On Wed, Mar 25, 2009 at 09:03:10PM +0100, Casper.Dik at Sun.COM wrote:
> I have no proof that AI will be better than jumpstart; the people 
> designing don't even know how people use jumpstart.  The AI documents 
> alone are a proof of that.  This is all sounds like designing from an
> ivory tower.  if you have not installed 100s of systems with jumpstart, 
> and run those systems in production (i.e., it wasn't some lab exercise at 
> Sun), then you're probably not the right person to redesign jumpstart.

I don't know about AI and ivory towers, but I agree with the rest of
your comment and would like to validate it with a tale from the past.

In a previous life I've built tools around jumpstart to automate the
concurrent installation of as many clients as the network will handle.
Which means: *ZERO* interaction.  The user makes gives the tool a list
of hostnames and off it goes.  This tool took care of installing
packages appropriate to the host type, and it took care of kicking off
"boot net - install" via a terminal server.  It supported jumpstart-in-
the-rack and jumpstart-in-the-lab (the first wave of servers had to be
installed before the network was available in their racks -- it was a
brand new site).  I'm glossing over a lot of details, obviously.  What
matters is it was simpler to install hundreds of customized hosts with
this tool than with JumpStart unmodified.

One reason I built such a tool: when I did it we needed to install some
600 systems in a short period of time, and the idea of interacting with
add_install_client 600 times was out of the question -- it's too error
prone, and anyways just painful.

Another reason: I wanted to be able to re-install everything in a flash
if we had to, the sort of thing one might have to do in case of a
disaster (think 9/11), which never goes as planned.

> To take existing software is sometime hard to develop it further.
> 
> However, continue to use existing software and developing has two huge 
> benefits:
> 
>       - it helps the current customers to take our new release
>       - it helps us fixing bugs in the current release
> 
> When you develop from tabula rasa you lose that; your end result must make 
> up for that.  I'm not seeing that.

I agree.

> Here's what I predict: someone will install OpenSolaris using the start
> and finish scripts from Solaris 10 because "AI doesn't do what I want".
> Because it can.

Heh.

Nico
-- 

Reply via email to