Research On Virtual Machine Placement For Energy And Access Latency Optimization In Cloud Data Center

In recent years,with the rapid development of Internet technology,the emerging commercial model:cloud computing has been widely used in IT enterprise program as an on-demand allocation of compute and storage resources.With the cloud computing is widely used in IT companies,cloud computing data center is growing.In the face of large-scale data centers,cloud service providers(CSPs)face two challenges.How to allocate virtual machines in order to:1)reducing energy consumption of data center;2)improving the Quality of Service(QoS),i.e.reducing data access delay,to meet user's needs.Previous research on data center resource allocation is often focused on one of the two objectives.It is noteworthy that the two objectives are mutually exclusive in some cases,and the existing allocation methods cannot be effective to optimize energy consumption and data access delay.In this case,this paper proposes a virtual machine placement strategy that simultaneously optimizes these two goals in order to better meet the needs of CSPs.Therefore,the main research content of this paper is how to effectively allocate data center resources to optimize the total energy consumption of the data center computing node(i.e.physical machine)and data access delay.Firstly,this paper formalize the virtual machine placement problem of energy consumption and data access delay optimization in cloud computing data center,and transform this bi-objective problem to a single objective(evaluation value)by setting a tradeoff weight between these two objectives.According to the research ideas from shallow to deep,this paper first studies virtual machine placement under static environment.In this case,only the virtual machine placement at fixed time is considered,and the virtual machine placement is completed at one time.According to the candidate set of nodes,this paper proposes a real ability of nodes based iterative greedy(RCG)algorithm.The RCG algorithm first tasks the maximum data access delay as a threshold.According to the real resources(virtual machines)of compute nodes(CNs),The RCG algorithm searches the placement strategy to minimize the number of active CNs,while satisfying the data access delay constraint.And then continue to reduce the threshold,RCG algorithm implement the above process,iteratively,until under the current threshold,a feasible allocation strategy does not exist.During the iteration process,the placement strategy with the smallest evaluation value is saved.Secondly,based on the study of virtual machine placement in static environment,this paper further studies the virtual machine consolidation under the dynamic environment,which is constantly allocated and reclaimed with the arrival and departure of the task.Based on the RCG algorithm,this paper proposes a multi-agent negotiation based virtual machine consolidation algorithm(MAC)in dynamic environment.The MAC algorithm assigns an agent to each CN.The agents themselves have some nature to facilitate negotiation(cooperation).Based on their own resource utilization and current maximum data access latency,agents carry out virtual machines migration.If an agent detects that the data access delay generated by the current virtual machine assigned,is equal to the maximum data access delay,it will request to other agents to migrate the virtual machine.According to the change of evaluation value,agents decide whether to execute the negotiation process.Then,the convergence of MAC algorithm is proved by analyzing the time complexity of the algorithm.Finally,the validity of the algorithm is verified by simulation experiments.In the case of static environment,the experimental results show that the energy consumption,maximum data access delay and evaluation value of RCG algorithm are close to optimization tool(OS),and superior to other contrast methods.Simulation experiments in dynamic environment show that the energy consumption and evaluation value of MAC algorithm are better than other contrast methods,and the maximum data access delay is slightly larger than the DRA algorithm.