Adaptive timeout based flow management scheme for per-flow processing in active networks

Abstract

Active networking approach encourages end-users or third party programmers to develop services on top of an active node in order to introduce network evolution from current static networks. The installed service may be instantiated to several flows. In this active networking model with per-flow processing, an efficient resource management scheme is very important. In this thesis, we aim to solve two major problems for flow management and also a minor problem caused by flow management scheme. One of two major problems is that ther is no mechanism for acquisition of generalized timeout value which may be applied to user injected programs. And the other is obtaining a flow replacement algorithm to help the limitation of resources of a node. The minor problem caused by flow management is the loss of flow state when a node unloads running flows. For the problem with timeout, we propose adaptive timeout mechanism, which is a simple algorithm only decreasing a timeout value for certain service and increasing at the occurrence of an unexpected expiration of a flow. For efficient replacement, we also propose two replacement algorithms, MRT and PMRT, both of which consider a remained timeout value for replacement. They show a better hit ratio than that of he LRU replacement algorithm in the case that the number of maximum allowable flows are small. To lessen the effects of the loss of flow state, we propose cache hierarchy, but this is quite general. This thesis porposes an effcient - in average flow population and flow hit ratio - and simple scheme - in computation complexity - for flow management. The flow management is an important issue for per-flow processing. More researches on this subject should be enforces.