Home // DEPEND 2013, The Sixth International Conference on Dependability // View article
Authors:
Martin Perner
Martin Sigl
Ulrich Schmid
Christoph Lenzen
Keywords: fault-tolerance; self-stabilization; clock distribution; fault-injection; simulation analysis
Abstract:
We present a prototype implementation and simulation-based evaluation of a recently proposed novel approach for Byzantine fault-tolerant and self-stabilizing clock distribution in multi-synchronous GALS architectures. Fault-tolerant clock generation and clock distribution is a mandatory prerequisite for highly dependable multicore processors and Systems-on-Chip, as it removes the single point of failure typically created by central oscillators and conventional clock distribution trees. Our scheme, termed HEX, is based on a hexagonal grid topology, which connects simple intermediate nodes implemented using the UMC 90 nm standard cell library. Their purpose is to (i) forward synchronized clock signals throughout the grid and (ii) to supply the clock to nearby application modules. To achieve (ii), we show how to construct a fast clock on top of the clock signal provided by HEX and analyze its properties. In sharp contrast to existing solutions, HEX is not only Byzantine fault-tolerant, but also self-stabilizing, i.e., it can recover from arbitrarily corrupted system states. ModelSim-based simulation experiments confirm the excellent performance and fault-tolerance properties of our approach achieved in practice, which were already suggested by an earlier theoretical worst-case analysis.
Pages: 6 to 15
Copyright: Copyright (c) IARIA, 2013
Publication date: August 25, 2013
Published in: conference
ISSN: 2308-4324
ISBN: 978-1-61208-301-8
Location: Barcelona, Spain
Dates: from August 25, 2013 to August 31, 2013