HP Labs Technical Reports

Hybrid and Predictive Admission Strategies to Improve the Performance of an Overloaded Web Server

Cherkasova, Ludmila; Phaal, Peter

HPL-98-125R1

Keyword(s): web servers; overloaded conditions; admission control strategy; performance analysis; optimization; SBAC

Abstract: In this paper, we use a session-based workload to measure a web server's performance. We define a session as a sequence of client requests. An overloaded web server can experience a severe loss of throughput when measured as the number of completed sessions. Moreover, the overloaded web server discriminates against longer sessions. This could significantly impact sales and profitability of commercial web sites because longer sessions are typically the ones that would result in purchases. Session based admission control (SBAC), introduced in [CP98], prevents a web server from becoming overloaded and ensures that longer sessions can be completed. If a server is functioning near its capacity a new session will be rejected (or redirected to another server if one is available). If there is enough capacity, the admission control mechanism will admit a new session and process all future requests related to it. In this paper, we propose two new admission control strategies: hybrid and predictive, aiming to optimize the performance of SBAC mechanism. The hybrid strategy is based on a self-tunable admission control function, adjusting itself accordingly to variations in traffic loads. It shows improved performance results for workloads with medium to long average session length. The predictive strategy estimates the number of new sessions it can accept and still guarantee processing of all future session requests. It consistently shows the best performance results for different workloads and different traffic loads.

28 Pages

Back to Index