Research On Virtual Machine Placement In Network-on-Chip

As one of the key technologies in cloud computing,the research on virtualization has been attracted widespread attention in recent years.There exists some critical issues such as Virtual resources allocation,virtual machine deployment,migration and servers consolidation.Most of current researches are only focused on how to solve the problem of virtual resources allocation between servers.However,with the high performance computer employed with multi-core architecture becoming popularization,the virtual machine resources allocation in a single server becomes a new hot topic.Therefore,how to deploy virtual machines in the multi-core chip with NoC architecture in order to realize the load balance of on-chip resources and energy saving will become a challenging problem.In this thesis,the key problem of virtual machine placement is analyzed firstly which focus on a single multi-core chip.And then we introduce the virtualization technology based on Network-on-Chip(NoC)architecture in current researches.In addition,the general process of virtual machine placement on NoC is given after the virtual machine placement model and classical optimization algorithms are summarized.On these bases,we construct the placement model and constraints with the objective of energy efficiency.By this way,the problem of virtual machine placement is translated into an optimization problem,which can be solved by designing an improved ant colony optimization algorithm.This algorithm enhances the ability of ants to explore the feasible solution by introducing a pheromone diffusion model.Thus,slow convergence speed and easily falls into stagnation of the solution in traditional ant colony algorithm could be avoided.Moreover,since the cloud computing environment is usually large scale and dynamic,we make a virtual machine placement model for jointly considering energy-aware and load balance problems.Given on the constraints of computing ability and storage ability of process cores,a dynamic multi-object algorithm is proposed for this model based on Markov approximation optimization theory.And the corresponding algorithm schedule is designed to finish the virtual machine placement.Finally,the simulation platform is built with Microsoft Visual C++.Simulation results show that our scheme achieves much higher energy efficiency,load balance and calculation speed compared with previous schemes under different network sizes and load traffic models.