Performance Research Of Virtual Machine Live Migration

In recent years, with the popularity of cloud services, users’ experience becomes the most important thing. Solution of load balancing and hot is needed to rely on the virtual machine live migration technology to ensure users experience. In the process of live migration, because of the large amount of information and frequent changes, the memory is the most complex part of the whole migration process, which determines the efficiency of the whole migration process.Now the mainstream method of live migration is pre-copy algorithm. In each round, the updated dirty pages in the memory are transmitted to the destination virtual machine in an iterative manner. Then stop source virtual machine to copy remaining dirty pages and other resources to the destination virtual machine until the amount of dirty pages is less than the threshold value.Because these remaining dirty pages are small enough, it makes the service interruption time short enough to ensure users’ experience.The traditional pre-copy algorithm transmits work set of dirty page granularity. When the dirty page rate is higher and the actual dirty data is less, the traditional pre-copy algorithm shows that the number of iterations is too much, the total migration time is too long, and too many network bandwidth is occupied. To solve these problems, the DIP algorithm that dirty data in page is work set is proposed in this paper. Compared with traditional pre-copy algorithm, work set granularity of DIP algorithm is smaller. It reflects actual migration environment more accurately. It makes less data transmitted each round. It makes working set converge to below of threshold value faster with the iterations. And it makes a higher migration efficiency. On this basis, this paper introduces the arithmetic coding and multithread to DIP to implements CDIP algorithm. CDIP has three different scenarios to use arithmetic coding. They are compared by experiments in this paper. And it finally get the CDIP algorithm of the optimal scene.At the end of this paper, it demonstrates that the DIP algorithm has higher migration efficiency than the traditional pre-copy algorithm by the experimental data first. Then it carries on the contrast experiment on those three different scenarios to get the optimal scene.It is suited to use arithmetic coding to compress the data of the first round of iteration because of it is large. The convergence speed of DIP algorithm isfast,and the data of the iteration is less after the first round, ifthe CDIP uses arithmetic coding to compress those data, it will slow down iteration’s convergence speed, and it will interfere with the iteration to end. Therefore, using the arithmetic coding in the first round of iteration is the best scene. It could further optimize the DIP algorithm to improve migration efficiency.