We study the performance of TCP in an internetwork consisting of both rate-controlled and nonrate-controlled segments. A common example of such an environment occurs when the end systems are part of IP datagram networks interconnected by a rate-controlled segment, such as an ATM network using the available bit rate (ABR) service. In the absence of congestive losses in either segment, TCP keeps increasing its window to its maximum size. Mismatch between the TCP window and the bandwidth-delay product of the network results in accumulation of large queues and possibly buffer overflows in the devices at the edges of the rate-controlled segment, causing degraded throughput and unfairness. We develop an explicit feedback scheme, called explicit window adaptation, based on modifying the receiver's advertised window in TCP acknowledgments returning to the source. The window size indicated to TCP is a function of the free buffer in the edge device. Results from simulations with a wide range of traffic scenarios show that this explicit window adaptation scheme can control the buffer occupancy efficiently at the edge device, and results in significant improvements in packet loss rate, fairness, and throughput over a packet discard policy such as random early detection (RED).

We discuss a hardware implementation of an explicit rate allocation algorithm for support of available bit rate (ABR) service in ATM switches. We then demonstrate the effectiveness of the algorithm at the network-level through measurements on the actual implementation in a network testbed. The rate allocation algorithm has several desirable properties, such as exact computation of the max-min rates, O(1) computations per resource management (RM) cell received, and the ability to provide minimum cell rate (MCR) guarantees. We show that the algorithm can be implemented with a modest amount of hardware (64,000 gates in an Altera 10K100 programmable logic device and 16 bytes of SRAM storage per VC), and that even a slow FPGA-based implementation with a 20 MHz internal clock rate can process RM cells within one cell time at an OC-3 port. We also outline the design for supporting OC-12 and OC-48 links. We present results from measurements of ABR traffic in network configurations with up to four bottlenecks and 100 connections. The results show that the algorithm is able to converge to the exact max-min fair allocations with no oscillations after convergence, maintains minimum rate guarantees, and is also able to maintain high link utilization in the presence of on-off traffic.

Loss of the routing protocol messages due to network congestion can cause peering session failures in routers, leading to route flaps and routing instabilities. We study the effects of traffic overload on routing protocols by quantifying the stability and robustness properties of two common Internet routing protocols, OSPF and BGP, when the routing control traffic is not isolated from data traffic. We develop analytical models to quantify the effect of congestion on the robustness of OSPF and BGP as a function of the traffic overload factor, queueing delays, and packet sizes. We perform extensive measurements in an experimental network of routers to validate the analytical results. Subsequently we use the analytical framework to investigate the effect of factors that are difficult to incorporate into an experimental setup, such as a wide range of link propagation delays and packet dropping policies. Our results show that increased queueing and propagation delays adversely affect BGP's resilience to congestion, in spite of its use of a reliable transport protocol. Our findings demonstrate the importance of selective treatment of routing protocol messages from other traffic, by using scheduling and utilizing buffer management policies in the routers, to achieve stable and robust network operation.

A novel parallel-prefix architecture for high speed module 2/sup n/-1 adders is presented. The proposed architecture is based on the idea of recirculating the generate and propagate signals, instead of the traditional end-around carry approach. Static CMOS implementations verify that the proposed architecture compares favorably with the already known parallel-prefix or carry look-ahead structures.

We study the performance of the TCP in an internetwork consisting of both rate-controlled and non-rate-controlled segments. A common example of such an environment occurs when the end systems are part of IP datagram networks interconnected by a rate-controlled segment, such as an ATM network using the ABR service. In the absence of congestive losses in either segment, the TCP keeps increasing its window to its maximum size. Mismatch between the TCP window and the bandwidth-delay product of the network will result in an accumulation of large queues and possibly buffer overflows in the devices at the edges of the rate-controlled segment, causing degraded throughput and unfairness. We develop an explicit feedback scheme, called explicit window adaptation based on modifying the receiver's advertised window in the TCP acknowledgments returning to the source. The window size indicated to the TCP is a function of the free buffer in the edge device. Results from simulations with a wide range of traffic scenarios show that this explicit window adaptation scheme can control the buffer occupancy efficiently at the edge device, and results in significant improvements in packet loss rate, fairness, and throughput over a packet discard policy such as drop-from-front or random early detection.

The sharing of a common buffer by TCP data segments and acknowledgments in a network or internet has been known to produce the effect of ack compression, often causing dramatic reductions in throughput. We study several schemes for improving the performance of two-way TCP traffic over asymmetric links where the bandwidths in the two directions may differ substantially, possibly by many orders of magnitude. These approaches reduce the effect of ack compression by carefully controlling the flow of data packets and acknowledgments. We first examine a scheme where acknowledgments are transmitted at a higher priority than data. By analysis and simulation, we show that prioritizing acks can lead to starvation of the low-bandwidth connection. Next, we introduce and analyze a connection-level backpressure mechanism designed to limit the maximum amount of data buffered in the outgoing IP queue of the source of the low-bandwidth connection. We show that this approach, while minimizing the queueing delay for acks, results in unfair bandwidth allocation on the slow link. Finally, our preferred solution separates the acks from data packets in the outgoing queue, and makes use of a connection-level bandwidth allocation mechanism to control their bandwidth shares. We show that this scheme overcomes the limitations of the previous approaches, provides isolation, and enables precise control of the connection throughputs. We present analytical models of the dynamic behavior of each of these approaches, derive closed-form expressions for the expected connection efficiencies in each case, and validate them with simulation results.

This paper provides an analysis of the source policy in the rate-based congestion control scheme developed by the Asynchronous Transfer Mode (ATM) Forum for available bit rate service and derives approximate analytical closed-form expressions to describe the rate increase process. These approximations are used to analyze the impact of the source algorithm on the TCP slow-start process operating over a rate-controlled ATM network. The results show that the increase in TCP congestion window ramp-up time is noticeable when the round-trip delay is small. The results are verified by simulation.

We study the performance of bidirectional TCP/IP connections over a network that uses rate-based flow and congestion control. An example of such a network is an asynchronous transfer mode (ATM) network using the available bit rate (ABR) service. The sharing of a common buffer by TCP packets and acknowledgment (acks) has been known to result in an effect called ack compression, where acks of a connection arrive at the source bunched together, resulting in unfairness and degraded throughput. It has been the expectation that maintaining a smooth flow of data using rate-based flow control would mitigate, if not eliminate, the various forms of burstiness experienced with the TCP window flow control. However, we show that the problem of TCP ack compression appears even while operating over a rate-controlled channel. The degradation in throughput due to bidirectional traffic can be significant. For example, even in the simple case of symmetrical connections with adequate window sizes, the throughput of each connection is only 66.67% of that under one-way traffic. By analyzing the periodic bursty behavior of the source IP queue, we derive estimates for the maximum queue size and arrive at a simple predictor for the degraded throughput, for relatively general situations. We validate our analysis using simulation on an ATM network using the explicit rate option of the ABR service. The analysis predicts the behavior of the queue and the throughput degradation in simple configurations and in more general situations.

We examine the performance of bidirectional TCP/IP connections over asynchronous transfer mode (ATM) networks using the available bit rate (ABR) service. The problem of "ack-compression" re-appears, although the queues are primarily at the end-systems. We further the understanding of the problem by quantitatively analyzing the periodic bursty behavior of the source IP queue. We are able to predict the peak values for the queue and arrive at a simple robust predictor for the degraded throughput, applicable for relatively general situations. The degradation in throughput due to bidirectional traffic can be significant. For example, even in the simple case of symmetrical connections with adequate window sizes, the throughput of each connection is only 66.67% of that under one-way traffic. We validate our analysis using simulation, where the ATM network uses the explicit rate option. We show that the analysis predicts the behavior of the queue and the throughput degradation. We observe the need to separate the flow of acknowledgments and data for the bidirectional TCP connection and for inter-leaving their processing at the end-systems to overcome the problem of ack compression.