It is widely accepted that large systems cannot efficiently
operate if a single component of the system is responsible, in
some part, for global operation. Such operation is called
centrally run. Centrally run systems are relatively easy to
develop but don't scale well because the central components become
bottlenecks. The Internet is not centrally run and indeed, it is
reasoned that this was necessary for its remarkably successful
growth. The Internet is said to be self-organizing in contrast to
being centrally run. If future systems are to achieve comparable
growth then they must operate using similar principles of
self-organization. The downside to self-organization is that
certain kinds of services, such as connection oriented
communication, become difficult to efficiently provide because
they tend to require complete knowledge of the underlying network --
which becomes unscalable. The purpose of this seminar is to
provide the background of this research topic, the current
research directions and some recent results. Specifically, the
problem of achieving a self-organized hierarchical structure. A
random hierarchy is proposed as a solution and some of its
properties are analysed. The use of a randomness directly
facilitates self-organization so is advocated as a key technique
for further research and development.
This seminar was also presented at I-SPAN'02, The Sixth International
Symposium on Parallel Architectures, Algorithms, and Networks.
Brief biography: Aaron Harwood is a new Lecturer in the Department
of Computer Science and Software Engineering at The University of Melbourne. |