Research On Key Technologies Of Resource Scheduling In Virtual Computing Environnment

In the age of cloud computing, with the increasing of application scale, computing e?ciency has become a huge problem, and the running of virtual computing environment has got more and more attention from academia and industry. A reasonable deployment and scheduling of virtual machines(VMs) can not only ensure stable and e?ective running of virtual computing environment, but also a?ect the utilization of physical resource directly. Fast accessing of image file and excellent network supporting are essential to further optimize the running e?ciency of VMs. Besides, they are also the basic guarantee for providing quality services to end users by virtual computing environment. In this thesis, we address some of the most important challenges of resource scheduling in virtual computing environment, which involve virtual machine scheduling, virtual cluster constructing, fast image accessing and incremental updating of network rules. The main contribution of this dissertation is summarized as follow:(1)We propose a real-time scheduler for virtual machine. Our scheduler introduces a new VCPU priority called urgent to identify real-time applications, re-designs the transforming relationship between VCPU states, and shortens the VCPU slice which is in urgent status. The implementation is based on Credit Scheduler used in Xen. Our evaluations using ping latency and throughput of web servers show that the proposed scheduler can response real-time tasks rapidly, and increase the stability of response time.(2)We propose an I/O-conscious strategy for virtual cluster constructing. Taking into consideration the I/O characteristics, the proposed strategy can obtain better scheduling results with lower overhead. Moreover, an exponential smoothing algorithm is suggested,which combined with the proposed strategy, can further improve the scheduling results.By selecting spot threshold and smoothing factor, the experimental results show that our strategy can avoid the migration oscillation of virtual machine, and reduce the overhead of virtual machine scheduling.(3)We propose a deduplication storage method that has been particularly designed for large scale VM deployment. Its design provides fast VM deployment with peer-topeer(P2P) data transfer, and low storage consumption by means of deduplication on VM images. It also provides a comprehensive set of storage features including instant cloning for VM images, on-demand fetching through a network, and caching with local disks by copy-on-read techniques. Experiments show that the features of our proposed storage method perform well and introduce minor performance overhead.(4)We propose an incremental update method of network rules of virtual computing,which constructs VM network with SDN technology. Based on OpenFlow protocol, it builds rule dependency during the policy compilation, and maintains a scattered priority distribution by using k-factor strategy. Our evaluation demonstrates that the proposed method can eliminate nearly all the priority updates, therefore, the network latency of virtual computing environment can be reduced.