Performance modeling of an all softcopy image analysis and exploitation system

Commercial computing technologies are now a practical option for creating an all softcopy environment for remote sensing imagery analysis and exploitation. An architecture based upon the client/server paradigm that optimally utilizes the mass storage capabilities of large servers, the data transport speed of Asynchronous Transfer Mode (ATM), and multiprocessor workstations has many attractive features. However, before such a system can become operational, issues relating to throughput need to be addressed. These issues include network connection speed, ATM switch size, server throughput and workstation processing capability. The total number of workstations that can be supported with a particular configuration is ultimately determined by the average response time required to perform typical operations. This paper details the use of a discrete event simulator to model an image archiving and retrieval system. Simulation modeling places few restrictions on the fidelity with which network elements can be represented. In addition, the actions of an analyst, as he/she goes through typical operations, making demands upon the system resources, can be realistically portrayed. The stochastic nature of demands for network resources is modeled using random number generators. The execution of a simulation model is similar to running an in vitro experiment on a candidate system. A candidate system is presented and performance issues relating to significant configuration items are identified. Included are typical user response time curves.