openstack team mailing list archive

[Pádraig Brady <P@xxxxxxxxxxxxxx>]

On 03/05/2013 11:03 AM, Balamurugan V G wrote:
> Hi,
> 
> I have a Folsom 2012.2 3 nodes setup like the one specified at https://github.com/mseknibilel/OpenStack-Folsom-Install-guide/blob/master/OpenStack_Folsom_Install_Guide_WebVersion.rst. Each node has a 1Tb hard disk with no raid.
> 
> Be it the image creation or the Instance creation, takes a long time(factor of the disk size of the image), about 15mins for a 20gb image. I can understand the slow image creation due to the http transfer of the file from a seperate web server hosting my images to the controller node. But even the instance creation takes that much or even a bit longer. I see that the image is copied from the Controller to Compute node and multiple copies are make in the compute node. So all this takes long an dis proportional to the image size. Is there any way to speed up this process? Is a SAN based backend the only way to go?

It sounds like you're using raw images throughout?
You might consider using qcow2 images in glance.
Then you can avoid the conversion to raw in the libvirt base
directory by setting force_raw_images=False in nova.conf
That will avoid some of the initial caching penalty.

Ensuring that you have use_cow_images=True set,
with use CoW images for the instances and improve
instance startup latency.

Details of the operations and tradeoffs involved are at:
http://www.pixelbeat.org/docs/openstack_libvirt_images/

thanks,
Pádraig.