부산대학교 도서관

UDP based Data Transfer (UDT) is a representative UDP based high speed transfer protocol which aims to efficiently utilize the available bandwidth in long fat pipe network. It estimates the Maximum Transfer Rate (MTR) with the arriving interval periodically for pair packets on receiver side, and then calculate the available bandwidth with the difference between MTR and the current transfer rate on sender side. Here, the major concern is to estimate MTR in order to exactly determine the available bandwidth. However, it is general to estimate the available bandwidth much greater than that of actual value, eventually it brings about a fairness problem. This situation mainly comes from the MTR estimation algorithm which assigns a fixed weight to the arriving interval to calculate the available bandwidth. Especially it is getting to be much difficult to determine an exact available bandwidth when network is congested. So, it is required an MTR calculation adaptive to network congestion. This thesis proposes a congestion degree based available bandwidth estimation method with reflecting current network congestion status. Here, the congestion degree means a relative index of current congestion status, and has a range from 0 to 1. To measure the degree, sender side records the pair packet arriving intervals when ACKs arrive from receiver side, and traces the maximum and minimum intervals. Then it compares them with the current interval to assess the network load. Finally, the current MTR estimation is taken place to calculate the available bandwidth adaptive to network traffic load. Here, MTR is calculated by subdividing estimation step according to the threshold. With the proposed method, fairness problem with considering multiple UDT flows can naturally be resolved by mitigating the aggressiveness of UDT. With a network simulator NS-2, the proposed method has been simulated the throughput in comparison with TCP, UDT, in terms of the link delay and the number of flows. It shown that the fairness problem can be significantly improved compared with that of UDT. Our method can be usefully applied to other protocols which require the fairness constraint.