Sung-hyuck Lee

Unfairness of assured service and a rate adaptive marking strategy

Assured Service, which is a service model of the Internet Differentiated Services (DiffServ) architecture is not currently well implemented on the Internet, mainly because TCP employs an AIMD (Additive Increase and Multiplicative decrease) mechanism to control congestion. In this paper, we analyze the problem and describe a marking rule, called RAM (Rate Adaptive Marking), which diminishes the problem. The RAM method marks sending packets, at a source node or at an edge router, in inverse proportion to throughput gain and in proportion to the reservation rate and throughput dynamics. We also discuss methods to implement the RAM strategy. Several experiments with the ns- 2 simulator are performed to evaluate the RAM scheme.

A Dynamic Marking Scheme of Assured Service for Alleviating Unfairness among Service Level Agreements

Assured Service, which is a service model of the Internet Differentiated Services (DiffServ) architecture, is not currently well implemented on the Internet, mainly because TCP employs an AIMD (Additive Increase and Multiplicative Decrease) mechanism to control congestion. Many studies have shown that the current Assured Service model does not assure the target rate of high profile flows in the presence of numerous low profile flows, and does not equably distribute the reservation rate of an SLA among multiple flows included in the SLA. In this paper, the former problem is referred to as inter-SLA unfairness and the latter problem is referred to as intra-SLA unfairness. We propose a marking rule, called RAM (Rate Adaptive Marking) that simultaneously diminishes both of these unfairness problems. The RAM method marks sending packets, at a source node or at an edge router, in inverse proportion to throughput gain and in proportion to the reservation rate and throughput dynamics. Three experiments with the ns-2 simulator are performed to evaluate the RAM scheme. The simulation results show that the RAM scheme may significantly reduce inter-SLA unfairness and the intra-SLA unfairness.