High Performance Interconnection Between High Data Rate Networks

Abstract

A rjc tm Ct , [13] are becoming a reality. This paper investigates TJie paper discusses the bridge/gateway system needed \ syslsms required to interconnecting high data rate netto mterconnect a wide range of computer networks to ', works Spedal attcntion ^ directed toward providing support a wide range of user quality-of-service require-suppon for a wide variety of applications which are imments. Vie bridge/gateway must handle a wide range of plied by multimedia applications. message types including synchronous and asynchronous traffic, large, bursty messages, short, self-contained messages, time critical messages, etc. TJte paper shows that messages can be classified into three basic classes, synchronous and large and small asynchronous messages. TJie first two require call setup so that packet identification, buffer handling, etc. can be supported in the bridge/ gateway. Identification enalbes resequencing of messages at the bridge/gateway which supports interconnection between networks having large differences in packet size. Tlie third class is for messages which do not require call setup. Resequencing hardware is presented in the paper based to handle two types of resequencing problems. TJie first is for virtual parallel circuitwhich can scramble Present network interfacing supports service, between connectionless networks usually via bridges or gateways [L]. Tthe major interconnection problems has been routing. Gateways, such as in Internet, provide a compatible protocol at the network layer which forwards packets to toward their destination on the selected "best route" and, where necessary, breaks up the packet to conform to any lower layer packet length restrictions [1,2]. Thus, all stations must implement the same network layer protocol.To over come the problem of common network layer, especially where not needed the transparent bridge approach is used. Bridges generally implement some form of spanning tree algorithm which eventually gets the packet to its destination but some times via a "non-shortest" route channel bytes. Vie secondsystem is effective in handling ri> 2 , 3> 4]. Packet reformatting in bridges is minimal at both synchronous and asynchronous traffic between networks with highly differing packet sizes and data rates. Vie two other major needs for the bridge/gateway are congestion and error control. Vie paper presents a new dynamic, lossless congestion control scheme which can easily support effective error correction. Results indicate that the congestion control scheme provide close to optimal capacity under congested conditions. Under conditions where error may develop due to intewening networks which are not lossless, intermediate error recovery and correction takes 1/3 less time than equivalent end-co-end error correction under similar conditions.