Comparison with Core i7 Based Cluster: The described system delivers higher efficiency per unit of power,
and energy, while reducing execution time in the same
power envelope. Due to increased power efficiency, the
maximum power drawn by the datacenter can be reduced
for the same performance levels and energy consumption
can be reduced. Alternatively, in a fixed power envelope,
performance of the datacenter can be improved by increasing the cluster size and scaling out the applications. Figure 8(a) shows the change in Perf./W of the four benchmarks when executed on the described system compared
to the Core i7 based cluster. The disk and CPU intensive benchmarks TeraGen and GridMix show 75.5% and
46.9% improvementrespectively, while TeraSort, which has
a CPU intensive Map phase, and disk intensive Reduce
phase shows a 147.8% improvement. WordCount, which
stresses neither the CPU, nor the disk due to its small dataset
size shows a degradation of 15.4%. Figure 8(b) shows the
change in performance per unit of energy (Watt-hour) when
comparing the two clusters. The trend is similar to Perf./W
and shows that our system is 2X-6X more energy efficient
than a commodity cluster.