An Interference-aware Virtual Machine Placement Strategy for Small-scale HPC Applications in Clouds

Abstract

The cross-interference problem may occur when applications are executed in virtual machines placed in a same physical machine. Although many previous works have proposed several different strategies for Virtual Machine Placement, neither of them have employed a suitable method for predicting cross-interference nor have considered the minimization of the number of used physical machines at the same time. In this thesis, we define the Interference-aware Virtual Machine Placement Problem for small-scale HPC applications in Clouds (IVMPP) that tackles both problems by minimizing, at the same time, the cross-interference of small-scale HPC applications, that can share physical machines, and the number of physical machines used to allocate them. We propose a mathematical formulation and a strategy based on the Iterated Local Search framework to solve this problem. Moreover, we also propose a quantitative and multivariate model to predict interference for a set of applications allocated to the same physical machine. Experiments executed in a real scenario, by using applications from the oil and gas industry and the HPCC benchmark suite, showed that our method outperforms several heuristics from the related literature in terms of interference, while using the same number of physical machines.