An Adaptive Feedback Controller for SIP Server Memory Overload Protection

摘要

Memory overload is one of the key challenges in managing enterprise application server performance such as in J2EE environment. Realizing this issue, J2EE middleware providers introduced management solutions to prevent memory overload. However, such solutions depend on configurable parameters, such as thread pool size and queue length limit, which are typically difficult to determine and unaccommodating to application behavioral changes. In this paper we introduce a dynamic memory overload protection solution based on applying adaptive feedback controller techniques. Our solution consists of (1) a model that captures the dynamics of system and workload characteristics, (2) a Kalman filter for model parameter estimation, and (3) an adaptive feedback controller that regulates the load admission rate to the server in order to prevent memory overload. In particular, we have designed and implemented a self-configurable memory overload controller, which is automatically tuned based on an analytical model of the system under control. We conducted experiments where we used our controller prototype to manage a commercially available SIP server and compared it to a static, fixed gain feedback controller. Our experimental results show that, with little configuration effort, the adaptive controller reacts nicely to workload dynamics.