Date of Award

Fall 2003

Document Type

Thesis

Degree Name

Master of Science in Telecommunications - (M.S.)

Department

Electrical and Computer Engineering

First Advisor

Lev A. Zakrevski

Second Advisor

Nirwan Ansari

Third Advisor

Sirin Tekinay

Abstract

Deadlock occurrence is a critical problem for any computer network. Various solutions have been proposed over last two decades to solve problem of deadlocks in networks using different routing schemes, like up/down routing algorithm used in Myrinet switches. However, most of existing approaches for deadlock-free routing either try to eliminate any possibility of deadlock occurrence, which can result in putting extra restrictions on the routing in the networks or put no restrictions on routing, which leads to other approach namely deadlock recovery. In this thesis emphasis is on developing hybrid approach for routing in wormhole networks, wherein some prohibition is imposed on routing along with some kind of deadlock recovery. This adaptive approach allows changing the amount of routing restrictions depending on network traffic, thus providing a flexible method to achieve better network performance compared to the existing techniques. The main idea of the proposed method consists in the sequential selections of some turns, which are prohibited to be selected during routing. After each additional turn is added, the probability of deadlock occurrence decreases gradually. Cost formula is proposed to estimate cost of implementing both strategies in a network which is basis of proposed adaptive model.

Share

COinS