Integrated Evaluation of Network, System, and Application Software Architecture and Performance Issues in ATM Networks
Project Award Date: 07-01-1997
Three major forces are interacting to create changes in networks, system software architectures, and the nature of application software. The latest generation of network technology supports huge bandwidths. Potentially, these provide system and application software with communication channels whose speed, capacity, and varied Quality of Service (QoS) greatly extend how software can use networks. Advances in system software and middleware continue to motivate changes in the software architecture and operational requirements of leading-edge applications, thus changing the services they require from the operating systems and networks supporting them. Finally, emerging applications such as multimedia conferencing, video delivery, interactive WWW interfaces, and distributed multi-user virtual reality environments (e.g. multi-user games) require new types of services from the underlying end system and network. The gradual but continuing emergence of real-time execution constraints in leading edge applications is of particular interest, since it is especially challenging to add real-time abilities to existing systems and software.
Initial experience shows that while the potential of the emerging network and middleware technologies is real, realizing that potential will require progress in addressing their influence on the end-to-end performance of applications. The isolated consideration of system components' properties by researchers is a necessary, but often not sufficient, condition for understanding the causes of the end-to-end application. The interest of the service providers in end-to-end evaluation is motivated by a simple principle: customers pay for application performance, not for network services. This project concentrated on creating the ability to take an integrated view of end-to-end performance.
We created the Adaptable Performance Evaluation Testbed (APET) within which the influence of all components of the system-from the network to the application-can be evaluated, individually and collectively, in detail and in a single consistent hardware and software context. Extensions to the Data Stream Kernel Interface (DSKI) improved our ability to gather performance data from the internal operating system's components. We extended and improved the KU real-time (KURT) extensions to Linux for support of applications with real-time execution constraints that we use to support real-time ORB experiments, to experiment with real-time support for video client and server software, and to support real and proportional time simulation and emulation of ATM networks. We created the Performance Measurement Object (PMO) to enable the specification and execution of ORB real-time and conventional ORB performance evaluation experiments under NetSpec control. We have created an initial ORB benchmark suite using the PMO and have a prototype for a Performance Pattern Language (PPL), which provides higher-level semantics for describing application-level performance experiments. We also considered how agent technology could be applied to network management in the emerging object-oriented, real-time environment.
Faculty Investigator(s): Douglas Niehaus (PI)
Primary Sponsor(s): Sprint