A Self-Adaptive Spike Detection Algorithm with Application in Performance and Capacity Management

This paper presents a simple, yet novel, approach to address the problem of spike detection. The definition of a spike varies with the domain under study. This results in domain-specific solutions that are often rendered unusable in other domains. Moreover, the spike detection algorithms proposed in the past are effective only when suitably tuned. This often limits their usability in real-life scenarios. In this paper, we present a generic approach to address the problem. We propose various spike definitions and present a computationally light-weight algorithm that can cater to all these definitions. The algorithm can automatically adapt itself to changing data characteristics. We demonstrate the soundness of our approach by experimental evaluation. We also compare the proposed algorithm with the existing implementations. We then present an application of spike detection in the domain of performance and capacity management in enterprise systems by presenting real-life case-studies.