Performance measurements and modeling of a java-based session initiation protocol (SIP) application server

Authors:
Greg Franks;Danny Lau;Curtis Hrischuk
Affiliations:
Carleton University, Ottawa, ON, Canada;Carleton University, Ottawa, ON, Canada;IBM, Durham, NC, USA
Venue:
Proceedings of the joint ACM SIGSOFT conference -- QoSA and ACM SIGSOFT symposium -- ISARCS on Quality of software architectures -- QoSA and architecting critical systems -- ISARCS
Year:
2011

Citing 7
Cited 0

The Stochastic Rendezvous Network Model for Performance of Synchronous Client-Server-like Distributed Software

IEEE Transactions on Computers
PC based precision timing without GPS

SIGMETRICS '02 Proceedings of the 2002 ACM SIGMETRICS international conference on Measurement and modeling of computer systems
The Method of Layers

IEEE Transactions on Software Engineering
The WebSphere application architecture and programming model

IBM Systems Journal
Layered Bottlenecks and Their Mitigation

QEST '06 Proceedings of the 3rd international conference on the Quantitative Evaluation of Systems
The Future of Software Performance Engineering

FOSE '07 2007 Future of Software Engineering
Enhanced Modeling and Solution of Layered Queueing Networks

IEEE Transactions on Software Engineering

Quantified Score

Hi-index	0.00

Visualization

Abstract

The Session Initiation Protocol (SIP) is an Internet protocol for establishing sessions between two or more parties. It is becoming ubiquitous in uses such as Voice over IP, instant messaging, Internet TV, and others. Performance is a chief concern with SIP because Quality of Service is important and SIP has internal timers that need to be honored or network efficiency suffers. The Java community has even provided a standardized API so that SIP applications can now be built using Java application servers. These new capabilities also bring with them new performance engineering methods, tools, and benchmarking needs. This paper describes the experiences and processes for the performance engineering of SIP applications in a Java environment. In this paper, a Java 2 Enterprise Edition (J2EE) SIP application server's performance is analyzed in a standalone and cluster environment, with network traces used to build a performance model of each environment. This included gathering data from test runs and extracting performance parameters from packet traces to construct the performance models. The models are then calibrated to match the model prediction with real system test data. Using the calibrated models, some bottlenecks were identified and suggestions to improve the overall maximum throughput were developed and were subsequently implemented in the system.