Autonomic Management of Scalable Load-Balancing for Ubiquitous Networks

In ubiquitous networks, a lot of sensors and RFID readers will be connected to the networks as well as PCs and mobile phones are connected, and huge numbers of transactions are expected to be issued by these devices. Since the loads caused by the transactions are getting increased gradually, it is difficult for system manager to estimate the required performance. The scalable load balancing method is, therefore, expected. This report proposes the scalable load-balancing method in which servers communicate each other, and they manage themselves for load-balancing. Because this method does not need a center server, it can avoid the center-server bottleneck. This system, however, is prone to diverge; the system continues load balancing repeatedly in certain configuration and never stops load-balancing operations. In this paper, we clarify a sufficient condition of the divergence, and prevent the system from the divergence.