Fast Retransmission Research Articles

Modeling TCP throughput

In this paper we develop a simple analytic characterization of the steady state throughput, as a function of loss rate and round trip time for a bulk transfer TCP flow, i.e., a flow with an unlimited amount of data to send. Unlike the models in [6, 7, 10], our model captures not only the behavior of TCP's fast retransmit mechanism (which is also considered in [6, 7, 10]) but also the effect of TCP's timeout mechanism on throughput. Our measurements suggest that this latter behavior is important from a modeling perspective, as almost all of our TCP traces contained more time-out events than fast retransmit events. Our measurements demonstrate that our model is able to more accurately predict TCP throughput and is accurate over a wider range of loss rates.

Improving the performance of TCP over the ATM-UBR service

In this paper we study the design issues for improving TCP performance over the ATM UBR service. ATM-UBR switches respond to congestion by dropping cells when their buffers become full. TCP connections running over UBR can experience low throughput and high unfairness. Intelligent switch drop policies and end-system policies can improve the performance of TCP over UBR with limited buffers. We describe the various design options available to the network as well as to the end systems to improve TCP performance over UBR. We study the effects of early packet discard, and present a per-VC accounting-based buffer management policy. We analyze the performance of the buffer management policies with various TCP end system congestion control policies, including slow start and congestion avoidance, fast retransmit and recovery and selective acknowledgments. We present simulation results for various small and large latency configurations with varying buffer sizes and number of sources.

Comparative performance analysis of versions of TCP in a local network with a lossy link

We use a stochastic model to study the throughput performance of various transport control protocol (TCP) versions (Tahoe (including its older version that we call OldTahoe), Reno, and NewReno) in the presence of random losses on a wireless link in a local network. We model the cyclic evolution of TCP, each cycle starting at the epoch at which recovery starts from the losses in the previous cycle. TCP throughput is computed as the reward rate in a certain Markov renewal-reward process. Our model allows us to study the performance implications of various protocol features, such as fast retransmit and fast recovery. We show the impact of coarse timeouts. In the local network environment the key issue is to avoid a coarse timeout after a loss occurs. We show the effect of reducing the number of duplicate acknowledgements (ACKs) for triggering a fast retransmit. A large coarse timeout granularity seriously affects the performance of TCP, and the various protocol versions differ in their ability to avoid a coarse timeout when random loss occurs; we quantify these differences. We show that, for large packet-loss probabilities, TCP-Reno performs no better, or worse, than TCP-Tahoe. TCP-NewReno is a considerable improvement over TCP-Tahoe, and reducing the fast-retransmit threshold from three to one yields a large gain in throughput; this is similar to one of the modifications in the TCP-Vegas proposal. We explain some of these observations in terms of the variation of fast-recovery probabilities with packet-loss probability. The results of our analysis compare well with a simulation that uses actual TCP code.

Improving the start-up behavior of a congestion control scheme for TCP

Based on experiments conducted in a network simulator and over real networks, this paper proposes changes to the congestion control scheme in current TCP implementations to improve its behavior during the start-up period of a TCP connection.The scheme, which includes Slow-start, Fast Retransmit, and Fast Recovery algorithms, uses acknowledgments from a receiver to dynamically calculate reasonable operating values for a sender's TCP parameters governing when and how much a sender can pump into the network. During the start-up period, because a TCP sender starts with default parameters, it often ends up sending too many packets and too fast, leading to multiple losses of packets from the same window. This paper shows that recovery from losses during this start-up period is often unnecessarily time-consuming.In particular, using the current Fast Retransmit algorithm, when multiple packets in the same window are lost, only one of the packet losses may be recovered by each Fast Retransmit; the rest are often recovered by Slow-start after a usually lengthy retransmission timeout. Thus, this paper proposes changes to the Fast Retransmit algorithm so that it can quickly recover from multiple packet losses without waiting unnecessarily for the timeout. These changes, tested in the simulator and on the real networks, show significant performance improvements, especially for short TCP transfers. The paper also proposes other changes to help minimize the number of packets lost during the start-up period.

Improving the performance of reliable transport protocols in mobile computing environments

We explore the performance of reliable data communication in mobile computing environments. Motion across wireless cell boundaries causes increased delays and packet losses while the network learns how to route data to a host's new location. Reliable transport protocols like TCP interpret these delays and losses as signs of network congestion. They consequently throttle their transmissions, further degrading performance. We quantify this degradation through measurements of protocol behavior in a wireless networking testbed. We show how current TCP implementations introduce unacceptably long pauses in communication during cellular handoffs (800 ms and longer), and propose an end-to-end fast retransmission scheme that can reduce these pauses to levels more suitable for human interaction (200 ms). Our work makes clear the need for reliable transport protocols to differentiate between motion-related and congestion-related packet losses and suggests how to adapt these protocols to perform better in mobile computing environments.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>