Research at Indiana University for Reliable Petascale Performance

As the rate of increase of individual processor
performance slows, the need to reliably and efficiently
use larger and larger numbers of processors
increases. With today’s top supercomputers
exceeding one petaFLOP/second, the number
of computational and network components has
increased by hundreds of thousands per machine.
With such a large number of components, the mean
time between failures of the whole machine decreases
dramatically. This presentation will discuss some
of the research projects now ongoing at Indiana
University that attempt to address this problem.