Under The Cloud Server Consolidation Algorithm Based On Shared Memory Research

Main stream virtualization products provide content-based page sharing, which identifies virtual machine memory pages with identical content and implement the pages with a single shared page. This technique, implemented at the host level, can applies to virtual machines placed on a given physical machine and reduce memory utilization of the physical machine. In cloud computing data center, two virtual machines holding lots of identical pages and residing on different physical machines will cause sharing lost. Consolidation based on page sharing can solve the problem. It monitors page value of all virtual machines in the data center and try to place virtual machines with similar memory into one physical machine. It can maximize the benefit of page sharing, increase memory utilization rate of data center, meet more virtual machines demand of memory and make cloud provider get more profit.In this paper, I improve existing page sharing based consolidation algorithm. I compare several data structures of set and choose bloom filter to present set of memory fingerprint. Bloom filter can efficiently determine the memory footprint of virtual machine. In addition, I gather lots of real data, analyze memory change rate under different memory utilization rate, get the relationship between the two rate. I propose two methods to reduce network traffic:bloom filter redundancy elimination and filtering out virtual machines with high memory utilization. I design the page sharing based consolidation algorithm, which take the memory sharing relationship between virtual machine being migrated and destination server into account. The algorithm is improved from first fit decreasing algorithm, but the number of active server is smaller than which of FFD.Evaluation of the algorithm demonstrates that page sharing based consolidation increase resource utilization compared to plane consolidation. To same initial virtual machine placement configuration, the number of active servers used by algorithms proposed in this paper is6percent less than that used by FFD and13percent less than BFD. The number of virtual machine migration is9percent less than that of BFD. In addition, the network traffic is more than50percent less than that of consolidation algorithm without redundancy elimination or high utilization virtual machine filtering out.