Techniques for Enhancing TCP Performance in Wireless Networks

Abstract

Transmission Control Protocol (TCP) carries the maximum Internet traffic, so performance of TCP largely affects the performance of Internet. However, end-to-end throughput in TCP degrades notably when operated in wireless networks, since random packet losses and packet reordering are considered as congestion. In wireless networks, due to high bit error rate and changing level of congestion, retransmission timeouts for packets lost in transmission is unavoidable. TCP misinterprets this loss to congestion and invokes congestion control by triggering fast retransmission and fast recovery, leading to under-utilization of the network resources. This paper presents various performance improvement mechanisms by which TCP does not consider every packet loss as congestion. These mechanisms help TCP to distinguish between congestion and packet loss and increase throughput performance.

How network monitoring and reinforcement learning can improve tcp fairness in wireless multi-hop networks

“…Using a learning rate of α = 1 (1+t) 2 brings down the convergence rate to a polynomial in 1 (1−γ ) , where γ is the discount factor [38]. We use a learning rate of α = 1 (1+t) 2 to ensure that our proposal complies with the dynamic nature of the environment. We have to emphasize that once the Q-learning TCP converges, the agent does not need to re-learn the environment as it has already discovered all the state-action pairs and their associated rewards and can cope with any changes in the environment.…”

“…Different approaches have been proposed in the literature to address the problem of TCP unfairness over wireless mesh networks [2][3][4][5][6][7][8][9][10][11]. The existing literature on TCP fairness solutions over multi-hop wireless networks can be categorized into cross-layer designs e.g [12][13][14][15][16][17][18][19][20][21], and layered proposals, e.g., [22][23][24][25][26][27][28][29][30][31][32][33][34].…”

“…A 0 discount factor prohibits the agent from acquiring future rewards, while a factor of 1 pushes the agent to only consider future rewards. We use a polynomial learning rate α = 1 (1+t) 2 as it has a faster convergence rate [38]. A discount factor of γ = 0.9 is suggested by the literature to encourage the agent into a comprehensive discovery of the (action, state) map [36].…”

“…However, in *Correspondence: nasima@ece.ubc.ca Department of Electrical and Computer Engineering, University of British Columbia, 2356 Main Mall, Vancouver, Canada a wireless multi-hop network, each node has a partial knowledge of the available network topology and creating a unified map of the links and collision domains using feedback messages is extremely costly in terms of overhead. As such, TCP is not capable of a fair resource allocation over WMN [1][2][3][4][5].…”

On the effectiveness of an optimization method for the traffic of TCP-based multiplayer online games

“…Some of these techniques were proposed long ago (some of them are surveyed in [37]), but new techniques are still being proposed, as e.g. [38].…”