|
Some of the existing networking topology techniques optimize the number of created replicas [16],
while others optimize the locations in which to place replicas [17]; still others
optimize how often replicas should be updated [18]. However, many of these
techniques have the shortcoming that they only consider a limited set of parameters
affecting the replication decision. The works by Loukopoulos and Ahmad
[12] has the same objective as ours. They design a genetic algorithm to find the
optimal replication strategy. In that work, two versions of the algorithm, a static
version and a dynamic adaptive version are proposed. However, they model the
problem as a single objective optimization problem. Specifically, they optimize
latency, while storage, bandwidth and other parameters are considered as constraints.
One of the limitations of this approach is that it can only maintain
the constraint parameters within certain bounds, but cannot explicitly optimize
them. Further, their work did not take into account the reliability of the system.
Thus, we believe that there is a need for a holistic approach to the overlay
replica placement problem which not only takes all the important factors into account
but also explicitly optimizes them. Motivated by this need, we propose the
MOE and MORG algorithms, both of which are based upon the multi-objective
optimization paradigm.
|
