Asymptotically tight bounds for computing with faulty arrays of processors

Authors:
C. Kaklamanis;A. R. Karlin;F. T. Leighton;V. Milenkovic;P. Raghavan;S. Rao;C. Thomborson;A. Tsantilas
Affiliations:
Aiken Comput. Lab., Harvard Univ., Cambridge, MA, USA;-;-;-;-;-;-;-
Venue:
SFCS '90 Proceedings of the 31st Annual Symposium on Foundations of Computer Science
Year:
1990

Citing 0
Cited 7

Optimal Sorting Algorithms on Incomplete Meshes with Arbitrary Fault Patterns

ICPP '97 Proceedings of the international Conference on Parallel Processing
Lower Bounds in Distributed Computing

DISC '00 Proceedings of the 14th International Conference on Distributed Computing
RACE: A Software-Based Fault Tolerance Scheme for Systematically Transforming Ordinary Algorithms to Robust Algorithms

IPDPS '01 Proceedings of the 15th International Parallel & Distributed Processing Symposium
The Robust-Algorithm Approach to Fault Tolerance on Processor Arrays: Fault Models, Fault Diameter, and Basic Algorithms

IPPS '98 Proceedings of the 12th. International Parallel Processing Symposium on International Parallel Processing Symposium
Routing complexity of faulty networks

Random Structures & Algorithms
Efficient automatic simulation of parallel computation on networks of workstations

Discrete Applied Mathematics
Priority queues resilient to memory faults

WADS'07 Proceedings of the 10th international conference on Algorithms and Data Structures

Quantified Score

Hi-index	0.00

Visualization

Abstract

The computational power of 2-D and 3-D processor arrays that contain a potentially large number of faults is analyzed. Both a random and a worst-case fault model are considered, and it is proved that in either scenario low-dimensional arrays are surprisingly fault tolerant. It is also shown how to route, sort, and perform systolic algorithms for problems such as matrix multiplication in optimal time on faulty arrays. In many cases, the running time is the same as if there were no faults in the array (up to constant factors). On the negative side, it is shown that any constant congestion embedding of an n*n fault-free array on an n*n array with Theta (n/sup 2/) random faults (or Theta (log n) worst-case faults) requires dilation Theta (log n). For 3-D arrays, knot theory is used to prove that the required dilation is Omega ( square root log n).