Proposal on dynamic load balancing algorithm in distributed network management

摘要

It is revealed that the centralized management paradigm, such as SNMP, does not scale, since management tasks are concentrated on a centralized management system (a manager), and the increasing number of managed systems (agents) results in bandwidth consumption for polling. As an alternative, distributed management paradigm has been proposed, in which management scripts are downloaded to and executed on agents. In this case, however, it could be identified as a problem that a heavy processing load might be imposed on the specific agents and the paradigm cannot necessarily strike a balance of the management tasks in overall the network, since the management scripts are downloaded regardless of the agents' processing load. In this paper, to solve the problem, we propose a new solution by a dynamic load balancing algorithm, taking account of both the processing load of managers/agents and the bandwidth for polling during the execution of the management scripts. Finally, we implement the proposed algorithm and evaluate it in an operational network.