A method to manage VM startup
Publication Date: 2016-Nov-15
The IP.com Prior Art Database
As the development of cloud infastructure, it's is common to see hundreds or thousands of virtual manchies(VM) running on one hypervisor. For system administrator, it hard to start these vms in an efficient way. Comparing to known solution, a group and tactics control methold is introduced in this disclosure to solve this problem.
Page 01 of 5
A method to manage VM startup
In data center or cloud infrastructure maybe there are hundreds of virtual machines(VM) running on one hypervisor, for example,hundreds of Linux virtual machines running on a powerful system. There will be a need to power on/off all(or most of all) these virtual machines at the same time. For example, power off all virtual machines and put hypervisor for maintainence. It is well known that operating system starting up process will almost make full use of hardware resources, such as CPU, I/O and etc. This also applies to virtual machines on hypervisor. Although one of the virtualization advantages is to improve system utilization, but if each of the virtual machine make full use of its resources ,this will make the hypervisor system overload. Start all these virtual machines at the same time will cost much time. This will be a critical problem for hypervisor.
Although virtual machines can be started one by one(or ten by ten and so on) instead of starting them at the same time, there are still problems. First, start hundreds of virtual machines one by one is low performance. it will cost much more time. Second if some of the virtual machinesshare devices with each other, it will cost more time, especially when some of the virtual machines share I/O devices. For example, five virtual machines share one disk where all operating system(OS) are stored, if these five machines are powered on at the same time it will cost many time to start up all of them due to the I/O bottleneck.
This is a common scenario. Following is an architecture of hundreds of virtual machines running on a hypervisor. VM11 - VM1Z share Disk 1 to store there OS, VM21 - VM2Z share Disk 2 and so on till VMN1 - VMNZ. After the holidays, the actor needs to power on all of these VMs. If Actor sends power on command to all the VMs at the same time, it will cost a long time to put all VMs online because of hypervisor overload and I/O bottleneck, although there are no hypervisor overload or I/O bottleneck after all VMs have been started and running. If Actor chooses the opposite way, send power on command to all these VMs one by one(or ten by ten, or other), then Actor may meets two kinds of problems:
Hypervisor overload and I/O bottleneck if Actor does not wait long enough to send command to start another set of VM(or VMs). Take too long time to start all the VMs if Actor waits longer time to send command to start another VM(or VMs)
So it is impossible for Actor to start all VMs manually to get best performance and response. This innovation brings a method of power on virtual machines to avoid these problems.
Page 02 of 5
Picture 1 Old method of VM Management
There are several virtual machine management tools with which can start/shutdown multi virtual machines at the same time. However, none of them can solve above problems. Followings are two typical solution.
Start VMs one by one (or ten by ten). This solution...