Applications, such as grid computing, remote backup, and contents distribution, transfer bulk data over wide-area networks. These applications schedule work to satisfy the user requirements. Guaranteeing the performance of data transfer is helpful for scheduling these applications. Diffserv AF is a QoS control method that guarantees data transfer performance. However, the throughput of conventional TCPs is much lower than the reserved bandwidth in congested Diffserv AF networks. Some high-speed protocols reduce the difference between the reserved bandwidth and the average throughput. However, other traffic in the networks influences the throughput of these high-speed protocols. The insufficient throughput stability exists when the background traffic throughput frequently changes in congested networks. Our purpose is to improve the network bandwidth utilization and to make it possible to reserve more bandwidth in Diffserv AF networks. We propose a data transfer method that uses two kinds of TCP flows. One TCP flow reserves network bandwidth and controls the window size to reduce the difference between the throughput and the reservation bandwidth. The other TCP flow transfers data without bandwidth reservation and flow control. This method allows for a higher throughput when the network is uncongested and reduces the difference between the reserved bandwidth and the throughput in congested networks.
This paper presents a Powerpoint presentation discussing a method that reduces the reserved bandwidth of large-scale data transfers over Diffserv AF networks.
