TCP is a popular transport protocol used in present-day internet. When packet losses occur, TCP assumes that the packet losses are due to congestion, and responds by reducing its congestion window. When a TCP connection traverses a wireless link, a significant fraction of packet losses may occur due to transmission errors. TCP responds to such losses also by reducing congestion window. This results in unnecessary degradation in TCP performance. We define a class of functions named loss predictors which may be used by a TCP sender to guess the actual cause of a packet loss (congestion or transmission error) and take appropriate actions. These loss predictors use simple statistics on round-trip times and/or throughput, to determine the cause of a packet loss. We investigate their ability to determine the cause of a packet loss. Unfortunately, our simulation measurements suggest that the three loss predictors do not perform too well. 1. Introduction TCP is a popular protocol for reliabl...

The success of the Internet can be attributed directly to the large number of useful applications running on it. TCP/IP has been the underlying communication protocols enabling these applications. Despite the widespread deployment of the protocols, TCP/IP continues to be a hot research topic among R&D teams in the academia as well as the industry. Much of the recent work has been centered on adapting the protocols for use in wireless environments because of its ease of deployment, convenience to users, and high bandwidth brought about by maturing of advanced technologies (e.g., 3G and IEEE802.11 LAN). The next big wave -- mobile Internet, in which wireless will play a necessary part of the Internet infrastructure -- is around the corner. The original TCP design assumes packet loss is always induced by congestion. This assumption can lead to significant performance degradation in wireless networks where random loss due to environmental noise is rampant. The detrimental effect of random loss on TCP performance is a well-known outstanding problem that remains to be unsolved. This dissertation proposes and studies a novel end-to-end congestion control mechanism called TCP Veno that is able to deal with random loss effectively. Veno differs from the conventional TCP in a major way: it monitors the congestion level using an end-to-end estimation algorithm and uses that knowledge to refine the congestion control algorithm of TCP. Specifically, 1) it refines the multiplicative decrease algorithm of Reno by setting the congestion threshold -- a key control parameter in TCP - according to the perceived congestion level of the network instead of a fixed drop factor when packet loss is detected; 2) it vi refines the linear increase algorithm by attempting to stay longer in an op...

Keywords: Wireless network, explicit congestion notification, TCP/IP, congestion control Abstract: TCP was designed for wireline networks, where loss events are mostly caused by network congestion. The congestion control mechanism of current TCP uses loss events as the indicator of congestion, and reduces its congestion window size. However, when a lossy link is involved in a TCP connection, non-congestion random losses should also be considered. The congestion window size should not be decreased if a loss event is caused by link corruptions. To improve TCP performance over lossy links, in this paper, we first present that zero congestion loss could be achieved by appropriately setting the ECN marking threshold in the RED buffer. Secondly, we propose a new TCP algorithm, called Differentiation Capable TCP (Diff-C-TCP). Diff-C-TCP makes an assumption that packet losses are caused by link corruptions, and uses ECN (Explicit Congestion Notification) to determine any loss that may occasionally happen due to network congestion. We have shown that Diff-C-TCP performs very well in the presence of a lossy link. 1

In the context of TCP, several researchers have proposed heuristics to detect or predict congestion in the network. In this paper, the term congestion predictors refers to such heuristics. Past proposals require TCP sender to reduce its window size when congestion is detected or predicted (otherwise, the heuristic may dictate that the sender window be held constant or increased). The proposed heuristics to detect/predict congestion typically use simple statistics on observed round-trip times and/or observed throughput. The primary objective of this paper is to investigate the ability of the congestion predictors to predict a packet loss. Our measurements indicate that the three congestion predictors studied in this paper are often poor in their ability to predict a packet loss due to congestion. To arrive at this conclusion we measure the frequency with which the predictors predict congestion, and how often they predict congestion just before a packet loss. A study of the variations in...

An asymmetric network is one in which there is a mismatch between the characteristics of the forward and reverse channel. Typically, the forward channel is capable of high transmission rates and the reverse channel has lesser bandwidth. Cable modem networks and satellite networks are some technologies that exhibit network asymmetry. TCP, due to its widespread acceptance, is also used in asymmetric networks. However, the reverse channel being a bottleneck causes many problems in the behavior of traditional TCP protocols. The TCP protocol is heavily dependent on ACKs. The variation in the ACK arrival rate, caused due to network asymmetry, results in slower initial window growth and burstiness on the forward channel. Existing approaches to improve the performance of TCP in asymmetric networks use techniques such as reduction in number of ACKs. However, these approaches do not take into consideration the high bandwidth availability in the forward direction for early ramp up of TCP window. ...

draft-ietf-pilc-error-00.txt This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Comments should be submitted to the PILC mailing list at

this document, we survey the different solutions available or under investigation, and issue the corresponding recommendations. The subsequent sections include solutions: unrelated to IP built on top of IP built on top of UDP built on top of TCP (modifying TCP) The latter solutions constitute the most numerous group, as it is very desirable to make TCP, the most widely used protocol in the internet, perform well over wireless networks. There is a large body of work on the subject of improving TCP performance over satellite links. The documents that the tcpsat working group of the IETF is working on [AG98, AGGHSSTT98] are very relevant. In both cases, it is essential to start by improving the characteristics of the medium by using forward error correction (FEC) at the link layer to reduce the BER (bit error rate) from values as high as 10

This document discusses the specific TCP mechanisms that are problematic in environments with high uncorrected error rates, and discusses what can be done to mitigate the problems without introducing intermediate devices into the connection.

Transmission control protocol (TCP), the widely used transport protocol in the Internet, assumes that packet losses are because of congestion. Wireless networks have higher error rates than wired networks; TCP misinterprets the error losses as congestion and wrongly invokes congestion control algorithms. In the proposed scheme, we show that by using congestion indication feedback from the network, performance of TCP can be improved in network paths containing wireless links. The routers in the network detect congestion and set a congestion indication bit on packets owing in the forward direction. The congestion indication is communicated back to the users through the transport-level acknowledgement. When the sender encounters a packet loss, explicit congestion indication feedback received for the packets sent before and after the packet dropped are used to identify the state of the network at the time of drop. If the network is identi ed as not congested when the packet was lost and the recent congestion indication feedback is also low, then the packet is considered to be lost because of transmission error. We compared our scheme against TCP-Reno which assumes that all packet losses are because of congestion. Under low congestion in the network, our scheme can lead to signi cant throughput improvement.

Abstract not found