Distributed
Parity: Protecting Shared Memory
Tuesday November 23, 2004
Hamerschlag Hall D-210
4:00 pm
Jangwoo
Kim
Carnegie Mellon University
In recent years, fault-tolerance has become an increasingly
important feature of shared-memory multiprocessor systems.
Researchers have investigated various mechanisms using some
level of redundancy to protect a single processor core, interconnection
network, or memory system. The distributed parity mechanism
was recently proposed to avoid a non-ECC recoverable data loss
in shared-memory multiprocessor systems.
In this talk, I will present our distributed parity proposal.
Our shared-memory protection mechanism minimizes the performance
overheads of past distributed parity solutions.
Jangwoo Kim is a PhD candidate in the Department
of Electrical and Computer Engineering at Carnegie Mellon University.
His research interests include fault tolerant computer architecture,
multiprocessor architecture and full computer system simulation.
He received his Bachelor and Master degress from Cornell University.
His advisor is Babak Falsafi.
|