DATE: Thursday, February 17, 2005
TIME: 12:00 pm - 1:00 pm
PLACE: Wean Hall - Room 8220

Jay Wylie

A read/write protocol family for versatile storage infrastructures

Today, high performance, high reliability storage systems are heavily engineered, very specialized, monolithic systems. The cost of such systems, in terms of raw storage capacity per dollar, is much higher than commodity storage. Storage bricks, scalable storage components built from commodity components, could reduce the cost of high performance, high reliability storage. I have developed the Read/Write Protocol Family that enables a versatile storage infrastructure to be built out of storage bricks. Where versatile means "able to meet radically different fault-tolerance and performance requirements without modifying storage bricks".

The Read/Write Protocol Family incorporates standard storage techniques (e.g., replication, striping, and RAID) with techniques for additional fault-tolerance (erasure codes and Byzantine fault tolerance) and scalability (witnesses and quorum systems). We believe that the versatility, scalability, and extraordinary fault-tolerance realized by the Read/Write Protocol Family are necessary for storage brick-based systems.

Evaluation of a prototype storage system based on the Read/Write Protocol Family shows that versatility can be provided efficiently and that a single storage infrastructure can meet diverse fault-tolerance and performance requirements. As well, the prototype allows the real performance and cost trade-off of different failure assumptions to be explored.

This talk is based on joint work with Michael Abd-El-Malek, Greg Ganger, Garth Goodson, and Mike Reiter.

Jay is a Ph.D. candidate at Carnegie Mellon University. He works in the Parallel Data Laboratory under the supervision of Greg Ganger and Mike Reiter. He received his Masters in Electrical and Computer Engineering from Carnegie Mellon University in 2000 and his B.A.Sc. in Systems Design Engineering from the University of Waterloo in 1998. Jay is interested in distributed systems, fault-tolerance & dependability, consistency protocols, and storage systems.

For Further Seminar Info Contact:

or visit