took 470 usec ( 2.179 MB/sec)
1024 bytes took 753 usec ( 2.720 MB/sec)
2048 bytes took 1076 usec ( 3.807 MB/sec)
4096 bytes took 1407 usec ( 5.822 MB/sec)
8192 bytes took 2141 usec ( 7.653 MB/sec)
16384 bytes took 3661 usec ( 8.950 MB/sec)
32768 bytes took 6428 usec ( 10.195 MB/sec)
65536 bytes took 12507 usec ( 10.480 MB/sec)
131072 bytes took 23708 usec ( 11.057 MB/sec)
262144 bytes took 46073 usec ( 11.380 MB/sec)
524288 bytes took 90696 usec ( 11.561 MB/sec)
1048576 bytes took 179756 usec ( 11.667 MB/sec)
Bi-directional asynchronous ping-pong
8 bytes took 197 usec ( 0.081 MB/sec)
16 bytes took 189 usec ( 0.169 MB/sec)
32 bytes took 196 usec ( 0.327 MB/sec)
64 bytes took 200 usec ( 0.640 MB/sec)
128 bytes took 230 usec ( 1.114 MB/sec)
256 bytes took 324 usec ( 1.580 MB/sec)
512 bytes took 454 usec ( 2.256 MB/sec)
1024 bytes took 735 usec ( 2.786 MB/sec)
2048 bytes took 1075 usec ( 3.810 MB/sec)
4096 bytes took 1421 usec ( 5.765 MB/sec)
8192 bytes took 2187 usec ( 7.491 MB/sec)
16384 bytes took 3630 usec ( 9.027 MB/sec)
32768 bytes took 6534 usec ( 10.030 MB/sec)
65536 bytes took 13344 usec ( 9.823 MB/sec)
131072 bytes took 25400 usec ( 10.321 MB/sec)
262144 bytes took 49689 usec ( 10.551 MB/sec)
524288 bytes took 124407 usec ( 8.429 MB/sec)
1048576 bytes took 261184 usec ( 8.029 MB/sec)
Max rate = 11.669048 MB/sec Min latency = 94.413757 usec
Og de adskiller sig jo ikke det store fra dem i Test Clusteret
Her er de så hvis man bruger Intel netkortet igennem en 100mbit/s switch:
Hello from 0 of 2
Timer accuracy of ~0.953674 usecs
8 bytes took 504 usec ( 0.032 MB/sec)
16 bytes took 266 usec ( 0.120 MB/sec)
32 bytes took 199 usec ( 0.322 MB/sec)
Hello from 1 of 2
64 bytes took 262 usec ( 0.489 MB/sec)
128 bytes took 262 usec ( 0.977 MB/sec)
256 bytes took 491 usec ( 1.042 MB/sec)
512 bytes took 460 usec ( 2.227 MB/sec)
1024 bytes took 1012 usec ( 2.024 MB/sec)
2048 bytes took 1084 usec ( 3.778 MB/sec)
4096 bytes took 1780 usec ( 4.602 MB/sec)
8192 bytes took 2179 usec ( 7.519 MB/sec)
16384 bytes took 3735 usec ( 8.774 MB/sec)
32768 bytes took 6395 usec ( 10.248 MB/sec)
65536 bytes took 12623 usec ( 10.384 MB/sec)
131072 bytes took 13000 usec ( 20.165 MB/sec)
262144 bytes took 24516 usec ( 21.386 MB/sec)
524288 bytes took 47452 usec ( 22.098 MB/sec)
1048576 bytes took 91873 usec ( 22.827 MB/sec)
Asynchronous ping-pong
8 bytes took 434 usec ( 0.037 MB/sec)
16 bytes took 189 usec ( 0.169 MB/sec)
32 bytes took 234 usec ( 0.273 MB/sec)
64 bytes took 221 usec ( 0.579 MB/sec)
128 bytes took 482 usec ( 0.531 MB/sec)
256 bytes took 325 usec ( 1.576 MB/sec)
512 bytes took 714 usec ( 1.435 MB/sec)
1024 bytes took 743 usec ( 2.756 MB/sec)
2048 bytes took 1255 usec ( 3.264 MB/sec)
4096 bytes took 1419 usec ( 5.773 MB/sec)
8192 bytes took 2247 usec ( 7.292 MB/sec)
16384 bytes took 3574 usec ( 9.168 MB/sec)
32768 bytes took 6691 usec ( 9.794 MB/sec)
65536 bytes took 13006 usec ( 10.078 MB/sec)
131072 bytes took 13254 usec ( 19.779 MB/sec)
262144 bytes took 24478 usec ( 21.419 MB/sec)
524288 bytes took 47442 usec ( 22.102 MB/sec)
1048576 bytes took 91929 usec ( 22.813 MB/sec)
Bi-directional asynchronous ping-pong
8 bytes took 490 usec ( 0.033 MB/sec)
16 bytes took 247 usec ( 0.130 MB/sec)
32 bytes took 248 usec ( 0.258 MB/sec)
64 bytes took 247 usec ( 0.518 MB/sec)
128 bytes took 248 usec ( 1.032 MB/sec)
256 bytes took 498 usec ( 1.028 MB/sec)
512 bytes took 528 usec ( 1.939 MB/sec)
1024 bytes took 748 usec ( 2.738 MB/sec)
2048 bytes took 1261 usec ( 3.248 MB/sec)
4096 bytes took 1521 usec ( 5.386 MB/sec)
8192 bytes took 2270 usec ( 7.218 MB/sec)
16384 bytes took 3785 usec ( 8.658 MB/sec)
32768 bytes took 6564 usec ( 9.984 MB/sec)
65536 bytes took 13499 usec ( 9.710 MB/sec)
131072 bytes took 19491 usec ( 13.450 MB/sec)
262144 bytes took 36962 usec ( 14.185 MB/sec)
524288 bytes took 71936 usec ( 14.577 MB/sec)
1048576 bytes took 160475 usec ( 13.068 MB/sec)
Max rate = 22.826658 MB/sec Min latency = 94.413757 usec
Hvordan den kan trække 22MB igennem et 100Mbit link går over min forstand...
Men Latency er 2x større på Intel end på Onboard 3Com.