| 1 |
On Wed, Aug 20, 2003 at 02:03:48PM +0930 or thereabouts, stephen white wrote: |
| 2 |
> I don't understand this reasoning, as this is like saying "not all |
| 3 |
> networks run zeroconf or want to run zeroconf"... DHCP is just a |
| 4 |
> program, and the gold server can run whatever programs it wants. The |
| 5 |
> network sits there and passes the packets. |
| 6 |
|
| 7 |
The other thing that is important to note is that, at this point, this |
| 8 |
installation script is spec'd out to have two distinct components: |
| 9 |
|
| 10 |
* a boot component |
| 11 |
* an installation component |
| 12 |
|
| 13 |
The boot component is responsible for obtaining an IP address and |
| 14 |
contacting the Gold Server for more information. That's really all it |
| 15 |
does. Then, the installation component takes over and does all of the |
| 16 |
heavy lifting, with most of the logic being stored server-side. |
| 17 |
|
| 18 |
The benefit of this is that anyone can write any boot component they want. |
| 19 |
Floppy boot with static IPs, PXE boot with dynamic IPs, whatever. |
| 20 |
|
| 21 |
So we can consider writing a reference implementation that uses zeroconf |
| 22 |
(which I personally like and hope we can get to work) but that does not |
| 23 |
preclude us (or someone else) from writing additional methods of booting. |
| 24 |
|
| 25 |
--kurt |
Archives