CSR: Small: Joint Power Optimization of Data Center Network and Servers with Correlation Analysis and Scalability

电源优化已成为当今数据中心设计中的一个关键挑战。最近的许多研究表明，数据中心中通常有三个主要的电力消费者：服务器、冷却系统和数据中心网络（DCN）。虽然近年来数据中心冷却的功率效率有了显著提高，但可以预见的是，服务器和DCN将成为未来最重要的两个功率消费者。不幸的是，虽然现有的研究工作主要集中在计算机服务器上，以降低其功耗，但只有很少的研究试图解决DCN的功耗，它可能占数据中心总功耗的20%左右。本项目旨在设计一个关联感知的功耗优化框架，共同实现数据中心DCN和服务器总功耗的最小化。这个及时的项目的成功可以通过显著降低DCN功耗来极大地影响数据中心的设计。功率优化框架的主要技术手段是关联感知服务器和流量整合。与服务器类似，DCN也经常未得到充分利用。因此，流量可以整合到一小组链路和交换机上，这样可以关闭未使用的网络设备以节省电力。服务器合并和流量合并应同时进行，因为不考虑DCN的服务器合并可能导致流量拥塞，从而降低网络性能。另一方面，服务器整合可能会改变DCN拓扑，从而为节省电力提供新的机会。该框架的设计基于一个关键观察，即不同服务器的利用率或不同流的带宽需求通常不会在完全相同的时间达到峰值。因此，如果考虑服务器和流量流之间的相关性，则可以在服务器和流量合并期间实现更多的节能。功率优化框架还包括具有流完成时间保证的多维DCN功率优化和适用于大型数据中心的高度可扩展优化算法。

Power optimization has become a key challenge in the design of today's data centers. Many recent studies have shown that there are typically three major power consumers in a data center: servers, cooling systems, and the data center network (DCN). While the power efficiency of data center cooling has been significantly improved in the recent years, it is foreseeable that servers and DCN are becoming the two most significant power consumers in the future. Unfortunately, while existing research efforts focus mainly on computer servers to lower their power consumption, only few studies have tried to address the power consumption of DCN, which can account for about 20% of the total power consumption of a data center. This project aims to design a correlation-aware power optimization framework that jointly minimizes the total power consumption of the DCN and servers in a data center. The success of this timely project can greatly impact the data center design by significantly reducing DCN power consumption.The main technical approach of the power optimization framework is correlation-aware server and traffic consolidations. Similar to servers, a DCN is also often underutilized. As a result, traffic flows can be consolidated onto a small set of links and switches, such that unused network devices can be shut down for power savings. Server and traffic consolidations should be conducted jointly because server consolidation without considering the DCN may cause traffic congestion and thus degraded network performance. On the other hand, server consolidation may change the DCN topology, allowing new opportunities for power savings. This framework is designed based on a key observation that the utilizations of different servers or the bandwidth demands of different flows usually do not peak at exactly the same time. Therefore, if the correlations among servers and traffic flows are considered, more power savings can be achieved during server and traffic consolidations. The power optimization framework also includes multi-dimensional DCN power optimization with flow completion time guarantees and highly scalable optimization algorithms for large-scale data centers.