Bottom-Up Fault Management in Service-Based Systems

Abstract

Service Oriented Architecture (SOA) enables the creation of distributed applications from independently developed and deployed services. As with any component-based system, the overall performance and quality of the system is an aggregate function of its component services. In this article, we present a novel approach for managing bottom-up faults in service-based systems. Bottom-up faults are a special case of system-wide exceptions that are defined as abnormal conditions or defects occurring in component services, which if not detected and/or managed, may lead to runtime failures. Examples of bottom-up faults include network outage, server disruption, and changes to service provisioning (e.g., new operation parameter required) that may have an impact on the way component services are consumed. We propose a soft-state signaling-based approach to propagate these faults from participants to composite services. Soft-state refers to a class of protocols where the state of a service is constantly refreshed by periodic messages, and user/service takes up the responsibility of communicating and maintaining its state. Soft-state-based protocols have a number of advantages including implicit error recovery and easier fault management, resulting in high availability for systems. Although soft-state has been widely used in various Internet protocols, this work is the first (to the best of our knowledge) to adopt soft-state for fault management in composite services. The proposed approach includes protocols for fault propagation (pure soft-state and soft-state with explicit removal) and fault reaction (rule-based). We also present experiment results to assess the performance and applicability of our approach.