|
The authors deal with the replication problem as a single objective optimization;
they optimize latency taking into consideration satisfying constraints
related to some parameters like storage availability. Dealing with these parameters
as constraints will ensure that the constraints are met. But they did not
find the best value for those parameters. This is not the right approach. A justification
of that is the following. In many cases, nodes of the system are not
dedicated to a specific service, and they might host different services. Those
services might consume nodes resources such as storage, each node according to
its behavior. At some point, the system could suffer from lack of storage, which
means that storage in this case is a vital resource, and we should maximize it as
much as possible instead of keeping it under a certain level using a constraint.
This is why the multi-objective optimization approach is better than the single
objective genetic algorithm approach. If we are in a situation where the network
is in a critical condition that requires a quick solution, then the best approach
to be used is MORG. Sometimes, switch network topology can have low traffic during certain
overnight hours, and in this case the best approach to choose is MOE because
of the wide variety of solutions that give us the best optimization for our objectives.
Since the multi-objective evolutionary approach takes long execution time,
one thing to be done is to use forecasting techniques that help us to estimate
a good time to execute the algorithm instead of executing the algorithm regularly.
Also, in cases where the system administrator is monitoring the network,
he can simply execute the algorithm whenever he finds a necessity to do so and
this can minimize the number of times in which the multi-objective evolutionary
algorithm needs to be executed.
|
