Information Technology Reference
In-Depth Information
due to the complex instruction set, the advanced branch prediction schemes and the
larger caches of the processors. Therefore, even in the case that the embedded
processor has longer execution time than the GPPs the overall energy that it consumes
is much lower than the GPP. Therefore, in data centers which require energy efficient
servers such as the microservers [9], embedded systems could be utilized efficiently
reducing the overall power consumption. Furthermore, as the most cloud applications
than are based on MapReduce framework are designed to run in parallel systems, the
servers could even achieve the same performance in terms of throughput by
replicating more embedded system cores but consuming much lower energy.
Normalized Energy Consumption
HP-GPP
LP-GPP
EP
9
Cloud Applications
8
7
6
5
4
3
2
1
0
Histogram
Linear_regr
String match
Word_count
Matrix_mult
Application
Fig. 6.
Normalized Energy Consumption for different applications
4
Conclusions
In this paper we evaluate high performance embedded processors in the domain of
cloud computing. We map typical cloud computing application in the ARM Cortex
A9-MPCore cores and we compare it with high performance and low power general
purpose processors. The performance evaluation shows that the execution time of the
embedded processors is up 5x higher than the general purpose processors in tasks
common in the cloud applications (word count, string match, etc.). However, the
power consumption of the embedded processors is significantly lower the general
purpose processors. Therefore high performance embedded processors can achieve up
to 7.8x better energy efficiency in cloud computing applications, compared with the
general purpose processors and they could be a viable alternative in data centers with
lower energy consumption requirements such as microservers. These embedded
processors could also be a promising alternative to any other cloud computing
applications that can tolerate a small increase in the overall execution time but
consuming much lower energy and thus reducing the operating cost of these data
centers.
