Tan, Jiann Shin (2005) Load balancing and congestion control in ATM using fuzzy logic and spiking neuron network / Tan Jiann Shin. Masters thesis, University of Malaya.
Abstract
The objective of this project is to propose an intelligent routing solution that is capable of increasing network performance in Asynchronous Transfer Mode (ATM) network for Available Bit Rate data traffic. The proposed intelligent routing algorithm, integrates spiking neuron network with fuzzy link cost for the dynamic routing computation. Although ATM network is a high speed network, it will encounter performance degradation due to heavy traffic or network congestion. Effective intelligent dynamic routing is able to minimize the impact of performance degradation. Among all the introduced concepts on dynamic routing, load balancing is the most popular. Load balancing is proven to be able to increase network performance in various studies [39]. However, distributing network load evenly to load balance network traffic is highly desirable only for single rate traffic. It may not be the best strategy for multi rate environment. Load balancing concept has the tendency to cause bandwidth fragmentation that leads to the result of rejecting connection earlier than it should be. Bandwidth packing is able to minimize the problem of bandwidth fragmentation [34]. Bandwidth packing is a concept of grouping similar traffic which requires similar bandwidth utilization on high utilization links to save bandwidth on other low utilization links for high requirement connections. Both load balancing and bandwidth packing concept may contradict each other, but it is possible to incorporate both concepts into a single routing solution. Fuzzy logic is introduced to incorporate both concepts (load balancing and bandwidth packing) using the network state metrics (current buffer size, buffer increase rate, bandwidth consumption and requested bandwidth) feedback from each network node. The capability to simultaneously maximize multiple objectives and easy maintenance makes fuzzy logic the perfect candidate to calculate link cost [5]. Spiking neuron network will use the computed fuzzy link cost to find the minimum cost route. The spiking neuron network is an efficient tool to solve the shortest path routing problem from source node to destination node. It has the flexibility to maximize solution space by recording details of all candidate routes for optimal route selection [37]. An experiment is conducted to measure the performance of the proposed model. A control model is created to work as a benchmark for comparison. According to the experiment result, the proposed model outperforms the control model by accepting more connections while maintaining lower overall cell loss ratio. The proposed model proves that artificial intelligence implementation (using fuzzy link cost and spiking neuron network) realize concept integration easily. The integrated concepts could cover each other’s weaknesses and strengthen each other’s strengths.
Actions (For repository staff only : Login required)
![[img]](http://studentsrepo.um.edu.my/style/images/fileicons/application_pdf.png)
