I’ve been working on drivers for unified I/O over a high speed fabric for a
while now and the marketing folks really would like to see booting from the
fabric in our first release again.
Last fall we had a discussion here about the requirements for a driver to be
the boot storage device, and a couple people who had been there done that
basically said “The device needs to be in the PnP tree of a real piece of
hardware”.
I saw the demo of iSCSI booting at WinHEC and have been pondering on what
changes must be in w2k3 R2 to make that work. It offhand seems like the
network stack must now be set to load via ntldr, and the iSCSI initiator is
NOT in the PnP tree of the NIC? I assume the iSCSI initiator must be root
enumerated, and connects sidways via TDI to the TCP stack. And this is
apparently capable of booting (and I assume shutting down).
Can anybody confirm that OS’s before R2 must have the boot device in the PnP
tree of real hardware, and R2 and later don’t? Some tests I had done earlier
seemed to show that a root enumerated boot load device in the correct class,
did start early enough to be a boot device, so am not sure why people here
said a boot device must be in a PnP tree of real hardware.
Thanks.
- Jan