Information Technology Reference
In-Depth Information
Fig. 4. Execution times in minutes for each program using 4 processors
Ta b l e 5 . Execution times in minutes and speedup for each program using 8 processors
0%
10%
20%
30%
40%
50%
Time
SU
Time
SU
Time
SU
Time
SU
Time
SU
Time
SU
P1
0 . 002 1 . 00 0 . 308 1 . 00 7 . 844 1 . 00 62 . 695 1 . 00 203 . 416 1 . 00 296 . 691 1 . 00
P2
0 . 003 0 . 67 0 . 193 1 . 59 4 . 610 1 . 70 26 . 578 2 . 36
78 . 644 2 . 58 117 . 415 2 . 53
P3
0 . 001 2 . 00 0 . 067 4 . 60 1 . 767 4 . 44 13 . 634 4 . 60
45 . 213 4 . 50
62 . 572 4 . 74
P4
0 . 008 0 . 25 0 . 047 6 . 55 1 . 155 6 . 79
9 . 310 6 . 73
32 . 142 6 . 33
54 . 165 5 . 48
P5
0 . 002 1 . 00 0 . 071 4 . 34 1 . 739 4 . 51 15 . 125 4 . 14
45 . 858 4 . 43
62 . 572 4 . 74
P6
0 . 005 0 . 40 0 . 055 5 . 60 1 . 258 6 . 23 10 . 158 6 . 17
35 . 275 5 . 76
54 . 006 5 . 49
P7
0 . 008 0 . 25 0 . 054 5 . 70 1 . 232 6 . 37 10 . 041 6 . 24
34 . 800 5 . 84
50 . 402 5 . 89
P8
0 . 007 0 . 28 0 . 051 6 . 04 1 . 248 6 . 28 10 . 143 6 . 18
35 . 376 5 . 75
53 . 031 5 . 59
Finally we analyze what happens if we use 8 processors to solve the problem. Once
more, we obtain for the parallelized Program 7 the least computation time, 50 . 402 min-
utes with a 50% CCA penetration rate. So if we compare this result with the execution
time of the sequential program we obtain an improvement of the 83% , that is, a speedup
factor of 5 . 89 .
6.4
Results Discussion
In conclusion, on the one hand, we have achieved an improvement of 83% in the compu-
tation time of the most complex case, what can be considered as a pretty much outstand-
ing improvement. On the other hand, if we compare the best execution times between
the two technical extremes under study, that is the use of 2 or 8 processors belonging
to the shared nodes architecture in the Arabi cluster, we reach to an improvement of
Search WWH ::




Custom Search