Fault Tolerance Through Automated Diversity in the Management of Distributed Systems

Nowadays the reliability of software is often the main goal in the software development process. Despite more and more improvements in fault preventing techniques, it is a fact that faults remain in every complex software system. In contrast to hardware-faults, no concepts or mechanisms for fault tolerance of general software-faults became widely accepted. In this paper we present a new concept for the design of system management, that enables the tolerance of software-faults of the executed applications. We explain how the underlying resource management of a distributed system affects the occurrence of errors of a faulty application and how the re-execution of the application with altered resource management decisions prevents the occurrence of errors with certain probability. Our contribution is to motivate a new approach for fault tolerance in distributed systems and to give a general concept for system designers which states what decision spaces can be used to tolerate faults and how different alternatives in one decision space can be evaluated and generated. The concept is applicable for a large class of system mod-