This paper will be an investigation on how empirical
modelling principles and tools can be used to model
grid computing middleware components and in
particular the contract management service. It will
be based on a model of a contract manager that will
actually be created.
In grid computing, a pool of heterogeneous
underutilised resources is available for users to
submit tasks to, which can be executed in parallel.
The middleware provides services to the users and
the resources connected to the grid. A contract
manager ensures that when a user and a resource
have come to an agreement about the execution of a
task, the agreement is carried out.
Part of the study will include creating a model
which will present three main agents, the user, the
resource and the contract manager which in case of
the failure of a job, will determine the reason and
will call on appropriate services to handle the
problem.
Users and resources will be connected in a network,
whereas the jobs will be observed as progress bars.
The model will thus show the parameters that a
contract manager needs to be watching and its
behaviour responding to unexpected changes. The
main dependencies involve the job execution
depending on the resources, network link and
user/owner of the job, but also the contract
manager’s behaviour depending on the progress of
the job. Problems in the execution of jobs will be
randomly generated and can be referring to the
network link, the position, the hardware and the
speed of the resource.
As this model presents a part of an agent system
and its behaviour to external stimuli, it falls
within the artificial intelligence category. As
such it is expected that empirical modelling will
prove to be a useful tool in modelling grid
computing components and thus helping in breaking
down a complicated problem and determining its
states.
