| dc.description.abstract | Process scheduling is a very important part of a computer system and defines the performance
of the same. There are several possibilities of polices to deal with it (e.g., FIFO, SJF, on
time, etc..). And, in addition, due to the heterogeneous nature of many applications, the choice
of a scheduling policy is not always a trivial task. Indeed, choosing the most appropriate policy
is related to the scenario and current demand. If the scenario and demand change, an adaptation
may be necessary. Thus, this work represents a step forward in adapting to the process
scheduling in computer systems.
More precisely, this work is about the application of scheduling policies and adaptation
in face of changes. Traditional policies may be combined into an adaptive method, which
changes its behavior as the workload of the processes changes. The adaptation of the scheduler
behavior occurs through the use of heuristics. It trades at a random policy. If the new
policy shows better result than the former, execution continues with the new policy until a new
evaluation occurs.
The policy evaluation was performed with the support of simulations, considering different
scenarios and demands. To allow the study and the evaluate of the policies, we used
a stable environment, which considers the network without delays and failures. Through our
experiments, the decisions for the development of adaptive scheduler may be taken. This study
considered the use of only one server. But the adaptation method can be applied to decentralized
systems as well.
To conduct the experiments, we used the Lua programming language due to the suitable
support in the development of prototype in distributed systems. With this work, we emphasize
that the language proves to be very simple and powerful for prototyping in computing systems. | |
