Abstract—Power consumption imposes a significant cost for data centers implementing cloud services, yet much of that power is used to maintain excess service capacity during periods of predictably low load. This paper investigates how much can be saved by dynamically ‘right-sizing ’ the data center by turning off servers during such periods, and how to achieve that saving via an online algorithm. We prove that the optimal offline algorithm for dynamic right-sizing has a simple structure when viewed in reverse time, and this structure is exploited to develop a new ‘lazy ’ online algorithm, which is proven to be 3-competitive. We validate the algorithm using traces from two real data center workloads and show that significant cost-savings are possible. I.

Abstract—Power consumption imposes a significant cost for implementing cloud services, yet much of that power is used to maintain excess service capacity during periods of low load. In this work, we study how to avoid such waste via an online dynamic capacity provisioning. We overview recent results showing that the optimal offline algorithm for dynamic capacity provisioning has a simple structure when viewed in reverse time, and this structure can be exploited to develop a new ‘lazy ’ online algorithm which is 3-competitive. Additionally, we analyze the performance of the more traditional approach of receding horizon control and introduce a new variant with a significantly improved worst-case performance guarantee. I.

My research interests are in the design, analysis, and implementation of algorithms for problems in networks, distributed computing, and data mining. In terms of techniques, I am interested in competitive analysis, optimization, and game theory. In particular, my current research focuses on designing self-tuning algorithms for distributed resource allocation. In the traditional “client-server ” paradigm, an end-user, or client, interacts with a network service by opening a connection to a particular machine, or server, running the service. Recently, it is more typical for a collection of services to execute on a shared collection of distributed resources. For example, a collection of services is executed on machines spread across multiple data centers or on machines in a peer-to-peer network. Dynamic name resolution techniques allow an end-user to connect to a service without any knowledge of physical resources associated with the service. A resource allocation algorithm dynamically optimizes the assignment of resources to services. We desire such an algorithm to be “self-tuning”, that is, to automatically adjust the allocation to maintain near-optimal performance. Significant research and development efforts have been expended towards the design, implementation, and deployment of efficient resource allocation algorithms. These efforts have been highly successful, but fall short of providing a general-purpose self-tuning resource allocation infrastructure. Numerous basic questions related to resource allocation