Thursday, April 17 , 2008
12:00 pm - 1:00 pm
Wean Hall 8220
Byzantine Fault-tolerant Erasure-coded Storage
As distributed storage systems grow in size and importance, they must tolerate faults other than crashes. Protocols that can tolerate arbitrary faulty behavior by some components of a system are said to be Byzantine fault-tolerant. Byzantine fault-tolerance, however, is not common in practice because it has often required additional hardware, network resources, and computational resources. This talk will describe a Byzantine fault-tolerant erasure-coded storage protocol and prototype that we have built that performs nearly as well as similar protocols that tolerate only crashes. Our protocol relies on a novel cryptographic primitive that we developed, homomorphic fingerprinting, which can be used to efficiently verify erasure-coded distributed data. Compared to a similarly-configured cluster-based erasure-coded storage system, our protocol requires the same number of storage nodes and similar network and computational resources, but our protocol tolerates a fraction of Byzantine faulty storage nodes and any number of Byzantine faulty clients.
Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.
James Hendricks is a PhD candidate in the Computer Science Department. He is interested in distributed systems, fault tolerance, security, and storage systems.
or visit http://www.pdl.cmu.edu/SDI/