Investigating power efficiency of mergesort

Abstract

Excessive power consumption emerged as a major obstacle to achieving exascale performance in next-generation supercomputers, creating a need to explore new ways to reduce those requirements. In this study, we present a comprehensive empirical investigation of a power advantage anticipated in the mergesort method based on identifying a feature expected to be physically power efficient. We use a highperformance quicksort as a realistic baseline to compare. Results show a generic mergosort to have a distinct advantage over an optimized quicksort lending support to our expectation. They also help develop some insights toward power efficiency gains likely meaningful in a future exascale context where trading some of the abundant performance for much needed power savings in a ubiquitous computation may prove interesting.

Exploring Power Advantage of Binary Search: An Experimental Study

“…The authors chose 300 after running test trials and found that averages tended to start converging after 230 runs. Dataset sizes were chosen based on cache miss data from [11], where they used the same CPU. Some experimentation with different sizes confirmed that power readings for datasets located in the same cache level were reasonably similar.…”

“…It encourages rethinking time optimization, which has long been a classic interest in computing, as it may sometimes interfere with the power concerns of systems where saving power may be of primary interest. With those concerns in mind, [11] examined the power efficiency of mergesort against that of quicksort, the standard general-purpose sorting algorithm. Their results showed that the power consumption of the mergesort was significantly better than an optimized quicksort when it exploited the barrel shifter, a digital binary shift circuit, to do the partitioning phase.…”

“…In [11], the author developed a rigorous experimental procedure to eliminate sources that could taint measurements based on a CPU with improved power instrumentation. They showed natural mergesort to have a clear power advantage against a highly optimized 3-way quicksort.…”

“…They showed natural mergesort to have a clear power advantage against a highly optimized 3-way quicksort. A meta-analytical comparison of the energy consumption of mergesort and quicksort was conducted in [33] based on an extensive literature review that included [11]. The study concluded that there was no significant difference in energy needs.…”

“…The study concluded that there was no significant difference in energy needs. Results in [11] had earlier agreed on energy but found the mergesort to consume less power based on direct measurement, stressing a concern when two systems go through the same energy budget at different rates. It is the basis of time-power tradeoffs.…”