摘要：A link layer protocol based on SR (Selective-Repeat) ARQ (Automatic Repeat reQuest) is required to achieve high performance over a lossy and large delay link. In spite of its effectiveness, SR ARQ has problems of large resequencing delay and bursty packet output due to the resequencing function. To mitigate these problems, this paper proposes a scheme where the resequencing is not performed by a receiver of SR ARQ, but by TCP end to end. The proposed scheme has advantages of scalability with regard to link bandwidth, the number of TCP connections on the link, and the size of the TCP window. It also preserves the end-to-end semantics of TCP and requires no modification to the existing TCP. To suppress unnecessary retransmissions by TCP, a sender of SR ARQ is aware of retransmitted TCP data packets and drops duplicate ACKs due to out-of-order packets, which are caused by the lack of resequencing by SR ARQ. The effectiveness of this proposed scheme is evaluated by simulations, which show that it attains high performance in comparison with other schemes.
Loading...
