Learning Actions in Complex Software Systems

Administering service-oriented architecture (SOA) systems could require sophisticated rules to decide for instance whether to add or remove servers and when. Rule construction often necessitates experts to study patterns that contribute to changes or events. This is a time consuming and error-prone process for complex software systems. In this paper we test the feasibility of automating this process by mining historical data such as past service requests (in time series) and server change events that the administrator committed. We propose a new method to relate frequent patterns in a given time series to changes recorded in the event's history. We implemented and tested our method on a simulation system for SOA applications. First, we use Euclidean distance, DTW, and FastDTW to identify frequent patterns in a time series that represents performance metric of a SOA simulation system. Then, we calculate the confidence and support of frequent patterns that contribute to changes to identify a set of rules for automating changes. We tested rules that are generated using the proposed method in a training set on a testing set. The average accuracy of generated rules for the change event "remove" exceeded 80% in our experiments.