"What would you say ya *do* here?"

David Huff dhuff at redhat.com
Wed Jan 20 15:45:34 UTC 2010

On 01/20/2010 08:50 AM, Greg DeKoenigsberg wrote:
> :)
> Right now we face a dilemma:
>     * The latest Fedora kernel available on Amazon EC2 is Fedora 8;
>     * Various people are creating "Fedora images" based on that AKI;
>     * Confusion results, e.g.:
> http://stackoverflow.com/questions/448794/how-does-an-amazon-ec2-instance-select-its-kernel
>     * Which (a) doesn't work and (b) gives Fedora a bad name in the process.
> We have an opportunity to fix this.  Justin Forbes has been working with
> some folks at Amazon to make this better, and it looks like we'll have an
> "official" account that will allow us to do something clever, like put a
> Gold Star next to "official" Fedora EC2 images.
> Once we actually get them up and running, of course.  Which, painted in
> broad strokes, maybe looks something like this:
> 1. Getting a basic image uploaded and working for anybody (Justin is on
> this, but maybe could use some help?);

So there are really several parts here....

a. create ami: Amazon Machine Image -> pre installed fedora image.  This 
is the easy part.  There are several tools for building pre-installed 
images form a ks file.  I have been working with the SPIN sig to define 
a minimal ks file (AOS) that describes minimal Fedora Image with dhcp, 
yum, selinux, that would work here.  We are also currently working to 
include a fedora-mini.ks file that can me extended for other "SPINS".

b. create aki, Amazon Machine Image, In the past I have just copied the 
vmlinuz file form a running image.  Is there a better more reliable way 
to do this?

c. create ari, Amazon Ramdisk Image. this needs a custom fstab per 
machine type, m1.small(32bit) vs. m1.large (64bit) etc, and needs the 
xen modules preloaded. Is there a good way to automate this?

d. edit image attributes of all of the above to 1)link ramdisk and 
kernel to ami, and 2)share with the world

I would be willing to help out here a much as possible however we need 
suggestions of good ways of automating the above.

> 2. Getting a basic image uploaded and working for everybody (which means
> coordinating a testing account for Fedora people to use free-of-charge,
> which we have funding for, and then finding actual people to test);

I can help test if needed

> 3. Getting that basic image created as part of the Fedora release process
> (which was discussed at FUDCon, but I don't know if any of those
> conversations continued);

This is just a matter of getting the stuff I described above into the 
Fedora release process which I am not real familiar with.  I have some 
suggestions here, related to how we build the ovirt node, but will wait 
to see if others have better ideas first.

> 4. Providing proper tools for people to create their own images, since
> creation of a custom "spin" is basically an essential activity in the
> cloud.

Im not going to touch this yet however I have some good ideas here as well.

> It may be that some of this activity is going on already, and that's a
> large part of what this SIG is about: to find those people and get them
> talking in one place, away from the clutter of fedora-devel and other
> lists.  So if you're working on tools that will help with this goal, by
> all means, speak up.
> And now, a couple of questions.
> * For Justin.  How's your current image looking?  Does it work well enough
> for other folks to try it out?  If not, what's the issue, and can anyone
> help?
> * For everyone.  Have you played with EC2 yet?  If not, do you have a
> side project that could benefit from its use, that furthers the goals of
> Fedora / free software / etc., and are willing to test our images?  If so,
> let us know -- we've got an account for this purpose, and can send you the
> info needed to get started.  Even if we don't have the latest Fedora up
> yet, there are plenty of images that you can play with to get a sense of
> how EC2 works.

Yes, and I would like to see the AOS/thincrust project be the base of 
these images.

> * For everyone.  Would an IRC meeting to talk through some of this stuff
> be useful -- sort of a kickoff where everyone takes an hour out of their
> life to think about all this -- or should we stick to the mailing list for
> now?

I could go either way here


More information about the cloud mailing list