Automated quality-of-service-aware configuration of publish-subscribe systems at design-time

Abstract

Configurable publish-subscribe middleware provides efficient support for the diverse Quality-of-Service (QoS) requirements of large-scale distributed applications. However, choosing the optimal middleware configuration to suit a specific application primarily remains a manual task within the responsibility of application developers. Existing configurable middleware approaches either do not allow extensive configuration or lack the possibility to specify QoS requirements in a domain-specific terminology. In this paper we present a novel methodology for the QoS-aware configuration of publish-subscribe middleware, which relies on discrete-event simulations and supervised learning techniques to automatically translate declarative descriptions of the application domain to an optimal configuration. The configuration is conducted at design-time to minimize possible run-time impacts. We implemented a design-time configurable middleware and evaluated the quality of the automated configuration workflow as well as the performance of event dissemination. Our results show that simulation experiments in conjunction with regression methods can be used to reliably derive optimal middleware configurations without further user interaction. We clarify that extensive design-time configurability does not compromise run-time performance and suits the requirements of demanding applications.