Information Technology Reference
In-Depth Information
Strengths and weaknesses
As Scheutz and Schermerhorn are focused on parallelizing the compu-
tation of the simulation of an agent-based model, the framework is
just developed and used to describe and verify their load distribution
algorithms. Unfortunately, as Scheutz and Schermerhorn just pub-
lished one article [114] on this topic, some important aspects remain
open.
The aspect of greatest interest in the context of this thesis is the de-
cision to assume a time-stepped execution of the simulation model with
fixed time increments. Besides the remark that 'to their knowledge,
all agent-based model simulation environments capable of distributed
simulation implement some form of stepwise synchronization mech-
anism' [114] similar to the one they proposed, they do not motivate
their decision in any way. Given the numerous arguments for and
against time-stepped simulation with fixed time increments, especially
when compared to discrete event simulation (e.g., ineciency due to
long idle periods, strictly periodical state updates), a more detailed
explanation for choosing a time-stepped execution of the model is
missing.
A further important aspect concerns the resolution of conflicting
actions of multiple agents which is not explained in detail. They do
not mention this problem at all, although it can be assumed that
the various update functions are responsible for this task. Again, as
this aspect is central and can be a potential bottleneck (as simply
verifying all possible agent interactions requires quadratic runtime),
it would be of interest how it is handled. This aspect is even more
interesting as the proposed load distribution algorithms make explicit
use of the
update independence
of the entities. In a way, Scheutz and
Schermerhorn deal only with update-independent entities, i. e., with
entities which do not conflict with each other during a simulation
step.
Besides these drawbacks, a highly appreciated result of Scheutz
and Schermerhorn is that they can proof the correctness of their
load balancing algorithm. This feature of being able to proof the
