Article Preview
TopIntroduction
With the popularity of cloud computing, the scale of data center is also increasing rapidly. Confront with huge cluster, it is an urgent problem to meet users’ needs to reasonably configure virtual resource in accordance with system operation status based on energy saving. As a manner to adequately utilize computing resource and optimizing performance of data center, the resource virtualization in cloud environment attracts more and more attention. In the Infrastructure as a Service (IaaS) cloud computing, one computer can be virtualized as several logical computers. The VM operates on different physical machine as stand-alone unit, which can also migrate among physical machines. Briefly pausing VM services, the VM can be transmitted from one physical machine to another. The mechanism enables dynamically adjustment of VM deployment based on system operation status, so as to adapt to service and computation changing, thus adequately utilizing computing resources in the system (Canali, & Lancellotti, 2014).
Many researches regarded dynamic VM management as an optimization problem, such as users’ QoS optimizing or energy consumption minimization (Karve et al., 2006; Hermenier et al., 2009; Jung et al., 2010; Goudarzi et al., 2012). There are also some research results based on saving migration cost and system performance. The VM placement algorithm in (Pagès et al., 2014) is based on migration benefit maximization, which is triggered in case of resource utilization exceeds the threshold. Aiming at same goal, the method in (Wood, Shenoy, Venkataramani, & Yousif, 2007). performs VM migration by targeting at reducing SLA violation and avoiding too high resource utilization of physical node. The method migrates VM from physical machine with the largest load to the smallest one, so as to reduce data amount generated in migration as possible. In order to satisfy SLA goal of dynamic work load, the dynamically remapping algorithm between physical machine and VM was put forwarded (Bobroff, Kochut, & Beaty, 2007). The application placement problem in the cluster among heterogeneous physical nodes considering about energy consumption and migration cost perception were studied (Goudarzi et al., 2012), where the improved first adaptation descending heuristics algorithm was used to obtain the local optimal solution. The VM placement is optimized by constraining programming mode, while striving to minimize migration cost of VM re-placement.
The point-to-point migration of VM is also an important topic in the VM dynamic management. Many systems stop service temporarily and copy status data to the destination host (Whitaker et al., 2004; Sapuntzakis et al., 2002; Kozuch, & Satyanarayanan, 2002; Osman et al., 2002). After migration completes, the VMs on the destination server will restart. However, the VMs cannot provide service within the migration period. Other methods tried to shorten downtime of VM by merely transmitting process groups (Osman et al., 2002; Nelson et al., 2005), which also result in long time of out of service. In order not to interrupt service in the migration process, the concept of pre-copy was introduced to improve the stop copy algorithm (Clark et al., 2005). It performs VM migration while providing service. On this basis, the VM migration program on WAN was also studied (Whitaker et al., 2004; Clark et al., 2005).
It can be known from above studies that most VM migration method only targets on one or two goals. In order to achieve goals of improving QoS, reducing resource competition and decreasing energy consumption, the multi-objective optimization VM dynamic management method is proposed in the work. The method mainly solves the VM re-allocate problem caused by change of system condition or dynamic application load.
The rest of the paper is organized as follows: Section 2 discusses the selection problem of VM migration. Section 3 designs the destination host selection algorithm. Section 4 performs experiment and Section 5 concludes this paper.