Distributed Fault-Tolerance
David Powell
LAAS-CNRS, Toulouse, France
david.powell@laas.fr

Full Paper in Postscript File




A distributed system can be defined as a set of computing nodes,
interconnected by a communication network, that cooperate to carry out 
some common work. The nodes can typically be considered as independent 
from the viewpoint of failures, so distribution can be a useful framework 
for providing fault tolerance.

However, distribution of system state and other dependencies between
cooperating tasks also mean that a distributed service can easily be
disrupted if any of the nodes involved should fail. With the added
potential complication of unreliable communication, distributed services
often need to be made fault-tolerant if they are to be useful.

These two opposing facets of the relationship between fault-tolerance and
distribution are strong motivations for distributed fault tolerance
techniques. An important characteristic of these techniques is that error
processing and fault treatment are implemented mainly by software using
distributed, message-passing algorithms.

The tutorial will address distributed system models and assumptions,
consistency issues, and fault-tolerance based on recovery and replication.

			   Speaker's biography

Dr. David Powell holds a position of "Directeur de Recherche" at LAAS-CNRS,
Toulouse, where he leads the Dependable Computing and Fault Tolerance
Research Group. He has written over ninety papers for international and
national journals and conferences, is co-author of two books on dependable
computing and holds a patent for a fault and damage-tolerant network for
data transmission. His current research work concerns distributed
algorithms for software-implemented fault-tolerance, stochastic Petri net
modelling for dependability evaluation and the use of fault injection for
estimating fault-tolerance coverage. He has managed several national and
European research contracts. In particular, he was the Scientific Director
of the Delta-4 Esprit project and is currently chief architect of the
GUARDS Esprit project. He has acted as a consultant for several aerospace,
telecommunication and data processing companies in France and for the
European Commission.  Dr. Powell is a member of ACM and IEEE.