NonStop Systems: Then and Now
Tuesday October 25, 2005
Hamerschlag Hall D-210
Carnegie Mellon University
For decades, fault-tolerant computing has been the purview of mainframes
such as the HP/Tandem NonStop and IBM z900 series. With rising concerns
over device reliability, fault-tolerant designs are making their
way into mainstream, commodity microprocessors. But before re-inventing
reliable computing, current architects must pay attention to the
lessons from past (and present) mainframes. Over the years, the
Tandem NonStop series has shifted from custom processors to commodity
microprocessors and, most recently, to chip multiprocessors (CMPs).
The shift to CMPs and modern microarchitecture does not come for
free, however, and in this talk I will focus on the challenges facing
the NonStop system by reviewing the recent paper, NonStop
Advanced Architecture from DSN '05.
Brian Gold is a 3rd year PhD student advised by Babak Falsafi.
He works on the TRUSS reliable server project, which has an upcoming
article in the Nov/Dec issue of IEEE Micro special issue on reliable
computing. He received BS degrees in Electrical Engineering and
Applied Math and an MS degree in Computer Engineering, all from