In this thesis, we deal with blackboard system performance issues. We show that
blackboard system performance can be improved using parallel processing strategies
and a novel blackboard architecture.
We study traditional blackboard architectures using a novel performance frame¬
work. This is a useful tool for directing system optimisation efforts. We present the
analysis of four blackboard systems present in the literature.
nalysis of four blackboard systems present in the literature.
Besides localised optimisation efforts, one of the most promising approaches for
improving blackboard system performance is the use of parallel processing techniques.
However, traditional blackboard architectures present both data and control contention
when implemented in parallel.
In this thesis we present a novel blackboard architecture, the Active Blackboard
Architecture (ABB). We based ABB on a novel variation of the traditional "Blackboard
and Experts" metaphor, called "Blackboard, Experts and Desks". This new metaphor
introduces a new element, the desks, used by the experts to perform their work.
The ABB architecture is based on an active blackboard, capable of processing on its
own, and a decentralised control model. This avoids control contention and bottlenecks.
We describe this architecture using the Z specification language, and implemented
and evaluated in the EPCC Meiko Computing Surface, a multi-transputer distributed
memory parallel machine.
The ABB Parallel prototype is an object oriented implementation of the ABB model
that overcomes both data and control bottlenecks by having a distributed blackboard
and using the ABB control model. Based on a series of experiments, we show that the
new architecture allows to achieve much greater effective parallelism in a blackboard
system. We also present some ways in which the system can be tailored to specific
application needs, improving in this way its overall performance.