Translated Abstract
In the IaaS cloud computing environment, due to the dynamic changs in virtual machine load, the hosts in the data center are prone to load imbalance. The load imbalance problem will reduce the performance of cloud applications, resulting in SLA violation, and cause waste of resources. This problem can be resolved by migrating the virtual machine from a heavily loaded host to a lighter loaded host. The traditional virtual machine scheduling methods for load balancing which ignore the applications running in the virtual machine, only regard the virtual machine as a black box. These methods are not well suited to frequent changs in load in the cloud computing environment due to the ignoring of potential resource contention between virtual machines. The researches on the above problem are as follows:
Firstly, a VMI-based virtual machine application sensing method is proposed. The method deploys an application sensing service on each host, which uses LibVMI, a virtual machine introspection library to read the memory data of virtual machine, and then uses Volatility, a memory analysis tool to extract the applications running in virtual machine from the raw memory data of virtual machine by semantic reconstruction. For the dynamic creation and migration of virtual machines in the cloud computing environment, an event-driven automatic configuration method for application sensing service is designed, which makes the application sensing working normally in the dynamic IaaS environment. By extending the existing functions of Ceilometer, the collection and storage of application sensing data is realized.
Then, based on the aformentioned virtual machine application sensing method, an application-aware virtual machine scheduling method is proposed. This method uses live migration of virtual machines to solve the host load imbalance problem, and reduces the probability of potential resource contention between co-located virtual machines by making virtual machines with relatively large differences in running applications co-located, and thus reduces the potential for host overloading.
In the real environment, the VMI-based virtual machine application sensing method is tested, the experimental results show that this method can senses the applications of virtual machines accurately, reliably and timely. Simulation results with CloudSim show that the application-aware virtual machine scheduling method can effectively reduce the number of overloaded hosts, reduce the number of migrated virtual machines, and reduce the SLA violations in the case of frequent load changes. And it can achieve a better load balancing in multi resources such as the CPU, memory and bandwidth.
Corresponding authors email
