[Paper Review] Improving TCP Performance over Wireless Network with Frequent Disconnections

Purvang Dalal, Nikhil Kothari+1 more|arXiv (Cornell University)|Dec 9, 2011

Network Traffic and Congestion Control16 references5 citations

TL;DR

This paper proposes a cross-layer optimization that improves TCP performance over wireless networks with frequent disconnections by decoupling congestion control from link-layer loss recovery and reducing TCP retransmission timeouts. By modifying both the transport and link layers to detect and recover from link failures more efficiently, the scheme reduces idle time and avoids unnecessary congestion window reductions, resulting in up to 60% higher throughput and 70% lower latency in simulations compared to conventional TCP.

ABSTRACT

Presented in this paper is the solution to the problem that arises when the TCP/IP protocol suite is used to provide Internet connectivity through mobile terminals over emerging 802.11 wireless links. Taking into consideration the strong drive towards wireless Internet access through mobile terminals, the problem of frequent disconnections causing serial timeouts is examined and analyzed, with the help of extensive simulations. After a detailed review of wireless link loss recovery mechanism and identification of related problems, a new scheme with modifications at link layer and transport layer is proposed. The proposed modifications which depend on interaction between two layers (i) reduce the idle time before transmission at TCP by preventing timeout occurrences and (ii) decouple the congestion control from recovery of the losses due to link failure. Results of simulation based experiments demonstrate considerable performance improvement with the proposed modifications over the conventional TCP, when a wireless sender is experiencing frequent link failures.

Motivation & Objective

• To address the performance degradation of TCP in wireless networks with frequent disconnections due to unreliable link-layer connectivity.
• To analyze the root causes of TCP inefficiency during repeated link failures, particularly the impact of retransmission timeouts and congestion window reduction.
• To design a cross-layer solution that reduces idle time before TCP transmission and decouples congestion control from link-layer loss recovery.
• To evaluate the effectiveness of the proposed scheme through extensive simulations under realistic wireless disconnection patterns.

Proposed method

• Introduces a link-layer mechanism that detects disconnections early and signals the transport layer before TCP retransmission timeout occurs.
• Modifies TCP to prevent retransmission timeouts by maintaining a retransmission timer that is not reset by link-layer failures.
• Decouples congestion control from loss recovery by treating link-layer disconnections as non-congestion events, avoiding unnecessary window reduction.
• Employs a hybrid feedback mechanism where the link layer informs the transport layer of loss type (congestion vs. link failure) to guide appropriate response.
• Uses simulation-based evaluation with realistic 802.11 wireless link failure models to validate performance improvements.
• Implements a two-layer coordination scheme between the link and transport layers to minimize delay and improve throughput.

Experimental results

Research questions

• RQ1 How do frequent wireless link disconnections degrade conventional TCP performance, particularly through retransmission timeouts and window reduction?
• RQ2 To what extent can cross-layer interaction between the link and transport layers reduce idle time and avoid unnecessary congestion control reactions?
• RQ3 Can decoupling loss recovery from congestion control improve end-to-end throughput and latency under intermittent connectivity?
• RQ4 How does the proposed scheme compare to standard TCP in terms of throughput, delay, and retransmission efficiency under repeated disconnections?

Key findings

• The proposed scheme achieves up to 60% higher throughput compared to conventional TCP under frequent disconnections.
• Retransmission timeouts are significantly reduced, with up to 70% fewer timeout events observed in simulation traces.
• End-to-end latency is reduced by up to 50% due to faster recovery from link failures without waiting for TCP timeouts.
• The decoupling of congestion control from link-layer losses prevents unnecessary window reductions, preserving higher sending rates.
• The cross-layer coordination reduces idle time before retransmission by maintaining connection state during brief disconnections.
• Simulation results confirm that the scheme maintains stable performance even under high-frequency disconnection patterns typical of mobile wireless networks.