Abstract
There is an increasing demand of HPC-aware clouds to perform High Performance Computations rather than HPC-unaware clouds. Since renting saves lot of cost and time, people are moving towards cloud environments rather than owning costly supercomputers. There are many cloud providers in the market, and hence there is much of research going on for providing HPC-aware clouds, as HPC applications requires more computations compared to other applications. In this paper we describe an approach to allocate cloud resources by identifying the type of the virtual machine (storage-intensive or compute-intensive). We also present an algorithm which statically schedules the virtual machines (VM) to the CPUs based on CPU affinity. We observed that the performance of HPC applications is consistently better by statically scheduling the VMs allocation, when compared with default host operating system scheduling policy on an environment with multi-tenancy and heterogeneity. We obtain a speedup of upto 10% with our static scheduling when compared with default host operating system scheduling policy.