| As network applications become more diverse, the requirements that they place on the network infrastructure become more varied and more stringent, particularly with respect to Quality of Service (QoS) requirements. Given this requirement, the network will intuitively yield more desirable service if its nodes have as much knowledge as feasible concerning the requirements of the applications whose traffic is traversing them. This goal can be achieved through mechanisms such as rate allocation, buffer management, congestion control, etc. As a large number of applications are served concurrently in the network, the fruits of this “application awareness” in the network will not be reaped unless it is efficiently scalable with respect to the number of traversing applications.; The objective of this thesis is to investigate “scalable application-aware” router mechanisms. This objective is addressed by discussing two types of mechanisms: buffer management and rate allocation. First, the queueing performance for several existing buffer management schemes is estimated and evaluated using Markov chains (in addition to simulation) with goodput as the metric instead of packet loss. Then, a new buffer management scheme is proposed to combine the efficiency and scalability extracted from existing schemes. The performance of the new scheme is compared with that of the existing schemes.; Second, a new scalable utility-based fair allocation architecture is proposed. This rate allocation architecture collects the utility functions from all flows that traverse a network link and uses them to allocate bandwidth such that all applications have similar utility (or application-level QoS). This is achieve without per-flow state, thereby achieving our goal of scalability for this thesis. |
