Application of soft computing techniques to adaptive user buffer overflow control on the Internet

Abstract

Two novel expert dynamic buffer tuners/controllers, namely, the neural network controller (NNC) and the fuzzy logic controller (FLC) are proposed in this paper. They use soft computing techniques to eliminate buffer overflow at the user/server level. As a result they help shorten the end-to-end service roundtrip time (RTT) of the logical Internet transmission control protocol (TCP) channels. The tuners achieve their goal by maintaining the given safety margin Delta around the reference point of the {0,Delta}2 objective function. Overflow prevention at the Internet system level, which includes the logical channels and their underlying activities, cannot shorten the service RTT alone. In reality, unpredictable incoming request rates and/or traffic patterns could still cause user-level overflow. The client/server interaction over a logical channel is usually an asymmetric rendezvous, with one server serving many clients. A sudden influx of simultaneous requests from these clients easily inundates the server's buffer, causing overflow. If this occurs only after the system has employed expensive throttling and overflow management resources, the delayed overflow rectification could lead to serious consequences. Therefore, it makes sense to deploy an independent user-level overflow control mechanism to complement the preventative effort by the system. Together they form a unified solution to effectively stifle channel buffer overflow.