Efficiency of the Matcher
The production match during the
elaboration phase
requires the
greatest amount of computational resources for Soar systems. Although this
process is implicitly parallel, it
is normally instantiated on serial computers. As the
size of the knowledge base grows,
Soar systems may exhibit a concomitant slow-down related to inefficiencies
in the match algorithm. Additionally, although the RETE algorithm
used in the Soar production system is quite efficient, there are physical
limits on the speed of the production match. Thus, the
scalability of Soar systems is an open
question for systems with very large knowledge bases.
Average Growth Effect
The size of the knowledge base
causes more than just a consideration of the
efficiency of the production matcher. An
increase in the number of productions represents a greater resource
drain for many of the algorithms underlying the architecture. Thus,
there may be a slow-down associated with increasing the knowledge base even
when the
elaboration phase is
actually parallelized. This slow-down with respect to an increase in the
size of the knowledge base is known as the average growth effect.
Although the average growth effect has not been a particular problem for
any Soar systems implemented thus far, it must be considered in
relevance to Soar's
scalability, especially
since the solution to
expensive chunks is to generate a indefinitely
large number of 'cheap' chunks.
Organization of Large Bodies of Knowledge
Soar's management of knowledge is inherently non-modular. Although
match-based attention relieves this
problem (to some degree) at run-time, the knowledge engineer/system
designer must organize this information to make it manageable for
human knowledge compilation. This somewhat arbitrary division has
repercussions for debugging and maintenance of Soar systems.
Return to the top of this architecture.
Go to a Discussion of this issue for multiple architectures.