How to Commit Concurrent, Non-Isolated Computations

  • Authors:
  • Edgar Nett;Michael Mock

  • Affiliations:
  • -;-

  • Venue:
  • FTDCS '95 Proceedings of the 5th IEEE Workshop on Future Trends of Distributed Computing Systems
  • Year:
  • 1995

Quantified Score

Hi-index 0.00

Visualization

Abstract

We face today the situation that although distributed system are potentially more dependable than centralized systems, they actually confront the users, programmers, and system administrators with more complex and hard to manage failure situations than centralized systems. Fault tolerance in distributed systems is still a promising and not yet mastered area of research. One key feature of every (software) fault tolerance mechanism is the ability to commit system progress, i.e., to guarantee that some reached state will no longer be subject to any considered fault. The paper focuses on a formal model extending the notion of commit, which is originally established in classical transaction theory for isolated transactions, to the notion commit-correctness, reflecting the possibility of interactions between different computations. The problem of the domino-effect, originally established in the context of a process oriented, message based paradigm, is inherently solved by this approach.