Causal synchrony in the design of distributed programs
The outcome of any computation is determined by the order of the events in the computation and the state of the component variables of the computation at those events. The level of knowledge that can be obtained about event order and process state influences protocol design and operation. In a centr...
| Main Author: | |
|---|---|
| Format: | Others |
| Language: | English |
| Published: |
W&M ScholarWorks
1994
|
| Subjects: | |
| Online Access: | https://scholarworks.wm.edu/etd/1539623855 https://scholarworks.wm.edu/cgi/viewcontent.cgi?article=3765&context=etd |
| Summary: | The outcome of any computation is determined by the order of the events in the computation and the state of the component variables of the computation at those events. The level of knowledge that can be obtained about event order and process state influences protocol design and operation. In a centralized system, the presence of a physical clock makes it easy to determine event order. It is a more difficult task in a distributed system because there is normally no global time. Hence, there is no common time reference to be used for ordering events. as a consequence, distributed protocols are often designed without explicit reference to event order. Instead they are based on some approximation of global state. Because global state is also difficult to identify in a distributed system, the resulting protocols are not as efficient or clear as they could be.;We subscribe to Lamport's proposition that the relevant temporal ordering of any two events is determined by their causal relationship and that knowledge of the causal order can be a powerful tool in protocol design. Mattern's vector time can be used to identify the causal order, thereby providing the common frame of reference needed to order events in a distributed computation. In this dissertation we present a consistent methodology for analysis and design of distributed protocols that is based on the causal order and vector time. Using it we can specify conditions which must be met for a protocol to be correct, we can define the axiomatic protocol specifications, and we can structure reasoning about the correctness of the specified protocol. Employing causality as a unifying concept clarifies protocol specifications and correctness arguments because it enables them to be defined purely in terms of local states and local events.;We have successfully applied this methodology to the problems of distributed termination detection, distributed deadlock detection and resolution, and optimistic recovery. In all cases, the causally synchronous protocols we have presented are efficient and demonstrably correct. |
|---|