Research And Implementation Of Key Infrastructure Cloud Algorithm

IaaS (Infrastructure as a Service)is in the bottom of the cloud computing services. And now, many IaaS platforms have been putting into use, so many users lead that the number of physical nodes in the IaaS cloud environment almost reach tens of thousands or even hundreds of thousands. On the premise of meetting user’s SLA, how to reduce the number of physical nodes to reduce energy consumption is a IaaS cloud platform’s problem to be solved urgently. Resource scheduling algorithm and load balancing algorithm have played a decisive role in solving this problem.At present, resources (this article refers to the physical node) scheduling algorithm and load balancing algorithm in the IaaS cloud environment have some drawbacks. On the one hand, when the resource scheduling algorithm search physical node which meets the resources demand of virtual machine, it need to judge the load of all the physical nodes, therefore, efficiency is lower; On the other hand, load balancing algorithm mainly consider raising the utilization rate of resources, but there are still problems:1) It do not consider reducing the number of physical nodes, there is a big waste of resources;2) The dynamic resource configuration operation of virtual machine which be used in load balancing algorithm can only solve the resource constraints of virtual machine itself, such as eliminating virtual machine’s hot spot, but can not change the load of the host physical node;3) The algorithm using the dynamic migration operation of virtual machine can easily lead a substantial load change of the host physical node and destination node, it is difficult to make the physical nodes’load close to the optimal state.To solve the above problems, we proposed a resource scheduling algorithm EPS (Efficient Part Search) and load balancing algorithm LRW (Lesser Resources Waste). The introduction of a new load balancing operation-the business load distribution in the LRW algorithm not only can multi-granularity adjust the load of the virtual machine, making host physical node’s load quantitatively change, but also be easy to eliminate the fragment of the physical node’s resources. The three kinds of balanced operations of business load distribution, dynamic resource allocation and virtual machine migration can effectively improve the overall resource utilization and reduce the number of physical nodes. It determines the load of each node and the overall load saturation by defining the critical value and the best point of the physical node’s load and the adjustment critical value of the virtual machine and system load saturation etc. and then divide all the physical nodes into six "state-pool"(cold-pool, accommodated-pool, fragment-pool, optimum-pool, risk-pool and hot-pool). The main research works are as follows:1) Research and implementation of physical node scheduling algorithm. When the virtual machine’s request arrives, it searchs the physical node what meets the conditions from accommodate-pool and cold-pool to create the instance of the virtual machine. if not meet, then open a new physical node;2) Research and implementation of load balancing algorithm. The load balancing algorithm orderly deals with each state-pool expect the optimum-pool. It selects the appropriate operation from three balanced operations to process adjustment in accordance with the current load of physical nodes in the state-pool and system load saturation, as much as possible making the load of physical nodes in the state-pool be close to the optimal state to reduce the waste of physical node’s resources and decrease the overall physical nodes’ number.Finally, test and verify the physical node scheduling algorithm and load balancing algorithm on the NwuC cloud platform. The experimental result demonstrates the feasibility and effectiveness of the EPS algorithm and the LRW algorithm.