Scalable service migration in general topologies

Service placement has typically been studied through the formulation and solution of a 1-median problem that is known to be complex and require global information. As such approaches are clearly not scalable in dynamic and large-scale environments, a more reasonable approach to service placement for large, ad hoc and autonomic environments would be through service migration. Such a migration policy has recently been proposed for tree topologies. The focus of the present paper is to devise effective service migration policies for general topologies with equal weight links. The proposed strategy - that incorporates a 2-phase polling mechanism of local scope and, consequently, induces in general a certain overhead - is shown to out-perform the previous one (i.e., getting closer to the optimal placement) for such general topologies. Finally, a hybrid migration policy (combining the present and the past ones) is considered and shown through simulations to be capable of reducing significantly the overhead effects of the afore-mentioned polling mechanism.