Bandwidth Allocation to Interactive Users in DBS-Based Hybrid Internet

M. Stagarescu

Masters Dissertation, Number: CSHCN MS 1998-3, Year: 1998, Advisor: John S. Baras

Abstract

We are motivated by the problem of bandwidth allocation to Internet usersin DBS-based Hybrid Internet, where the Network Operations Center (NOC)-scheduler controls the amount of service provided to each user, by using packet scheduling and buffer management. Such a system exploits the ability of satellites to offer high bandwidth connections to large geographical areas, and it delivers low-cost hybrid (satellite-terrestrial) high-speed services to interactiveInternet users. In this system, it is important to reduce the delay thatusers experience.We analyze several bandwidth allocation policies at the NetworkOperations Center (NOC) of a DBS-based hybrid Internet network.We consider the problem of optimal scheduling of the services of interactive users in the DBS-based hybrid Internet configuration.We show that, for the interactive Internet users, the Most Delayed Queue First (MDQF) policy, which serves the queues starting with the most delayed queue, is providing the minimum delay when compared with the Equal Bandwidth (EB) and Fair Share (FS) allocation policies. The MDQF policy is shown to be optimal with respect to a performancemetric of packet loss due to queuing time constraints.The impact of the scheduling policies on the Hybrid Internet system'sperformance is analyzed in the context of the interplay between the NOC queuing system (and its bandwidth allocation policies)and the underlying transport protocol (TCP), and we show the effectiveness of the MDQF policy in the presence of the TCP congestion control algorithm.We also present simulations in which the Internet server sources sendself-similar ("fractal") data traffic to the NOC-scheduler.The results confirm our calculations, that the MDQF policy is a better performing policy for minimizing the mean delay at high load factors, when comparing it to the EB and FS policies. Finally, we propose two solutions, a buffer allocation policy and a "virtual delay" mechanism, which make the MDQF policy work in the presence of greedy sources.