RIA: Adaptive Load Balancing for Heterogenous Distributed Systems

点击翻译按钮获取中文摘要

Chaudhary This project explores a new scheme for achieving load balancing in a heterogeneous distributed computing system. With the advent of optical links it is now feasible to migrate tasks from one computer to a remote computer. This scheme is being implemented and tested using real-world problems in a real heterogeneous distributed computing environment. A salient feature of the scheme is its semi-distributed, adaptive nature, i.e., the parameters taken into consideration and the algorithm for load balancing itself change during run time. The algorithm is stable, i.e., no processes are allowed to migrate without making any progress. No assumptions are made about the dependency relations of the various tasks. Various job classes with precedence constraints are allowed. Other important real- world constraints such as the inability of a process to execute on particular computers, a process needing more than one computer simultaneously for its execution, a set of processes needing to be executed on a particular subset of computers, etc., are also incorporated into the scheme. The scheme starts with an initial static load balance and then evolves into an adaptive algorithm.