Performance engineering of enterprise computing systems

Client/server(CIS) system architectures for enterprise-wide computing are becoming increasingly popular because of the advantages such as scalability, portability, interoperability,' and reliability. However, the distributed nature of such systems, as compared to centralized architectures, requires careful performance engineering of each system component so that the over" system performance meets the customer's requirements. Otherwise, the design has excess capacities in subsystems that future growth cannot take advantage of because they are not the potential performance bottlenecks. In this talk we first describe our performance engineering methodology for designing C/S systems. Our approach is top down in ftt we start with the throughput and response time requirements of the tasks that system supports. Then we traverse down the layers in the OSI model, from the application layer to the physical layer, mapping the activities in the various wks into resource consumption in the subsystems, such as client workstations, local area network, database servers, communication gateways, wide area networks and mainfrarne computers. System models are then evaluated for different workload scenarios using anatlytical techniques and/or simulation. Trade-offs in hardware, software, and networking technologies can then be performed for capacity planning and system design purposes. Then we describe the recent enhancements to Q+ designed specifically to support the design of C/S systems. Q+, the AT&T Queueing+Simulation Software is the visual interactive simulation tool developed at AT&T Bell Laboratories by the Performance Analysis Department to meet a wide range of design and analysis needs in communications, and computers. We have added a rich library of submodels for modeling subsystems in C/S systems, along with an interface to connect these submodels together quickly for modeling entire systems.