Research On Reliability And Performance Optimization Of Virtual Machine In Cloud Environment

Cloud vendors provide cloud resources to users accurately because the cloud in-frastructure has high abilities in computing,communicating,and storing.As a key technology for cloud computing,virtualization enables more resources virtualized from the physical to the cloud objects.Traditionally,researchers only concerned about hardware virtualization that emulates hardware devices and provides virtualized hard-ware resources to cloud entities.In recent years,virtualization technology began to provide software virtualized resources.All these techniques guarantee the sufficient resources from cloud provider such that all the cloud users or tenants could maintain their Quality of Service(QoS).In recent years,with the increasing of the diversity and complexity from cloud Virtual Machine(VM),and the augmentation of cloud platform,how to better guaran-tee per-VM performance has become an essential research point.In this process,we found the following challenges:1.In cloud environment,because there are many mission-intensive and mission-critical applications,memory is frequently accessed.Thus,without memory high availability,a memory's crash will cause the termination of all in-memory activities.Unfortunately,the state of art for memory high availability employ specific,expensive hardware,which only monitors limited bits of memory error.As a result,how to design a universal and cost-effective software solution is a challenge for service quality in VM.2.Cloud VMs could run either CPU-intensive or GPU-intensive applications.Dif-ferent applications should acquire respective CPU or GPU resources.However,in such hybrid environment,most of the existing policies are designed based on sole resource.In other words,they could only schedule one kind of core but occupy too much computation of the other core.To this end,this paper will address them.3.Cloud network provides the fairness the bandwidth allocation according to per-user's payment.However,in real-world network environment,a malicious VM could increase its bandwidth by increasing the connections optionally.There-fore,the last challenge in this paper is to address the problem of the bandwidth manipulation.Motivated by the above,in this paper,we present the following work:1.We implement a memory High Availability(HA)architecture called kMemvisor.The design and implementation of kMemvisor are based on Direct Page Table(DPT)and Shadow Page Table(SPT).kMemvisor can provide system and appli-cation levels memory mirroring.On the other hand,kMemvisor can guarantee the mirror memory write synchronously when it writes native memory.2.We propose a GPU/CPU hybrid framwork,vHybrid,under multiple VMs en-vironment.vHybrid could allocate CPU and GPU resources to different ap-plications with no modification of guest OS kellel in the cloud environment.In particular,vHybrid designs and implements open-loop control algorithm and adaptive control algorithm for GPU-intensive application;SLA-aware algorithm for CPU-intensive application.3.We model the problem of bandwidth proportional allocation with Weight Node Graph(WNG)in multiple VMs environment.We propose the policy of of band-width proportional allocation,NetLoft,in multiple VMs environment.NetLoft includes two policies:NetLoft-R and NetLoft-G.NetLoft-R achieves per-VM network proportionality,while NetLoft-G achieves per-tenant network propor-tionality.In this paper,we have emulated the two policies using Transmission Control Protocol(TCP)stack.