Novel techniques in large scaleable ATM switches
Master Thesis
2000
Permanent link to this Item
Authors
Supervisors
Journal Title
Link to Journal
Journal ISSN
Volume Title
Publisher
Publisher
University of Cape Town
Department
License
Series
Abstract
This dissertation explores the research area of large scale ATM switches. The requirements for an ATM switch are determined by overviewing the ATM network architecture. These requirements lead to the discussion of an abstract ATM switch which illustrates the components of an ATM switch that automatically scale with increasing switch size (the Input Modules and Output Modules) and those that do not (the Connection Admission Control and Switch Management systems as well as the Cell Switch Fabric). An architecture is suggested which may result in a scalable Switch Management and Connection Admission Control function. However, the main thrust of the dissertation is confined to the cell switch fabric. The fundamental mathematical limits of ATM switches and buffer placement is presented next emphasising the desirability of output buffering. This is followed by an overview of the possible routing strategies in a multi-stage interconnection network. A variety of space division switches are then considered which leads to a discussion of the hypercube fabric, (a novel switching technique). The hypercube fabric achieves good performance with an O(N.log₂N)²) scaling. The output module, resequencing, cell scheduling and output buffering technique is presented leading to a complete description of the proposed ATM switch. Various traffic models are used to quantify the switch's performance. These include a simple exponential inter-arrival time model, a locality of reference model and a self-similar, bursty, multiplexed Variable Bit Rate (VBR) model. FIFO queueing is simple to implement in an ATNI switch, however, more responsive queueing strategies can result in an improved performance. An associative memory is presented which allows the separate queues in the ATM switch to be effectively logically combined into a single FIFO queue. The associative memory is described in detail and its feasibility is shown by laying out the Integrated Circuit masks and performing an analogue simulation of the IC's performance is SPICE3. Although optimisations were required to the original design, the feasibility of the approach is shown with a 15Ƞs write time and a 160Ƞs read time for a 32 row, 8 priority bit, 10 routing bit version of the memory. This is achieved with 2µm technology, more advanced technologies may result in even better performance. The various traffic models and switch models are simulated in a number of runs. This shows the performance of the hypercube which outperforms a Clos network of equivalent technology and approaches the performance of an ideal reference fabric. The associative memory leverages a significant performance advantage in the hypercube network and a modest advantage in the Clos network. The performance of the switches is shown to degrade with increasing traffic density, increasing locality of reference, increasing variance in the cell rate and increasing burst length. Interestingly, the fabrics show no real degradation in response to increasing self similarity in the fabric. Lastly, the appendices present suggestions on how redundancy, reliability and multicasting can be achieved in the hypercube fabric. An overview of integrated circuits is provided. A brief description of commercial ATM switching products is given. Lastly, a road map to the simulation code is provided in the form of descriptions of the functionality found in all of the files within the source tree. This is intended to provide the starting ground for anyone wishing to modify or extend the simulation system developed for this thesis.
Description
Bibliography: p. 172-178.
Keywords
Reference:
Lawrence, M.A. 2000. Novel techniques in large scaleable ATM switches. Thesis. University of Cape Town ,Faculty of Engineering & the Built Environment ,Department of Electrical Engineering. http://hdl.handle.net/11427/7672