Regarding NFS, below are a few practical reasons to not use local hard disks to host the images as they run.

1) Say each image is 10GB, on a 72GB hard drive (the standard "large" drive in a blade), that means as a maximum consolidation you can only have 7 VM's on a blade. A blade can likely handle between 4 - 30 VMs depending on how powerful it is, so this would be a gross underutilization.

2) It takes a VERY long time to copy images to and from the blades (3-6 MB/s) because of the difficult file system translation between to and from the VMFS filesystem format on the hypervisor.

3) ESX 3i (the free, efficient hypervisor) is designed to boot as a disksless hypervisor, so ESX3i environments would be out which would reduce the number of folks who could do this dramaticlly

4) Although it seems performance counter-intuative, actually if booting from an actual storage system with multiple disks, VMs will perform quicker when booting over the network. Consider the HD thrashing that would occur if a single disk were really handling the I/ O operations of 20VMs on a single hypervisor. Believe it or not, the network is faster.

5) While not possible with VCL today, if the instance images live on the network natively, the opportunity to do things like Vmotion aren't totally out of the question, but those are not at all possible with local disk boot.

In regards to image re-using, your configuration is not correct. It sounds like you need to create a few more computer "slot" entries in the computer table and link those slots to your hypervisor entry in the computer table. The number of "slots" you have equals the number of virtual machines that can run concurrently. Notice it actually copies it from /golden/<image_name>/ to /inuse/<slot_name>. Once a slot is created, remember to associate it with a hypervisor through the "Virtual Hosts" area. If you still can't get this, post up some of the lines in your computer table's DB.


Brian Bouterse
Secure Open Systems Initiative

On May 6, 2009, at 2:07 AM, 乔木 wrote:


We have setup the environment for VM provision on ESXi recently. We are
confused about some mechanisms:

  - Images are stored in NFS

As I see, VCL stores images in NFS and just copy it locally when using it. It means that the hardisk in the esxi hypervisor will not be used. All the VMs are stored in remote computer and all the operations will be remote operation (CPU exsits in Hypervisor but hardisk exsits in NFS). I think it may lead performance problems. Although we can setup mutiple NFS servers, it
is not good that the hardisk in the hypervisor is not used and all the
operations will be remote in my opinion. Why not transfer the image to the
esxi hypervisor which will run the image?

  - Image reusing

When we use an image from repository, VCL does: 1. copy the image form
/golden/<image_name> to /inuse/<image_name>, 2. start the image. It's
difficult for image reusing. It's impossible for mutiple machines to use the same image because the copy operation will only copy one image to the /inuse folder and that image can be used only for one esxi hypervisor. Other copy operation will overwrite that image. Again, if we transfer the image to
hypervisor, there won't be such problem.

Best wishes,
MOE KLINNS Lab and SKLMS Lab, Xi'an Jiaotong University
Department of Computer Science and Technology, Xi’an Jiaotong University
TEL: 15991676983

Reply via email to