Performance Measurement of Orphan Detection in the Argus System

In distributed systems, anomalies known as orphans can arise because of component failures and explicit aborts. An orphan is a computation that continues to execute even when its result is neither wanted nor needed, Orphans waste system resources and can see inconsistent data, which may cause them to behave erratically. In the Argus system, an orphan detection algorithm has been designed and implemented to detect and eliminate orphans before they can see inconsistent data. This algorithm is impractical since it piggybacks an unreasonable amount of data on system messages. As a part of this thesis, an optimization to the Argus orphan detection algorithm known as deadlining was implemented. Furthermore, the performance of the orphan detection algorithm was measured statically and observed under some simple dynamic loads. Measurement results show that although the algorithm is more practical to use after it has been extended to include deadlining, it still suffers from the current implementation of the Argus system, which is only a prototype system.