FAWN A Fast Array of Wimpy Nodes - CMU Computer Science.pdfVIP

To appear in 22nd ACM Symposium on Operating Systems Principles (SOSP’09) This version is reformatted from the official version that appears in the conference proceedings. FAWN: A Fast Array of Wimpy Nodes 1 1 2 David G. Andersen , Jason Franklin , Michael Kaminsky , 1 1 1 Amar Phanishayee , Lawrence Tan , Vijay Vasudevan 1Carnegie Mellon University, 2Intel Labs Abstract 1 Introduction This paper presents a new cluster architecture for low-power Large-scale data-intensive applications, such as high- data-intensive computing. FAWN couples low-power embed- performance key-value storage systems, are growing in both ded CPUs to small amounts of local flash storage, and balances size and importance; they now are critical parts of major In- computation and I/O capabilities to enable efficient, massively ternet services such as Amazon (Dynamo [ 10]), LinkedIn parallel access to data. (Voldemort [41]), and Facebook (memcached [33]). The workloads these systems support share several charac- The key contributions of this paper are the principles of teristics: they are I/O, not computation, intensive, requiring the FAWN architecture an