[rescue] x86 vs. SPARC BIOSes (was: Small servers (was Re: WTT: 1.5G of PC2700 for 1G of PC100))

[Andrew Gaylard]

On Thu, May 8, 2008 at 4:07 AM, Shannon Hendrix <shannon at widomaker.com>
wrote:

> On May 5, 2008, at 16:15 , Ethan O'Toole wrote:
>
>  I don't mind the PC stuff really. I mean, once a server is setup and
>> deployed you don't really have to deal with it that much. Even on Sun and
>> SGI boxes you don't really mess with the boot prom stuff once it's up and
>> running.
>>
>
> The issue comes at 0400 when you have to drive out to a server, or even fly
> out to it, all because they could not spend $1.50 in quantity to have a real
> console.
>

I agree.  And for me, it can take *days * to travel to some of my machines.
And transporting the video data requires links that are fast & fat (by our
standards, anyway).  Obviously, since these boxes are in remote places, the
cost of such links is prohibitive.  And the rest of the time, we don't
*need* video.  So a serial BIOS is essential.

However, the reality is that x86 boxes are built to run Windows, and Windows
assumes a keyboard, mouse, and hi-res bitmapped display.  So the x86 BIOS
writers can assume the same I/O will be available.  This is a major reason
why I favour SPARC machines -- they were never designed under this
assumption.

Some of the things that drive me nuts about x86 BIOSes are:

   - There's no command-line, so there's no way to script a setup.  OK, you
   can capture the settings with dmidecode, but how do you enter them
   programmatically? (The OS hasn't booted the first time yet, remember.)
   - There's no standard way to enter the menu system.  It could be <DEL>,
   or <F11> or who knows what.
   - The menus contain all kind of stuff that you don't understand (what
   does "optimised settings" mean?  And why wouldn't I want it?)
   - BIOSes on motherboards, video boards, RAID cards (don't get me started
   on RAID cards...) all assume that they can clear the screen, wiping out that
   vital error that appeared for a brief, tantalising, moment.  Why can' they
   just scroll?
   - How do you tell an x86 box the equivalent of "setenv auto-boot? false",
   so that it comes up in the BIOS, where you can run "probe-scsi", "test-all",
   and then "boot"?

And here are some of the things that I love about the OBP:

   - emacs keybindings
   - devalias: "boot disk" can point whereever.
   - diag-mode
   - FORTH: you can actually *program* the darn thing without booting an OS.
   Or even having disks or CDROMs plugged in.  And F-code is portable to other
   CPUs, so it should suit the peripheral manufacturers too.
   - Plugin-cards can extend the boot environment with their own F-code.  So
   the basic PROM can be lean & mean. Video cards, SCSI cards, etc.can do what
   they need to.
   - TFTP net-booting (actually, HP's ILO feature where it boots from a
   http:// URL is pretty neat)
   - OBP is designed to be upgradeable, without needing to boot an OS.

IEEE 1275-1994 must be the most under-rated feature of the SPARC boxen.

OK, enough now.  Back to work.  Sorry, I get carried away...

Andrew