Date Awarded

1997

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Computer Science

Abstract

Distributed systems are characterized by the fact that the constituent processes have neither common memory nor a common system clock. These processes communicate solely via message passing. While providing a number of benefits such as increased reliability, increased computational power, and geographic dispersion, this architecture significantly complicates many of the tasks of software development and verification, including evaluation of the program state. In the case of distributed systems, the program state is comprised of the local states of the constituent processes, as well as the state of the channels between processes, and is called the global state.;With no common system clock, many distributed system protocols rely on the global ordering of local process events imposed by the message passing that occurs between processes. This leads to a partial global ordering of local process events, which can then be used to determine which process states could (or could not) have occurred simultaneously.;Traditional predicate evaluation protocols evaluate predicates on the global state of a distributed computation using consistent global states. This evaluation is complicated by the fact that the event ordering imposed by message passing is only partial. A complete history of the global states that occurred during an execution cannot always be constructed. This introduces inefficiency into predicate detection protocols and prohibits detection of certain predicates.;This dissertation explores the use of this rough global time base for global state predicate evaluation within distributed systems. By structuring the evaluation on the assumption that a global time base exists, we can develop simple and efficient protocols for both stable and unstable predicate evaluation. Further, we can evaluate certain predicates which are not easily evaluated using consistent global states. We demonstrate these advantages by developing protocols for detection of distributed termination, distributed deadlock detection, and detection of certain unstable predicates as they occur. as the global time base is rough, we can only detect unstable predicates which remain true for a sufficient duration. We additionally develop several formalizations which assist the protocol developer in dealing with the fact that the global time base is not perfect. We demonstrate the application of these formalizations within the protocols that we develop.

DOI

https://dx.doi.org/doi:10.21220/s2-hst4-gb04

Rights

© The Author

Download

Share Feedback

Share

COinS