Increasing throughput in ATM/B-ISDN using simple buffer management and selective discarding

Abstract

If the reliability provided by the network is lower than that requested by the application, the error and flow control systems in the end-nodes must make up for the difference. While most applications use retransmission based schemes to recover lost data, ARQ based closed loop techniques may not be suitable for many new applications. Time sensitive information requiring no-loss of data would constitute a major segment of the emerging traffic (interactive web, videophones, CSCW) that typically employ Forward Error Correction (FEC) codes. Such codes send redundant information which help to recover lost data without retransmission. This paper shows that just plain FEC is not enough when one considers the fragmentation effects that takes place when legacy networks interwork with ATM. Simple yet powerful buffer management techniques can yield higher throughput thereby reducing block loss rates. A combination of a gatekeeping operation along with janitor routine that cleans up the buffer to preserve frames on an end-to-end basis is proposed. A simulation study demonstrates positive results. The proposed schemes even point towards efficient sizing of ATM switch buffers which guarantee no-loss transport of real-time information.