Case Study III: Performability Management in Distributed Database Systems

Abstract

Among the applications of adaptive methods, those for distributed systems present greater challenges. In particular, capturing system conditions in the context of distributed computing for adaptive decision making is very difficult and costly for at least two reasons. First, obtaining information about remote sites requires additional message passing such that the benefits from adaptive mechanisms could be negated by a performance penalty. Secondly, there is always a latency between the time when the information about remote sites is collected and the time when an adaptive decision is made, which may make the adaptation ineffective since the information (based on which the adaptive decision is made) could become obsolete through the latency.