Performance analysis of fuzzy thresholding-based buffer management for a large-scale cell-switching network

Abstract

This paper addresses three key issues in buffer management in cell switching networks including asynchronous transfer mode (ATM) networks. First, it develops a model of fuzzy thresholding-based buffer management for a 50-switch representative cell-switching network, to study its performance under realistic conditions. Previous studies have been confined to a single switch and, consequently, the results are not necessarily applicable to a real-world cell-switching network consisting of multiple switches. Second, it presents an approach to reroute to their final destinations, the fraction of the selectively blocked cells that correspond to the difference of cell loss due to buffer overflow between the fixed and fuzzy schemes. While the fixed threshold, through its abrupt nature, causes a relatively higher cell drop through buffer overflow, intuitively, the fuzzy threshold may trade off cell loss through buffer overflow for increased selective blocking at the sending switch. While the goal of the second issue is to improve throughput and thereby achieve higher reliability, the delays incurred by the cells in the network are likely to increase. Third, this is the first paper to report on the influence of the buffer management scheme on the end-to-end delay performance of a representative cell switching network. The model is simulated on a testbed consisting of a network of 25+ Pentium workstations under linux, configured as a loosely coupled parallel processor. Simulation results reveal that, even for a large-scale representative cell-switching network, the fuzzy approach adapts superbly to different bursty input traffic distributions, yielding lower cell loss rates. A total of 10 000 user calls, generating between 1.0 and 1.5 million ATM cells, is stochastically distributed among the nodes. Performance analysis reveals that for different input traffic distributions ranging from light to moderate to heavy traffic, the fuzzy threshold scheme consistently succeeds in lowering the cell loss due to buffer overflow relative to fixed thresholding by blocking them at the sending switch. The re-routing approach, in turn, successfully routes these blocked cells, although it causes the average end-to-end cell delay in the network to increase compared to the fixed scheme by a factor ranging from 1.65 for relatively light traffic to 6.7 for heavy traffic.Index Terms-asynchronous transfer mode (ATM) networks, buffer management, cell-switching network, fuzzy thresholds, large-scale switching network.

High-speed networks: definition and fundamental attributes

“…The exact units of the quantities in the above equation will depend on the network under consideration. Estimates of 't' may be obtained through sophisticated modeling and simulation [18].…”

A modeling and simulation methodology for analyzing ATM network vulnerabilities

“…In this study, ATMSIM 1.0 is executed on a testbed of Pentium laptops, configured under the Linux operating system and interconnected through a 100 Mb/s Fast Ethernet. While ATMSIM 1.0 incorporates input traffic [16,17], designed to emulate a representative ATM network, it permits users to generate call SETUP requests externally and insert them into the network. Traffic consists of audio, video, and data.…”

An adaptive, dynamic buffer management (ADBM) approach for input buffers in ATM networks

“…The fuzzy buffer management technique utilizes an asymmetric "sigmoid" membership function from which fuzzy threshold values are determined corresponding to the discrete buffer occupancy levels [3] [4], and for the fixed thresholding scheme, the admission policy is encapsulated by the equation:…”

“…They proposed the use of fuzzy thresholds and comparative analysis on a representative, large-scale, ATM network [4] revealed that this approach adapts superbly to sharp changes in cell arrival rates and maximum burstiness of bursty traffic sources. Under this approach, the threshold is adjusted continuously to reflect the dynamic state of the buffer.…”