Information Technology Reference
In-Depth Information
3
Performance Evaluation
In this section we evaluate the Phoenix MapReduce framework in terms of
performance and energy efficiency. The Phoenix MapReduce framework has been
mapped to three different processors. The first processor is based on a high
performance general purpose processor (HP-GPP: Intel i7-2600). This processor has 4
cores and the clock speed is 3.4GHz. The second processor is based on a low power
general purpose processor (LP-GPP: Intel U5400 processor) with 2 cores and
maximum clock frequency 1.2GHz. The third processor is an embedded system
processor that is based on the OMAP4430 SoC with 2 ARM Cortex A9 cores [14].
The detailed characteristics of the processors are shown in the following table.
Table 1.
Processor architecture characteristics
HP-GPP
LP-GPP
EP
i7-2600
U5400
OMAP4430
Processor
# of Cores
4
2
2
Intel i7
Intel Pentium
ARM Cortex A9
Cores
Process
32nm
32nm
45nm
3.4GHz
1.2GHz
1GHz
Frequency
ISA
CISC
CISC
RISC
64KB (I),64KB (D)
64KB (I),64KB (D)
32KB (I),32KB (D)
L1 Cache
L2 Cache
256KB per core
256KB per core
1MB (shared)
8MB
3MB
-
L3 Cache
Instruction
Set
64-bits
64-bits
32-bits
YES
YES
YES
Integrated
Graphics
As it is shown in the table the high performance processor has larger cache size
and higher clock frequency while the other two processors that are optimized for low
power consumption have much smaller caches and lower clock frequencies. The main
difference is that the first two processors are based on the Pentium x86 CISC
instruction sets while the OMAP4430 processor is based on the ARMv7 RISC
instruction set and is optimized for embedded systems applications. Furthermore, the
Intel processors are 64-bits wide while the ARM cores are based on 32-bits. In both
cases the Phoenix MapReduce framework was hosted on the same operating systems
(Ubuntu Linux 11.10).
3.1
Performance Evaluation
Figure 3 depicts the performance evaluation of 4 different common cloud tasks using
the Phoenix MapReduce framework in terms of execution time. Besides the 4
common cloud tasks it shows also the execution time of a typical benchmark (matrix
multiplication) in order to show the differences between the cloud tasks and other
common tasks used in benchmarking. The figure shows the normalized execution
