Scalable resource scheduling: design, assessment, prototyping

Resource scheduling in distributed systems aims at achieving maximal system performance by utilizing the available system resources efficiently. Large distributed systems, comprising hundreds or thousands of nodes and spanning vast geographical distances (e.g. Internet), require resource scheduling to be scalable. Scalability has become a common requirement in the design and development of distributed software. This paper describes a comprehensive approach to software development, leading from the stage of requirements specification, through design and algorithm assessment to a prototype implementation of a scalable resource scheduling policy. Scalability is achieved by system partitioning. Communication delays may limit scalability and degrade system performance. In this work, delays are handled to improve the performance of a scheduling policy. The paper demonstrates performance results obtained in simulation under communication and computation overload conditions. The simulation code is later used for prototype implementation. Finally, we examine the software design issues and applicability of the prototype to different distributed environments, providing the example of PVM.