| Popis: |
This chapter discusses two methods that can potentially save megawatts of power on leadership class systems and provide significant reductions in power consumption on small clusters over the long term. The chapter first examines how a developer can make simple nonalgorithmic application changes, such as thread distribution, in concert with power measurements, which can significantly change the application’s performance per watt. The chapter also takes a look at how administrators and developers can use big data techniques like waterfall plots to examine profiling information across cluster environments, and their potential impact on data center performance. |
