Keith Mitchell wrote:
> 
> 
> Alok Aggarwal wrote:
>>
>> On Tue, 7 Jul 2009, Karen Tung wrote:
>>
>>>>> There's an assumption here that a "bootable AI iso"
>>>>> will be a separate entity from the current AI iso.
>>>>> Is there any reason that the current AI iso can't
>>>>> be made bootable with the end result being that we
>>>>> have a single AI image?
>>> The bootable AI image could be the same as the "regular" AI image, it 
>>> all depends on
>>> how we code it.  Right now, the GRUB menu for an AI image is 
>>> re-created during
>>> installadm time anyway.  The one from the image is not used.
>>> So, it is not hard to make the grub menu
>>> entries that will help VMC to boot right,  and still use the same 
>>> image for both
>>> bootable AI and "regular" AI.
>>
>> Okay, that sounds reasonable.
>>
>>>> Yes. The assumption for a bootable AI was that it would install a 
>>>> liveCD set of packages (as the VMC project needed a self-contained 
>>>> AI, that didn't access IPS). So the bootable AI would be, in my 
>>>> assumption, based off the liveCD (in terms of installed packages).
>>> I don't think that's a valid assumption.  The VMC project will be 
>>> using the bootable AI image
>>> to install whatever it is specified in the AI manifest from IPS.
>>
>> I agree, although I think the VMC func spec needs to call
>> that out more clearly.
>>
>> Currently sections 1.4.1 and 1.5 of the spec seem to imply
>> that an IPS based installation is the mode of operation
>> from within the bootable AI image. But section 5.1 seems
>> contradict that and says it could be cpio or IPS based.
>>
>> Alok
>  From a Virtual Machine standpoint, I think the cpio-based version would 
> be preferred. The references to IPS are because the existing networked 
> AI installs via IPS, and we were basing some assumptions off of that. 
> But, installing from a remote repository requires extra set-up to make 
> sure the VM can contact the repo. Granted, that set-up should be 
> straightforward. However, I thought that a reason for having a 
> media-based AI was so that one could do an automated installation onto 
> systems that don't necessarily have network access - in which case cpio 
> would make more sense, yes?

I believe the reason was purely to avoid the overhead of setting up and 
AI service here, not to avoid using IPS repositories.

Dave


Reply via email to