Dealing with the Performance and Power Budgeting Challenges of Datacenters

Опубликовано 27 июня 2016, 18:52

Energy consumption is a major challenge for datacenters as it represents one of the larger components of ownership costs. In this talk I will present techniques to maximize the performance of datacenters within dynamic power budgets. I will first describe a technique that enables datacenters to partition the total power budget among the cooling and computing infrastructure in a self-consistent way, where the cooling power is sufficient to extract the heat of the computing power. To allocate the computing power among the servers such that the performance is maximized, I will then describe a number of optimal techniques including: (1) dynamic programming methods for servers equipped with power cap controllers, (2) matching-based algorithms for datacenters with heterogenous server makeup, and (3) linear programming techniques for workload consolidation and virtual machine migration. Using traces of execution from real servers, our techniques are integrated into a workload management simulator to assess the impact on performance and power consumption. The effectiveness of the proposed techniques as a function of the datacenter load and configuration will be illustrated. Significant improvements over state-of-the-art methods will be demonstrated.