A Kalman Filter based Hill-climbing Strategy for Application Server Configuration

Application server needs to adjust the parameters steadily to maintain the performance of system, especially in the complicated and dynamic Internet environment. However, the common manual adjustment is always difficult, time-consuming and error-prone. Therefore there is an urgent requirement of the self-adaptive application server which adjusts its server parameters at runtime. We adopt an approach of hill-climbing based optimization algorithm, which is guided by fuzzy control. After testing it in real scenarios, we find this method may lead to worse performance in some critical conditions compared with the manual way. After further analyzing, we conclude the worse performance resulting from the inaccuracy of monitor and the delay of the process of decision-making. To deal with this problem, we extend the approach with Kalman filter to reduce the deviation of the measurement. At the same time, we deploy a web service in the application server and test the corresponding workload to validate the effectiveness of this self-adaptive strategy.