Simulation study of World Wide Web traffic over the ATM ABR service

ABSTRACT
1. INTRODUCTION
In this section, we give a brief overview of TCP congestion control, the ABR ow control mechanism, and the ERICA+ rate allocation algorithm. TCP provides a reliable, connection-oriented service. TCP connections provide window-based end-to-end ow control. The receiver's window (rcvwnd) is enforced by the receiver as a measure of its bu ering capacity. The congestion window (cwnd) is used at the sender as a measure of the available capacity of the network. The sender cannot send more than the minimum of rcvwnd and cwnd. The TCP congestion control scheme6 consists of the \slow start" and \congestion avoidance" phases. In the \slow start" phase, cwnd is initialized to one TCP segment. The cwnd is incremented by one segment for each acknowledgment received, so the cwnd doubles every round trip. The cwnd can reach up to ssthresh (initialized to 64K bytes). In the \congestion avoidance" phase, the cwnd is incremented by 1=cwnd for every segment acknowledged. If an acknowledgment is not received by the timeout period, the segment is considered to be lost, and \slow-start" phase is entered. The cwnd is set to one, and ssthresh is set to max(2, min(cwnd/2, rcvwnd)). Figure 1 shows the di erent phases of the TCP congestion control mechanism. It has been shown that TCP/IP performance can be improved with the fast retransmit and recovery7 and selective acknowledgments8 options.
Proceedings of SPIE Symposium on Voice, Video and Data Communications, Vol. 3530, Conference on Performance and Cooston, MA November 1998, pp. 415-422. c SPIE. Further author information: Tel: (614) 688-4482; Fax: (614) 292-2911; E-mail: fvandalor, jaing@; WWW: /~jain/
Asynchronous transfer mode (ATM) is the technology chosen for implementing the Broadband Integrated Services Digital Network (B-ISDN). The performance of Internet protocols over ATM is an extremely important research area. As web tra c forms a major portion of the Internet tra c, we model world wide web (WWW) servers and clients running over an ATM network using the available bit rate (ABR) service. The WWW servers are modeled using a variant of the SPECweb961 benchmark, while the WWW clients are based on a model proposed in 2]. The tra c generated is typically bursty, having active and idle transmission periods. A timeout occurs after a certain idle interval. During idle periods, the underlying TCP congestion windows remain large until the timer expires. When the application becomes active again, these large windows may be used to send data in a burst. This raises the possibility of large queues at the switches, if the source rates are not controlled by ABR. We study this problem and show that ABR scales well to a large number of bursty TCP sources in the system. Keywords: ATM, WWW model, TCP/IP, ABR service ATM is a high speed, connection-oriented, cell switching technology. It uses small xed-size packets (also called cells) to transport its tra c. It is designed to provide an integrated service for supporting audio, video and data tra c. ATM provides multiple categories of service to support di erent quality of service (QoS) requirements. The current set of service categories speci ed are: the constant bit rate (CBR), real-time variable bit rate (rt-VBR), non-real time variable bit rate (nrt-VBR), available bit rate (ABR), and unspeci ed bit rate (UBR). The CBR service is aimed at transporting voice and synchronous data applications. The VBR (rt- and nrt-) services provide support for video and audio applications which do not require isochronous transfer. The ABR service transports data tra c. It guarantees a minimum cell rate (MCR) allocation and uses closed-loop feedback to control source rates. Cell loss is small in ABR, though ABR does not give bounded cell loss guarantees. UBR provides \best-e ort" delivery for data applications and does not give any guarantees. The ATM technology is already being used in the backbone of the Internet and many campus backbones. The transport control protocol/internet protocol (TCP/IP) suite of protocols are the most important protocols of the Internet. The performance of TCP/IP over ATM has been addressed in 3{5]. As large ATM networks are built, it is important to study the performance of real-world applications like the World Wide Web over ATM. Such applications typically have low average bandwidth demands from the network, but desire good response time when they become active. It is interesting from the tra c management perspective to study the aggregate e ect of hundreds of such applications browsing or down-loading large documents over an ATM backbone. This study focuses on the performance of WWW tra c over the ATM ABR service. Section 2 gives an overview of the TCP congestion control algorithm, the congestion control mechanism in the ATM ABR service, and the ERICA+ algorithm. Section 3 discusses the issues involved in transporting WWW tra c over ATM. Section 4 discusses the implications of using the HTTP/1.1 standard and presents the WWW server model and the WWW client model used in this study. Section 5 explains the simulation con gurations, and section 6 gives the TCP/IP and the ERICA+ parameter values used in the simulations. Section 7 discusses the simulation results. Finally, section 8 provides a summary of the paper.