Cloud platform scheduling strategy based on virtual machine resource behaviour analysis
Aobing Sun, Tongkai Ji, Jun Wang · International Journal of High Performance Computing and Networking · 2016
Virtual machines (VMs) are the main scheduling and management objects of cloud computing platform. Currently, it is short of an efficient scheduling strategy for virtual machines' motion (VMotion) to guarantee their QoS and avoid the 'rolling snowball effect' of whole cloud platform with high resource occupation rate. In this paper, we present our VMotion scheduling strategy based on the analysis of VMs' resource access behaviour. According to the monitoring data of VMs, we can acquire the property curve of VMs' resource behaviour including CPU, disk I/O, net I/O usage, etc. of one day. Through processing the curve with filtering and segmentation algorithm, the movable span of one VM can be determined. We add a pre-motion step for VMotion to forecast the host's CPU, disk and network I/O through the overlapping of VM's curves to avoid the motions of VMs will not affect their QoS each other so as to improve the QoS of whole cloud platform, especially when the resource occupation rate of cloud computing platform keeps at a high level. The resource behaviour can also be used to monitor the abnormal exceptions of VMs for security.