• Corpus ID: 211215

FAWNdamentally Power-efficient Clusters

@inproceedings{Vasudevan2009FAWNdamentallyPC,
  title={FAWNdamentally Power-efficient Clusters},
  author={Vijay Vasudevan and Jason Franklin and David G. Andersen and Amar Phanishayee and Lawrence Tan and Michael Kaminsky and Iulian Moraru},
  booktitle={HotOS},
  year={2009}
}

• V. Vasudevan, Jason Franklin, +4 authors Iulian Moraru
• Published in HotOS 18 May 2009
• Computer Science

Power is becoming an increasingly large financial and scaling burden for computing and society. The costs of running large data centers are becoming dominated by power and cooling to the degree that companies such as Microsoft and Google have built new data centers close to large and cost-efficient hydroelectric power sources [8]. Studies have projected that by 2012, 3-year data center energy costs will be double that of server equipment expenditures [15]. Power consumption and related cooling costs have become a primary design constraint at all levels, limiting the achievable density of data centers and large systems, and pushing processor manufacturers towards alternative architectures. While power constraints have pushed the processor industry toward multi-core architectures, power-efficient alternatives to traditional disk and DRAM-based cluster architectures have been slow to emerge. As a power-efficient alternative for data-intensive computing, we propose a cluster architecture called a Fast Array of Wimpy Nodes, or FAWN. A FAWN consists of a large number of slower but efficient nodes that each draw only a few watts of power, coupled with lowpower storage—our prototype FAWN nodes are built from 500MHz embedded devices with CompactFlash storage that are typically used as wireless routers, Internet gateways, or thin clients. Through our preliminary evaluation, we demonstrate that a FAWN can be up to six times more efficient than traditional systems with Flash storage in terms of queries per joule for seek-bound applications and between two to eight times more efficient for I/O throughput-bound applications (§3). Long-lasting, fundamental trends in the scaling of computation and energy suggest that the FAWN approach will become dominant for increasing classes of workloads. First, as we show in §4, slower processors are more efficient: they use fewer joules of energy per instruction than higher speed processors. Second, dynamic power scaling techniques are less effective than reducing a cluster’s peak power consumption. We conclude with an analysis of the design space for seek-bound workloads, showing how one should use a FAWN architecture for various dataset sizes and desired query rates (§5). 2 Data-intensive Computing