Optimal static distribution of prioritized customers to heterogeneous parallel servers

-The main aim of this paper is to present a model for distributing prioritized customers to heterogeneous parallel servers. A queuing optimization model is proposed in the form of a nonlinear programming problem having linear constraints, so that the average mean response time taken over all job requests is minimized. A solution procedure is developed by employing the convexity of the performance objective. A distributed system is illustrated to demonstrate the applicability of the model. In addition, the model is extended to the case of class-dependent policies. 1. I N T R O D U C T I O N The main aim of this paper is to present a queuing optimization model for distributing prioritized customers in distributed processing environments. Customers are originators of job requests. Consider a system with heterogeneous parallel-processing servers, where jobs arrive randomly. Jobs are classified into a finite number of distinct types. The servers have different expected processing times for different classes of jobs. To distribute jobs among parallel servers is of fundamental importance in enhancing the performance of such a distributed system. To realize this benefit, however, system designers must solve the problem of assigning jobs so that the system is used to its fullest advantage. The objective is to determine the portion of job requests transmitted to each server so that a minimum overall average job response time is achieved. Jobs on a singleprocessing server are processed according to an absolute preemptive or non-preemptive priority queuing discipline. Algorithms for distributing customers on queuing systems are grouped into two classes: static and dynamic. Most distribution algorithms have been studied in a distributed computer configuration [1]. A static algorithm is one in which decisions concerning scheduling of job do not depend on the state of the system. Dynamic algorithms may outperform static algorithms by using short-term fluctuations of the system load to improve the quality of their decisions [2]. However, a static policy is simple to implement and maintain while a dynamic policy involves "t'Heeseok Lee is Assistant Professor of Management Information Systems at the Korea Advanced Institute of Science and Technology, Seoul, Korea. He taught MIS for the University of Nebraska at Omaha and the State University of Minnesota at Moorhead. He received his bachelor's degree in Industrial Engineering from Seoul National University, his master's degree in Industrial Engineering from the Korea Advanced Institute of Science and Technology, and his doctorate in MIS from the University of Arizona. His research interests include building an enterprise database and telecommunication infrastructure, database design, IS planning, and application of OR techniques in MIS. He has his research paPers published or forthcoming in Information and Management, Information Systems, Information Processing and Management, Omega, Computers and Industrial Engineering, Journal of Systems and Software, Computers and Operations Research, European Journal of Operational Research, Engineering Economist, Systems Science, and Annals of Operations Research.