Jon Aimone wrote:
> Hi,
> 
> I've been musing through the design of AI recently. AI and the 
> underlying tools, libraries, and such are a great improvement over the 
> nearly hidden code of the old Solaris installer. Yet AI still strikes me 
> as a monolithic solution to a modular and potentially iterative problem.
> 

Agreed.  The implementation you see right now is quite some ways from 
the intended final form; the current version leveraged what we had to 
get something available in OpenSolaris, but isn't providing an 
architecture for the long term.

We'll soon be publishing an updated architecture spec that describes the 
way in which all of the various existing and future installer pieces fit 
together.  That architecture is quite modular, leveraging the 
Distribution Constructor's execution engine to tie the elements 
together; that obviously has capabilities for checkpoint and restart 
that map to the use cases you're implying further on.  You may also see 
some additional SMF service decomposition as we work through the design, 
but I don't presently expect it to be the primary glue used since there 
are uses of the architecture (such as Distribution Constructor) that are 
not really appropriately modeled as system-wide services, which is the 
only scope supported for SMF service definitions.

Thanks for the thoughts, and do review the spec once we publish it; we 
certainly value additional perspectives.  Good news is that I think 
you'll see your decomposition is pretty similar to the one we envision :-)

Dave

Reply via email to