Steven D. Gribble
University of California at Berkeley
About the talk:
This talk presents the design, implementation, and evaluation of highly available, durable, scalable distributed data structures (DDS's) on clusters of workstations, intended primarily to vastly simplify the construction of scalable Internet services. A DDS is a self-contained, self-managing, consistent and available repository that exports a data-structure interface. A DDS platform decouples data persistence and consistency requirements from the rest of cluster-based Internet service logic, greatly simplifying service design and implementation. The main hypothesis of this work is that by providing service authors a small but carefully chosen selection of DDS's (such as a log, a hash table, and a tree), these authors will have enough flexibility to implement a wide variety of interesting services, but will also be shielded from many of the complexities of scalable, available, consistent state management on clusters. The DDS's are built on an asynchronous I/O layer that uses state machines to achieve high concurrency and data throughput, and design of the DDS's exploits properties of clusters (such as ample bandwidth, low latency, and a very small probability of a network partition) in areas such as the design of consistency protocols and recovery techniques. Example services built on the DDS platform, such as the instant messaging gateway and translation proxy "Sanctio", will be discussed in addition to the core platform.
About the speaker:
Steven D. Gribble (firstname.lastname@example.org) is a graduating Ph.D. student from the University of California at Berkeley, working primarily with Dr. Eric A. Brewer as part of the Ninja research group. Steve's research interests include the design, implementation, and evaluation of highly available, scalable Internet service platforms on cluster of workstations, and in the use of such platforms to aid the deployment and adoption of "Post-PC" devices and their infrastructure. Steve received his B.Sc. in Computer Science and Physics from the University of British Columbia in 1995, and his M.Sc. in Computer Science from Berkeley in 1997. He was also a co-founder of ProxiNet, Inc. (recently acquired by Puma Technology), which is commercializing thin client technology created at UC Berkeley.