Autonomic Admission Control for Congested Request Processing Systems

Abstract

Consider a Transaction Server or any other kind of Request Processing System under congested workloads. The most probable outcome under peak congestion is for the system to degrade ungracefully and most sessions to receive timeouts after waiting for some piece of processing for a long period of time. Wasteful computation occurs as the system is unable to service in-time. Setting a session limit is the most common approach to minimize the problem, but while this approach is useful, it is neither autonomic nor dynamic. How do we determine a good value for the limit? Can we also have a strategy that automatically adapts to changes (e.g. the performance of backend database accesses can vary widely as it grows, as hardware is upgraded or as administrative tasks tune and modify settings)? In this paper we propose external adaptive control. We use a simulator environment to design and compare the strategies and also test against a transactional system.