Developing and Analysing Dynamic Resource Management Algorithms with a Generic Framework
Joshua D. King · 2005
Distributed systems need to share tasks efficiently to exploit their performance. A combination of process migration and dynamic resource management algo-rithms can enable this. We investigate one such migration technique and five algorithms in order to create a generic resource management framework for dis-tributed systems. We use openMosix’s transparent process migration under Linux for experimentation. Experiments that were run on 4-, 8- and 12-node clusters showed that per-formance depends heavily upon process migration, and less so on the algorithms. Some algorithms also monitor multiple resources to improve performance. Scal-ability of the framework is an important consideration, and we found that even with distributed control a completely connected network would soon become un-manageable as the number of nodes is increased. We find that the opportunity cost algorithms are the most consistent but upon consideration of scalability sug-gest that simpler algorithms, such as the distributed MINIX load balancer, are desirable.