Performance Prediction of Distributed Enterprise Applications with Session Persistence

Distributed enterprise applications store client-specific data in application server memory, in a data structure commonly referred to as the session. The backing up of this session data (in a process known as session persistence) is essential for load balancing between application servers and for managing server failures, and hence is critical for the success of distributed enterprise applications. This paper investigates how session persistence affects web application performance and models the performance of distributed enterprise applications with and without session persistence using a layered queuing model. Experimental results are presented showing measured versus predicted mean response time and throughput under a range of workloads, with session persistence both active and inactive. These results i.) show that the layered queuing model gives extremely accurate predictions (typically over 90% accurate); and ii.) quantify the significant degradation in system performance that session persistence incurs.