Performance Modeling of a Web Server with a Dynamic Pool of Service Processes.pdfVIP

Performance Modeling of a Web Server with a Dynamic Pool of Service Processes Tien Van Do, Ram Chakka, Thang Le Nhat, Udo R. Krieger 1 Abstract This paper considers the multi-processing module Prefork in the Unix version of Apache, one of the most widely applied software solutions for Web servers in todays Internet. It analyzes its dynamic pool size behavior. For the first time, we propose a queueing model to approximate the performance of the latter. Numerical results computed by advanced matrix-analytic methods are compared with measurement results. They clearly indicate that the proposed analytic model can indeed accurately predict the performance of the dynamic pool size behavior of an Apache Web server. Key words: Web server performance, queueing network model, QBD-M processes 1 Introduction Previous analytic studies of a Web server have not sufficiently considered the internal structure and operation of its software architecture, which has a crucial impact on the performance. In particular, the dynamic behavior of the available number of operating processes (or threads) serving HTTP requests inside the software architecture plays a critical role. However, so far all presented analytic performance models are not able to capture explicitly the dynamic pool size behavior that handles the concurrency of HTTP requests. We propose a new queueing model with a variable number of servers to describe the dynamic pool size implemented in an Apache Web server. Applying a batch Markovian arrival process and advanced matrix-geometric solution techniques, results on the system performance are derived and compared with actual measurements. 2 Analysis of a Web server with dynamic pool size We consider the UNIX software architecture of an Apache Web server, in particular, the operation of the Apache multi-processing module (MPM) and the dynamic pool of the