Dynamic Load-Balancing Based on a Coordinator and Backup Automatic Election in Distributed Systems

摘要

In a distributed system environment it is likely that some nodes are heavily loaded while others are lightly loaded or even idle. It is desirable that the work-load is fully distributed among all nodes so as to utilize the processing time and optimize the whole performance. A load-balancing mechanism decides where to migrate a process and when. This paper introduces the load-balancing mechanism as a new scheme to support the reliability and to increase the overall throughput for distributed systems environment. The idea is to assign one node as a coordinator in addition to a backup node, with the possibility of automatic election in case both coordinator and backup fail. The presented scheme has been integrated into a Zap system. Zap provides a transparent checkpoint-restart mechanism for migrating a PrOcess Domain (POD). A POD provides a group of processes with a private namespace that presents the process group with the same virtualized view of the system. Experimental results show that the load among all nodes is balanced and the freezing time is low compared with other load-balancing mechanisms such as random selection of the destination, unless the number of communication messages needed for migrating a POD becomes high .