Dynamic Load-Balancing for Distributed Management in Active Networks

Abstract

As a solution to the lack of scalability of the centralized management paradigm,the di tributed management paradigm ha been introduced. In the paradigm,on the basis of active node in active networks, management script are downloaded to and executed on managed systems. The paradigm,however,contain a eriou and un olved issue that ill-balanced of system load over a network is cau ed by unexpected amount of the management script added on the systems’ own task. Therefore,considering network resource utilization,how to distribute and place the management script in an appropriate manner in a significant issue to be studied in order to hold the balance of the management task in the overall network. For a solution to the issue, we propose a new dynamic load-balancing algorithm for distributed management in active networks. The decision on which ystems to execute the management script is dynamically made on the basis of the deviation from the average CPU utilization of all the systems and the bandwidth needed for executing all the management scripts. We theoretically how how to find optimal value for a tolerable deviation and a maximum tolerable bandwidth for the better load-balancing. We evaluate the proposed algorithm by applying it to an operational LAN. The results how that the proposed algorithm perform so well with a trivial overhead that it can hold the balance among management task in an overall network. The proposed algorithm could be one of the essential techniques enabling the distributed management paradigm more promising.