Node Selection for Placement of Migratory Tasks in Wide-Area Shared Computing Environments

In a wide-area shared computing environment such as the PlanetLab the available resource capacities at nodes can fluctuate significantly. A node selected based on its available resource capacities for executing a task may become unsuitable for it within a few minutes or hours timeframe. This motivates the need to find another node and relocate the task. In this paper we present a comparative evaluation of different strategies for selecting nodes in the PlanetLab environment for the placement and relocation of migratory tasks. We have developed a system for monitoring PlanetLab nodes for their available resource capacities. Using this system we characterize the distribution of the number of nodes that meet a given resource capacity requirement, and the distribution of the duration for which a node satisfies the requirement.