Predictive Self-Healing of Web Services Using Health Score

Existing self-healing mechanisms for Web services constantly monitor services and their computational environment, analyze system state, determine failure occurrences, and execute built-in recovery plans (MAPE loop). We propose a more pro-active self healing mechanism that uses a multi-layer perceptron ANN and a health score mechanism to learn about the occurrences of failures or quality of service degradation in advance, without requiring modifications to the framework of services used by applications. Highest score is assigned to the system upon start and is degraded during system execution whenever a service fails to operate or the time-to-leave (TTL) of the client side requests increases. Application of the proposed mechanism to a set of vehicle tracking Web services decreased the probability of out of service occurrences by 70% and increased system quality of service by 13%. The overhead of the mechanism was nearly 3% and negligible, whilst TTL for a request from the client side decreased by 20%.