Information Technology Reference
In-Depth Information
3.2
Energy Efficiency
In this section we evaluate the energy efficiency of the embedded processors
compared to the general purpose processors. The energy efficiency is measured by the
product of the power consumption by the total execution time of a specific task [15].
Power
Energy
=
ExecTime
⋅
For the ARM processor we measure the power consumption of the processors using
the
Pandaboard
[16] which integrates the OMAP4430 chip with the ARM processors
and the DRAM memory. The power measurements are based on the current that is
drawn by the ARM processors [17]. On the Intel processors we measure the power
consumption using the
powerstat
application [18]. In all cases the CPU utilization is
above 80% for the cloud computing applications which means that all processors
consume almost the maximum power consumption (Figure 5). Again this figure
shows the difference between the matrix-multiplication applications with the typical
cloud computing applications. In the case of
matrix multiplication
the processor is fast
enough to perform the tasks and the lower utilization is due to the system calls.
CPU Utilization of cloud computing applications
100
90
80
70
60
Histogram
Linear_reg
String_match
Word_count
Matrix_Mult
50
40
30
20
10
0
1
2
3
4
5
6
7
8
9
10 11 12 13 14 15 16 17 18 19 20
Seconds
Fig. 5.
CPU Utilization of the applications in time
Figure 6 depicts the normalized energy consumption of the HP-GPP, the LP-GPP
and the embedded processor for different applications of the Phoenix MapReduce
framework. The figure shows the normalized energy based on the energy
consumption of the embedded processor. As is it shown in this figure the embedded
processors can achieve up to 7.8x lower
energy
consumption compared with the HP-
GPP. This is due to the fact that the
power
consumption of the embedded processor is
much lower than the power of the GPP. The high power consumption of the GPP is
