Managing boot process remotely
What I mean by 'managing boot process' is the following:
1) Normally the OS should load from the hard disk
2) If the server's hard disk failed or boot config is screwed, be able to load from other media and provide access to hard disk after it is loaded
3) If nothing else helps, be able to re-install the OS remotely
All of the above should be done remotely without requiring any attention from datacenter personal. Let's assume that we have at least remote power control of the system, so we can reboot it if it hangs.
So far I came up with the following ideas:
A) Some sort of remote management hardware like Compaq Remote Insight board. Very nice reviews and impressive feature set, including the ability to load your server from a floppy in your machine 5,000 miles away. Solves all problems mentioned above (by the way, will anyone share the experience?)
One problem I have is I can't confirm with Compaq if this board works with non-Compaq servers. Does anybody know? Do you know any similar hardware solution from other vendors?
Another issue: if you buy your servers and send them directly to co-location provider without actually seeing them, you will need someone, either the server's seller or your colo provider, to install this for you. Can be as expensive as the board itself.
B) Boot the OS from the network using NIC's BIOS. By my opinion it is not an option, as it does not allow the owner of the server, who only has access to servers hard disk, to modify/upgrade their OS to their needs.
It is nice thing to have NIC boot if you manage an enterprise with standard OS load across all your machines, but in case of dedicated servers every server's owner has their own needs, wants specific OS versions and kernel patches etc., maybe even different OSes.
Again, not all NICs come with its own BIOS, you may need to buy EEPROM separately (it is cheap) and send to your colo provider to install it for you (that could be expensive).
C) Have a boot floppy with write protection (or bootable CD) permanently installed in the server with the following boot options:
(i) boot normally from the hard disk (default)
(ii) boot from the network
(iii) re-install the OS from the network
and make it to boot OS from the hard disk by default.
While I don't have a problem creating such a disk, I am struggling to find the best way to remotely control which of the entries to be booted.
I was looking into two solutions for this:
(a) Remote KVM switch from the likes of Raritan. Again, it is a hardware solution, expensive.
(b) Make the boot floppy mentioned above to check some file off the network and behave depending on its existence/content. I think this could be achieved by tweaking Bart's Network Boot Disk and using loadlin, but this requires some work.
So, I would like you to hear your thoughts about all this. By my opinion such solution is badly needed.
