-The transmission control protocol (TCP) has been designed to provide reliable transport of packets by adjusting the transmission rate to the network congestion level. While TCP can adapt to small fluctuations in the delay between the sender and the receiver, adverse affects (most importantly spurious timeouts) have been observed under large delay variability. In this paper, we exhibit the presence of such delay spikes in wireless networks and discuss their possible origins. We then investigate a new methodology for avoiding spurious TCP timeouts by appropriately injecting additional random delay along the communication path. Different algorithms for the delay injection are presented and we assess their relative performances and merits through simulations. In particular we show by numerical examples that the delay injection methodology can significantly decrease the number of timeouts and increase the achieved TCP throughput by about 8% in the network scenario considered in this paper. One of the attractive features of the new methodology is that it does not require any changes to the TCP protocol and can be applied independently of the TCP version used.

No abstract

A new transport layer protocol called TCP-Swift is proposed for enhancing the TCP performance over satellite /P networks. TCP-Swift replaces the conventional TCP slow start and fast recovery algorithms by speedy start and speedy recovery. With speedy start, a TCP-Swift sender opens up its congestion window in only two round trip times. This signiJicantly shortens the tinie needed in probing the network for equilibrium state. With speedy recovery, we can infer the cause of a packet loss by observing the ACK stream received at the sender. lfthe loss is due to wireless transmission error, the sender's congestion window can be reopened up more aggressively to fully utilize the available satellite link bandwidth. We show that TCP-Swift outperforms existing TCP schemes by simulations.

Focusing on a general wireless network where a wireless link can be at any link along the sender-toreceiver path, a new TCP enhancement scheme, called Generalized-Snoop (G-Snoop), is proposed. Since many existing applications are built on top of TCP, it is essential that any TCP enhancement scheme should be transparent to the end-systems as well as the faed networks. To achieve this;G-Snoop only needs to be implemented at the wireless gateways, no other parts of the network require modifications. With G-Snoop, TCP senders are shielded from non-congestion packet loss and thus no unnecessary congestion control mechanisms will be performed. Simulation results show that significant throughout gain can be obtained with G-Snoop.

No abstract