Vegas is a new implementation of TCP that achieves between 40 and 70 % better throughput, with one-fifth to onehalf the losses, as compared to the implementation of TCP in the Reno distributionof BSD Unix. This paper motivates and describes the three key techniques employed by Vegas, and presents the results of a comprehensive experimental performance study—using both simulations and measurements on the Internet—of the Vegas and Reno implementations of TCP. 1

Vegas is an implementation of TCP that achieves between 37 and 71 % better throughput on the Internet, with one-fifth to one-half the losses, as compared to the implementation of TCP in the Reno distribution of BSD Unix. This paper motivates and describes the three key techniques employed by Vegas, and presents the results of a comprehensive experimental performance study—using both simulations and measurements on the Internet—of the Vegas and Reno implementations of TCP.

We evaluate the feasibility of using Migratory TCP (M-TCP), a reliable connection-oriented transport layer protocol that supports connection migration, for building highly available Internet services. M-TCP can transparently migrate the server endpoint of a live connection and assists server applications in resuming service on migrated connections. M-TCP provides a generic solution for the problem of service continuity and availability in the face of connectivity failures.

The congestion control mechanisms used in tcp have been the focus of numerous studies and have undergone a number of enhancements. However, even with these enhancements, tcp connections still experience alarmingly high loss rates, especially during times of congestion. To alleviate this problem, the ietf is considering active queue management mechanisms, such as red, for deployment in the network. In this paper, we first show that even with red, loss rates can only be reduced marginally in times of congestion. We then show that the solution to this problem lies in decoupling congestion notification from packet loss through the use of explicit congestion notification (ecn) and in understanding the traffic generated by an aggregation of a large number of sources. Given this, we then propose and experiment with more adaptive active queue management algorithms (Adaptive red) and more conservative end-host mechanisms (subtcp) which can significantly reduce loss rates across conges...

Wide-area distributed applications are frequently limited by the performance of Internet data transfer. We argue that the principle cause of this effect is the poor interaction between host-centric congestion control algorithms and the realities of today's Internet traffic and infrastructure. In particular, when the duration of a network flow is short, then using end-to-end feedback to determine network conditions will be extremely inefficient. We propose an incremental approach to the problem, in which congestion information is shared among many co-located hosts and transport protocols make informed congestion control decisions. We argue that the resulting system can potentially improve the performance experienced by each network user as well as the overall efficiency of the network. 1 Introduction While there has been considerable research invested in mitigating the effects of inadequate memory performance[22, 5], disk performance [23, 24], and LAN performance [34], for the growing ...

With the rapid proliferation of the WWW, the Internet has seen an enormous growth in both the demand for access from its users and in the demand for new services from its applications. One of the primary reasons for the Internet's success is TCP and its congestion control algorithms which allow the network to deliver service in times of extreme demand. This dissertation focuses on these two, extremely important challenges to today's Internet: supporting an explosion in the number of users and supporting a myriad of new applications which require more out of the network than the best-effort service that the Internet currently provides. The first half focuses on the problem of maximizing network efficiency in times of extremely heavy congestion. In particular, a number of significant weaknesses in TCP's congestion control and current active queue management algorithms (such as RED) are uncovered. These weaknesses are the key reasons why an alarming rise in packet loss rates has been observed across a number of network links. With additional growth in demand, these weaknesses canseverely degrade network efficiency. Given this, several new queue management (Adaptive RED, Blue) and congestion control algorithms (SubTCP) are introduced and analyzed. These algorithms are shown to eliminate packet loss and maximize link utilization effectively even in the presence of an extremely large number of competing sources. In addition to packet loss, another problem with TCP is the fact that it only provides applications with a single service, best-effort. As the need for new services has grown, the inability to provide service differentiation across flows beyond best-effort service has become problematic. The second half addresses the problem of supporting additional services across the Internet. In particular, these chapters focus on building scalable, deployable mechanisms for supporting bandwidth guarantees across the Internet based on priority mechanisms as outlined by the Differentiated Services working group of the IETF. Using minor extensions to the queue management (Enhanced RED) and congestion control algorithms in place today, these mechanisms provide applications more predictable behavior out of the network and enable the deployment of a large number of bandwidth-sensitive applications.

The main goal of a congestion avoidance algorithm is to maximize throughput and minimize delay (Jain & Ramakrishnan 1988). While TCP Reno achieves high throughput, it tends to consume all of the buffer space at the bottleneck router, causing large delays. In this paper we propose a simple scheme that modifies TCP Reno's congestion avoidance algorithm by throttling back the opening of the congestion window once an increase in round-trip time is perceived. We call the scheme TCP-BFA and have implemented it in the ns network simulator and in BSD 4.4. We show through simulations and measurements of real traffic on the Internet that TCP-BFA results in lower router buffer occupancies and lower delays while maintaining a throughput similar to that of TCP Reno. The advantages of TCP-BFA are (1) smaller router buffer size requirements, (2) an order of magnitude improvement in network power (the ratio of throughput to delay), (3) fewer packet losses, (4) faster detection of multiple losses due to lower retransmission timeout estimates, and (5) smoother traffic patterns.

OF THE THESIS MTCP : Transport Layer Support for Highly Available Network Services by Kiran Srinivasan Thesis Director: Liviu Iftode We propose a transport layer protocol designed for building highly available Internet network services. The transport layer protocol would provide a mechanism to transparently migrate the server endpoint of a live connection. The connection migration is dynamic and can happen multiple times during a client-service session. Our migration mechanism is client-initiated, integrated in a migration architecture that uniformly addresses various types of events that affect the quality of service perceived by the client. In the architecture, the migration mechanism is independent and decoupled from any migration policies. The migration mechanism can be applied for different contexts like server load balancing, to provide fault tolerance across server crashes, to improve the performance as perceived by the client etc. We examine and address a number of challenges posed by the transfer of connection state required for a connection migration to both the OS and the application layer. We describe the details of our design and an implementation, as well as experimental data that suggests the usefulness of this approach.

This paper studies TCP performance over static, ad hoc networks that use IEEE 802.11 protocol as the access method. Our study reveals some interesting results. First, there exists an optimal value for TCP congestion window size, at which the TCP throughput is maximized. However, TCP does not operate around this optimal point, and typically grows its window much larger; this leads to decreased throughput and increased packet loss. To better understand this behavior, we further study the characteristics of TCP packet loss. Our results show that, network overload is mainly signified by wireless link contention. As long as the buffer size at each node is reasonable (larger than 10 packets), buffer overflow-induced packet loss is rare and packet drops due to link-layer contention dominate. Link-layer drops offer the first sign for network overload. We further observe that multihop wireless links collectively demonstrate Random Early Detection (RED) like graceful drop behavior, but the current TCP protocol does not adapt well to this built-in grace drop characteristic. We further propose two techniques of link RED and adaptive spacing at the link layer, and simulations show that they can improve TCP throughput by 5% to 30%.

In this paper, we present an experimental transport protocol which employs alternative methods of startup and congestion avoidance to those commonly used in implementations of TCP. Through modelling and simulation, we demonstrate how a packet-pair probe may be used to replace traditional slow-start methods and how proactive congestion avoidance algorithms provide smoother trac ow with lower levels of packet loss and retransmission. We also show how a token bucket may be used to allow bounded burstiness for an Internet connection in order to facilitate an application's Quality of Service. 1 Overview STTP is an experimental transport protocol designed to operate primarily over traditional IP-based Internetworks. It is a Shaped (through use of leaky and token bucket trac shapers), Token-based (packet acknowledgements are treated as tokens for use in its internal shaping algorithms), Transport Protocol (STTP). STTP is designed as an alternative for the reliable transport transport...