Software Environments - Heterogeneous Multicore Processor Technologies for Embedded Systems - page 157

Table 5.10 Software version

Software Version number

Linux 2.6.16

glibc 2.3.5

5000

4500

4000

3500

3000

2500

4cores
3cores
2cores
1core

2000

1500

1000

500

0

0 500 1000 1500 2000 2500

Total Frequency (CPU cores) [MHz]

Fig. 5.2 Power consumption with CPU hot-plug and CPUfreq control

the total energy consumption of idle reduction with application loads. The Linux
version on which we implemented idle reduction is given in Table 5.10 :

1. Power consumption of CPU hot-plug and CPUfreq.
The total power consumption of up to four CPU cores of the RP-2 using CPUfreq
and CPU hot add/remove is shown in Fig. 5.2 . Each CPU core executes the Dhrystone
2.1 program at 75-MHz, 150-MHz, 300-MHz, or 600-MHz frequency controlled by
CPUfreq, or is stopped and powered off using CPU hot remove.
This figure shows that the total frequency that can be translated to total instructions
per second (IPS) is the same, but the power consumption is different. For example, if
each core frequency is set to 600, 300, 150, and 150 MHz, and the total is 1,200 MHz,
the power consumption is about 3.2 W where the chip voltage is 1.4 V. In another
case, all four core frequencies are set to 300 MHz and the total is 1,200 MHz; the
power consumption is about 1.9 W where the chip voltage is 1.2 V [ 6 ] .

2. Energy consumption of idle reduction.
We evaluated the energy consumption of idle reduction using a multi-thread
benchmark (Splash2 RAYTRACE) [ 7 ]. The evaluation was done by comparing it
with the old governors.
Table 5.11 and Fig. 5.3 show the energy consumption with no load (all cores are
in an idle state). The table presents a comparison of 10 s of energy consumption of
each governor. Idle reduction clearly had the lowest energy consumption, and