Abstract-This paper presents Random Early Detection (RED) gateways for congestion avoidance in packet-switched networks. The gateway detects incipient congestion by computing the average queue size. The gateway could notify connections of congestion either by dropping packets arriving at the gateway or by setting a bit in packet headers. When the average queue size exceeds a preset threshold, the gateway drops or marks each arriving packet with a certain probability, where the exact probability is a function of the average queue size. RED gateways keep the average queue size low while allowing occasional bursts of packets in the queue. During congestion, the probability that the gateway notifies a particular connection to reduce its window is roughly proportional to that connection's share of the bandwidth through the gateway. RED gateways are designed to accompany a transport-layer congestion control protocol such as TCP. The RED gateway has no bias against bursty traffic and avoids the global synchronization of many connections decreasing their window at the same time. Simulations of a TCP/IP network are used to illustrate the performance of RED gateways. lNTRODucTION I N high-speed ne!works with connections with large delay bandwidth products. gateways are likely to be designed with correspondingly large maximum queues to acccomodate transient congestion. In the current Internet, the TCP transport protocol detects congestion only after a packet has been dropped al the gateway. However, it would clearly be undesirable to have large queues (possibly on the order of a delay bandwidth product) that were full much of the time; this would significantly increase the average delay in the network. Therefore, with increasingly high-speed networks, it is increasingly important to have mechanisms that keep throughput high but avemge queue sizes low. In the absence of explicit feedback from the gateway, there are a number of mechanisms that have been proposed for transport-layer protocols to maintain high throughput and low delay in the network. Some of these proposed mechanisms are designed to work with current gateways [15] could infer congestion from the estimated bottleneck service time or from changes in throughput or end-to-end delay, as well as from packet drops or other methods. Nevertheless, the view of an individual connection is limited by the time scales of the connection, the traffic pattern of the connection, the lack of knowledge of the number of congested gateways, the possibilities of routing changes, or other difficulties in distinguishing propagation delay from persistent queueing delay. The most effective detection of congestion can occur in the gateway itself. The gateway can reliably distinguish between propagation delay and persistent queueing delay. Only the gateway has a unified view of the queueing behavior over time; the perspective of individual connections is limited by the packet arrival patterns for those connections. In addition, a gateway is shared by many active connections with a wide range of round-trip times, tolerances of delay, throughput requirements, etc. Decisions about the duration and magnitude of transient congestion to be allowed at the gateway are best made by the gateway itself. The method of monitoring the average queue size at the gateway, and of notifying connections of incipient congestion, is based on the assumption that it will continue to be useful to have queues at the gateway where traffic from a number of connections is multiplexed together with FIFO scheduling. Not only is FIFO scheduling useful for sharing delay among connections and reducing delay for a particular connection during its periods of burstiness [4], but it scales well and is easy to implement efficiently. In an alternate approach, some congestion control mechanisms that use variants of Fair Queueing [20] or hop-by-hop flow control schemes [22] propose that the gateway scheduling algorithm make use of per-connection state for every active connection. We would suggest instead that per-connection gateway mechanisms be used only in those circumstances where gateway scheduling mechanisms without per-connection mechanisms are clearly inadequate. The DECbit congestion avoidance scheme [ 18], described later in this paper, is an early example of congestion detection at the gateway; DECbit gateways give explicit feedback when the average queue size exceeds a certain threshold. This paper proposes a different congestion avoidance mechanism at the gateway, RED (Random Early Detection) gateways. with somewhat different methods for detecting congestion and choosing which connections to notify of this congestion. While the principles behind RED gateways are fairly general and RED gateways can be useful in controlling the average queue size even in a network where the transport protocol 1063-6692/93$03.00" O 199.7 IEEE

We propose an optimization approach to flow control where the objective is to maximize the aggregate source utility over their transmission rates. We view network links and sources as processors of a distributed computation system to solve the dual problem using gradient projection algorithm. In this system sources select transmission rates that maximize their own benefits, utility minus bandwidth cost, and network links adjust bandwidth prices to coordinate the sources' decisions. We allow feedback delays to be different, substantial and time-varying, and links and sources to update at different times and with different frequencies. We provide asynchronous distributed algorithms and prove their convergence in a static environment. We present measurements obtained from a preliminary prototype to illustrate the convergence of the algorithm in a slowly time-varying environment.

This paper discusses the use of Explicit Congestion Notification (ECN) mechanisms in the TCP/IP protocol. The first part proposes new guidelines for TCP’s response to ECN mechanisms (e.g., Source Quench packets, ECN fields in packet headers). Next, using simulations, we explore the benefits and drawbacks of ECN in TCP/IP networks. Our simulations use RED gateways modified to set an ECN bit in the IP packet header as an indication of congestion, with Reno-style TCP modified to respond to ECN as well as to packet drops as indications of congestion. The simulations show that one advantage of ECN mechanisms is in avoiding unnecessary packet drops, and therefore avoiding unnecessary delay for packets from low-bandwidth delay-sensitive TCP connections. A second advantage of ECN mechanisms is in networks (generally LANs) where the effectiveness of TCP retransmit timers is limited by the coarse granularity of the TCP clock. The paper also discusses some implementation issues concerning specific ECN mechanisms in TCP/IP networks.

Abstract-This paper examines the performance of TCP/IP, the Internet data transport protocol, over wide-area networks (WANs) in which data traffic could coexist with real-time traffic such as voice and video. Specifically, we attempt to develop a basic understanding, using analysis and simulation, of the properties of TCP/IP in a regime where: 1) the bandwidth-delay product of the network is high compared to the buffering in the network and 2) packets may incur random loss (e.g., due to transient congestion caused by fluctuations in real-time traffic, or wireless links in the path of the connection). The following key results are obtained. First, random loss leads to significant throughput deterioration when the product of the loss probability and the square of the bandwidth-delay product is larger than one. Second, for multiple connections sharing a bottleneck link, TCP is grossly unfair toward connections with higher round-trip delays. This means that a simple first in first out (FIFO) queueing discipline might not suffice for data traffic in WANs. Finally, while the recent Reno version of TCP produces less bursty traffic than the original Tahoe version, it is less robust than the latter when successive losses are closely spaced. We conclude by indicating modifications that may be required both at the transport and network layers to provide good end-to-end performance over high-speed WANs.

This paper presents a control-theoretic approach to reactive flow control in networks that do not reserve bandwidth. We assume a round-robin-like queue service discipline in the output queues of the network’s switches, and propose deterministic and stochastic models for a single conversation in a network of such switches. These models motivate the Packet-Pair rate probing technique, and a provably stable rate-based flow control scheme. A Kalman state estimator is derived from discrete-time state space analysis, but there are difficulties in using the estimator in practice. These difficulties are overcome by a novel estimation scheme based on fuzzy logic. We then present a technique to extract and use additional information horn the system to develop a continuous-time system model. This is used to design a wuisnt of the control law that is also provably stable, and, in addition, takes control action as rapidly as possible. Finally, practical issues such as correcting parameter drift and cmmlination with window flow control are described.

Abstract not found

We use simulation to study the dynamics of the congestion cent rol algorithm embedded in the BSD 4.3-Tahoe TCP implementation. We investigate the simple case of a few TCP connections, originating and terminating at the same pair of hosts, using a single bottleneck link. This work is an extension of our earlier work ([16]), where one-way traffic (i.e., all of the sources are on the same host and all of the destinations are on the other host) was studied. In this paper we investigate the dynamics that results from two-way traffic (in which there are data sources on both hosts). We find that the one-way traffic clustering and loss-synchronization phenomena d~cussed in [16] persist in this new situation, albeit in a slightly modified form. In addition, there are two new phenomena not present in the earlier study: (1) ACK-compression, which is due to the interaction of data and ACK packets and gives rise to rapid fluctuations in queue length, and (2) an out-of-phase queue-synchronization mode, which keeps link utilization less than optimal even in the limit of very large buffers. These phenomena are helpful in understanding results from an earlier study of network oscillations ([19]). 1

We describe ways in which the transmission control protocol of the Internet may evolve to support heterogeneous applications. We show that by appropriately marking packets at overloaded resources and by charging a fixed small amount for each mark received, end-nodes are provided with the necessary information and the correct incentive to use the network efficiently.

In this paper we explore the bias in TCP/IP networks against connections with multiple congested gateways. We consider the interaction between the bias against connections with multiple congested gateways, the bias of the TCP window modification algorithm against connections with longer roundtrip times, and the bias of Drop Tail and Random Drop gateways against bursty traffic. Using simulations and a heuristic analysis, we show that in a network with the window modification algorithm in 4.3 tahoe BSD TCP and with Random Drop or Drop Tail gateways, a longer connection with multiple congested gateways can receive unacceptably low throughput. We show that in a network with no bias against connections with longer roundtrip times and with no bias against bursty traffic, a connection with multiple congested gateways can receive an acceptable level of throughput. We discuss the application of several current measures of fairness to networks with multiple congested gateways, and show that diff...

this paper we define the notion of traffic phase in a packet-switched network and describe how phase differences between competing traffic streams can be the dominant factor in relative throughput. Drop Tail gateways in a TCP/IP network with strongly periodic traffic can result in systematic discrimination against some connections. We demonstrate this