Abstract: Delay-sensitive Internet applications are required for high service
quality in IP networks. Unfortunately, the existing load balancing tools are
developed to distribute Internet traffic between routes without considering the
current status of applications and network resources. This results in a
degradation of quality of the applications when these load balancing tools are
implemented. Since the characteristics of Internet applications such as traffic
patterns are uncertain, and the usage of network resources is time-variant, an
adaptive load-balancing tool is needed to cope with the changes in the current
status of the system. In our previous work, we have presented a per-application
load balancing for voice over IP based on a neuro-fuzzy integration. The
previous concept of load balancing is to classify applications with similar
characteristics to the same class and results in achieving the desired targets.
This paper intends to extend upon the previous mechanism in order to apply it
to other delay-sensitive applications. However, since the previous concept and
mechanism are not applicable to general Internet applications, this paper
proposes the idea of load balancing to classify applications with different
characteristics to be possible in the same class provided that the QoS of the
desired application is not degraded. This results in optimizing both the QoS
requirement for the general delay-sensitive applications and the utilization of
network resources, simultaneously.
Keywords: Intelligence adaptive control, hybrid neuro-fuzzy system, adaptive classification, load balancing, and Quality of Service (QoS)
