Power routing: dynamic power provisioning in the data center
Power routing: dynamic power provisioning in the data center
复制标题
DOI：
10.1145/1736020.1736047
复制
发表时间：
2010-03
期刊：
影响因子：
--
通讯作者：
Steven Pelley;David Meisner;Pooya Zandevakili;T. Wenisch;Jack Underwood
中科院分区：
文献类型：
--
作者：
Steven Pelley;David Meisner;Pooya Zandevakili;T. Wenisch;Jack Underwood
数据中心电力基础设施会产生巨大的资本成本，通常超过设施寿命期间的能源成本。为了从基础设施中获得最大价值，研究人员提出了超额订购电力电路，这依赖于峰值负载很少的观察结果。为了确保可用性，这些建议采用功率封顶，在利用率高峰期间限制服务器性能，以实施安全的功率预算。然而，由于预算必须在本地实施-在每个配电单元（PDU）处-即使在其他地方有电力输送容量时，本地利用率峰值也可能强制节流。此外，需要在电力输送路径上保持用于容错的储备容量，这放大了利用率尖峰的影响。在本文中，我们开发的机制，以更好地利用已安装的电力基础设施，减少备用容量的利润率和避免性能节流。与传统的高可用性数据中心不同，在传统的高可用性数据中心中，并置的服务器共享相同的主电源和辅助电源，我们重新组织电源，以创建混合的配电拓扑。混洗拓扑将二次电源馈电分布在多个PDU上，从而降低了容忍单个PDU故障所需的备用容量。其次，我们提出了电源路由，它在冗余电源之间动态调度IT负载，以：（1）将松弛转移到电力需求不断增长的服务器，以及（2）平衡交流相位之间的电力消耗，以减少发热并提高电气稳定性。我们描述了高效的调度服务器的PDU（NP完全问题）。使用从三个生产设施中的近1000台服务器收集的数据，我们证明了这些机制可以将所需的电力基础设施容量相对于传统的高可用性数据中心减少32%，而不会降低性能。
Data center power infrastructure incurs massive capital costs, which typically exceed energy costs over the life of the facility. To squeeze maximum value from the infrastructure, researchers have proposed over-subscribing power circuits, relying on the observation that peak loads are rare. To ensure availability, these proposals employ power capping, which throttles server performance during utilization spikes to enforce safe power budgets. However, because budgets must be enforced locally -- at each power distribution unit (PDU) -- local utilization spikes may force throttling even when power delivery capacity is available elsewhere. Moreover, the need to maintain reserve capacity for fault tolerance on power delivery paths magnifies the impact of utilization spikes. In this paper, we develop mechanisms to better utilize installed power infrastructure, reducing reserve capacity margins and avoiding performance throttling. Unlike conventional high-availability data centers, where collocated servers share identical primary and secondary power feeds, we reorganize power feeds to create shuffled power distribution topologies. Shuffled topologies spread secondary power feeds over numerous PDUs, reducing reserve capacity requirements to tolerate a single PDU failure. Second, we propose Power Routing, which schedules IT load dynamically across redundant power feeds to: (1) shift slack to servers with growing power demands, and (2) balance power draw across AC phases to reduce heating and improve electrical stability. We describe efficient heuristics for scheduling servers to PDUs (an NP-complete problem). Using data collected from nearly 1000 servers in three production facilities, we demonstrate that these mechanisms can reduce the required power infrastructure capacity relative to conventional high-availability data centers by 32% without performance degradation.