An application-level implementation of causal timestamps and causal ordering

Abstract

Maintenance of causality information in distributed systems has previously been implemented in the communications infrastructure with the focus on providing reliability and availability for distributed services. While this approach has a number of advantages, moving causality information up into the view and control of the application programmer is useful, and in some cases, preferable. In an experiment at the University of Queensland, libraries to support application-level maintenance of causality information have been implemented. The libraries allow the collection and use of causality information under programmer control, supplying a basis for making causal dependency information available for application management and troubleshooting. The libraries are also unique in supporting existing distributed systems based on the remote procedure call paradigm. This paper describes the underlying theory of causality, and the design and implementation of the libraries. An event reporting service example is used to motivate the approach, and a number of previously unresolved practical problems are addressed in the design process.