Performance of Application-Specific Buffering Schemes for Active Networks

As the cost of computing power decreases and network traffic patterns become more complex, it becomes worthwhile to consider the benefits of allowing users to specify policies for managing their traffic within the network. Active networking is a new design paradigm in which the network is architected not merely to forward packets, but also to be dynamically programmed in order to support per-user services. Active networks export uniform "meta-level" interfaces that expose network-level resources, policies and mechanisms to its users. In this work, we present a new scheme for managing network buffer space, aggregate application level buffering, in which a group of flows of similar types are managed by an active node as a single entity. Thus, resources within the network are provisioned on a "per application" basis rather than a "per user" basis. We justify this scheme by comparing it to previously accepted buffering schemes such as a single shared buffer, per flow queueing, and an extension to per flow queueing which accommodates application specific packet dependencies. Within this work, we justify the benefits of this scheme in terms of required computational processing, information maintained by a node within a network, goodput achieved per application type, and mean delay. Based on the work herein, we show that the impact of an aggregate buffer management scheme can be significant depending on the interdependencies which may exist within an application's traffic stream