The CPU Overhead Evaluation System In The Virtualization Environment

In recent years, computing systems virtualization as a powerful and convenient technique has been widely used. In the virtualization system, the physical computing resources are no longer monopolized by a single operating system, but are shared by multiple virtual machines. Because of that, the virtualization introduces some CPU overhead which can't be ignored of the system itself. After the announcement of the concept of the CPU overhead in virtualization system, a series of questions come out. First of all, how to define the overhead and divide it from the normal utilization of CPU slice. The second question is how about the relationship between the overhead and the count of VMs, and the third one is how the different types of workload on the VMs would affect the overhead.Based on the purpose, we propose the principle to define the CPU overhead in the virtualization environment: the CPU overhead in the virtualization environment is the inevitable CPU overhead which is due to the introduction of virtualization technology and differ from the CPU overhead in traditional systems. Based on this principle, we point out two indicators to measure the CPU overhead: the overhead of scheduling between virtual machines and the additional overhead generated by the privileged domain agent executing some privileged instructions. By taking use of the Xentrace and Xenrelay as engine, the virtualization CPU overhead measuring system, VOMS, could accurately obtain the CPU usage of all the virtual machines, and the CPU overhead in Xen Virtual Machine Monitor during the measurement period. After formatting the measuring results, the system could dynamically reflect the usage of physical CPU and virtual CPU and the mapping relationship between them in the virtualization environment. In order to better research the different factors impact on this overhead, we design a test platform as the extensions of VOMS, making it much more convenient to research the relationship between the overhead and the number and workload types of virtual machine.By measuring the overhead and analyzing the relationship between the factors and the CPU overhead, the overhead exponential grow with the growth of the number of virtual machines. Comparing to the IO-intensive applications, compute-intensive applications in the same circumstances would generate a greater CPU overhead. When the virtual machine number reached 8 and all the virtual machines are running compute-intensive load, the overhead accounts for 14.7% of all computing resources, and with the increase of the number of virtual machines this overhead will keep growing. This law reveals how to allocate and balance the utilization of resources and the count of the effective computing resources, and give some proposals to the virtual machine designers and users.