Authors:
Christian Dauer Thorenfeldt Sellberg
1
;
Michael R. Hansen
2
and
Paul Fischer
2
Affiliations:
1
IBM Denmark A/S, Denmark
;
2
Institute of Mathematical Modelling, Technical University of Denmark, Denmark
Keyword(s):
Fault tolerance, Dependable systems, Distributed systems, Process Algebra, Pi-calculus.
Related
Ontology
Subjects/Areas/Topics:
Dependable Computing
;
Enterprise Information Systems
;
Enterprise Software Technologies
;
Information Systems Analysis and Specification
;
Reliable Software Technologies
;
Requirements Analysis And Management
;
Software Engineering
Abstract:
We live in a time where we become ever more dependent on distributed computing. Predictable quantitative properties of reliability and resource requirements of these systems are of outmost importance. But today quantitative properties of these systems can only be established after the systems are implemented and released for test, at which point problems can be costly and time consuming to solve. We present a new method, a process algebra and simulation tool for estimating quantitative properties of reliability and resource requirements of a distributed system with complex behaviour hereunder complex fault-tolerance behaviour. The simulation tool allows tailored fault injection e.g. random failure and attacks. The method is based upon π-calculus (Milner, 1999) to which it adds a stochastic fail-able process group construct. Performance is quantitatively estimated using reaction rates (Priami, 1995). We show how to model and estimate quantitative properties of a CPU scavenging grid wi
th fault-tolerance. To emphasize the expressiveness of our language called Gπ we provide design patterns for encoding higher-order functions, object-oriented classes, process translocation, conditional loops and conditional control flow. The design patterns are used to implement linked lists, higher-order list functions and binary algebra. The focus of the paper is on practical application.
(More)