Very large networks with thousands of applications and millions of users
pose serious scalability problems to the current technology. Moreover, the
limitations are strongly felt if synchronised client behaviour exists. This
dissertation proposes the usage of a co-operative mobile agent system with a
very dynamic and scalable name service (called location service). It proposes a
new algorithm to control the self-deployment of application servers on the
network to respond to demand, and a simplified theoretical model to configure
the algorithm parameters.
Current name services are adapted to names with long lifetimes but scale
poorly on this scenario because references to the servers have a short
lifetime, invalidating the use of client controlled caching. The amount of
servers and their mobility invalidate the pure hierarchical structures of name
servers.
The location service identifies applications using anycast
names. The usage of anycast names improves
scalability because clients can interact with any of the application servers. A
new structure for a location service is proposed, which scales better because
it is dynamic and adapts to the overload situations. The adaptation is done
both in terms of the internal structure and the number of location servers.
Sets of simulations were performed to study the dynamic behaviour of the
overall system, and identify the relevant tuning parameters.
Scalable Networking Architectures; Scalable Applications; Dynamic
Replicated Servers; Location Service; Anycast Names;