Where to start.

Images today are one of the other major stumbling blocks for most
users of clouds -- not just building, but the versioning and
distribution of the images are still too painful.  People revert back
to using traditional configuration management tools, because
re-rolling a server image is a time consuming and painful procedure.

In addition, there is no abstraction layer between generating the most
common image format, Amazon's AMI, and other formats, preventing wider
adoption and support of image formats by other hosting providers.

I have talked about it on IRC in #libcloud, and in person with a few
people, about maybe expanding the scope of the libcloud project, to
include "Cloud Images".

I think if libcloud took the same approach to images that it took to
APIs -- start simple, support everyone, get the hosting providers
involved, and iterate quickly, we could get good adoption of an image
format and specification quite quickly, where other complicated
standards have faltered.   In addition, because libcloud is about
producing actual software, not just a specification, I feel like we
could provide the tools to providers to make integration of the image
format extremely simple.

I think the most interesting project in this space right now is
dotCloud's cloudlet:
http://bitbucket.org/dotcloud/cloudlets/src/

The short version is they use a filesystem, integrated with version
control, and that is able to export to various formats, from a chroot,
to a tarball, or an AMI.  There is also support for templating things
like configuration files inside the image, which will make it very
easy to support multiple cloud providers with a single baseline image.

I think the most interesting possibility is to build on Cloudlet
(maybe even convince cloudlet's authors to join us inside the ASF),
but I am getting ahead of myself there :-)

What does everyone else think about Cloud Images?  Does it belong in a
place like libcloud, or do you feel that the standardization efforts
elsewhere are sufficient?

Thanks,

Paul

Reply via email to