Resource allocation heuristics for the miriaPOD platform

Cloud based services have seen a steady increase in popularity in recent time due to high availability, low cost and ease of use. The performance of such services is determined by the resource allocation algorithms. Generic allocation algorithms like best fit or first fit are an adequate solution for generic services. When additional information is available about the applications that are deployed over the cloud, the information can be used by the resource allocator to make a more informed, and thus, more efficient, decision. We propose an alternative resource allocator for the miriaPOD platform, a distributed system for the virtualization of entire network topologies. The allocator exploits the unique memory sharing features of the virtualization backend. Although custom built for miriaPOD, the applied principles can be easily extended to other applications where memory sharing between guest systems is possible. A simulator is implemented to determine the performance of the proposed allocator compared to standard allocation algorithms.