Long Atomic Computations,
Abstract
Distributed computing systems are becoming commonplace and offer interesting opportunities for new applications. In a practical system, the problems of synchronizing concurrent computations and recovering from failures must be dealt with effectively. Atomicity has been suggested as a tool that masks concurrenty and failures from the users of a system. With synchronization and recovery mechanisms, atomic computations appear to execute indivisibly. This dissertation addresses the issues in implementing long atomic computations, such as computations that last for hours or even days. Long computations make synchronization more difficult because their execution is more overlapped. They are also more likely to encounter failures in their execution. Three issues are raised: 1. Should long computations be executed automatically? Or should atomicity be replaced with other correctness criteria to increase the concurrency of a system? 2. If long atomic computations can be implemented practically, are there implementation paradigms that application programmers can follow to simplify the programming effort? 3. How can long atomic computations be made resilient to transient failures? (Theses)
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1986
- Accession Number
- ADA174788
Entities
People
- Pui Ng
Organizations
- Massachusetts Institute of Technology