DC: Small: Collaborative Research: DARE: Declarative and Scalable Recovery

点击翻译按钮获取中文摘要

One dominant characteristic of today's large-scale computing systemsis the prevalence of large storage clusters. Storage clusters at thescale of hundreds or thousands of commodity machines areincreasingly being deployed. At companies like Amazon, Google, Yahoo,and others, thousands of nodes are managed as a single system.As large clusters have brought many benefits, they also bring a newchallenge: a growing number and frequency of failures that must bemanaged. Bits, sectors, disks, machines, racks, and many othercomponents fail. With millions of servers and hundreds of datacenters, there are millions of opportunities for these components tofail. Failing to deal with failures will directly impact thereliability and availability of data and jobs.Unfortunately, we still hear data-loss stories even recently. Forexample, in March 2009, Facebook lost millions of photos due tosimultaneous disk failures that "should" rarely happen at the sametime (but it happened); in July 2009, a large bank was fined a recordtotal of 3 millions pounds after losing data on thousands of itscustomers; more recently, in October 2009, T-Mobile Sidekick, whichuses Microsoft's cloud service, also lost its customer data. Theseincidents have shown that existing large-scale storage systems arestill fragile to failures.To address the challenges of large-scale recovery, the goal of thisproject is to: (1) seek the fundamental problems of recovery intoday's scalable world of computing, (2) improve the reliability,performance, and scalability of existing large-scale recovery, and (3)explore formally grounded languages to empower rigorous specificationof recovery properties and behaviors. Our vision is to build systemsthat "DARE to fail": systems that deliberately fail themselves,exercise recovery routinely, and enable easy and correct deployment ofnew recovery policies.For more information, please visit this website:http://boom.cs.berkeley.edu/dare/