Re: [tlug] auto-shutdown at certain temp, testing, and CF boot

[Joe Larabell <fred62@???>]

I'd add a third possibility:

> PROBLEM
>
> Making a system with a 3 disk RAID5 array bootable in the even of disk
> failure.
>
> METHOD 1
>
> Create a partition on the start of all three drives and RAID1 mirror the
> partitions[1]....
>
> METHOD 2
>
> Get an IDE flash card[3] place it on your first IDE port and put your
> boot[4] partition and MBR on it...

METHOD 3

Take the simple way out and assume that once the disk fails, someone will
have to visit the machine in person to replace the disk anyway so at that
time they can also reboot the system from alternate media. Obviously, in
the case of an unattended machine that simply *must* come back up on its
own despite a failed disk, something else is needed. In my case, I really
use raid as an additional later of protection for my data. It's a desktop
machine so I can easily jam in a boot floppy in the case of a catastrophic
failure. And because it's cool. YMMV.

Nonetheless, for the sake of completeness...

> 1. Partitions in a RAID1 mirrors can be directly booted from grub, lilo etc.

I didn't know that. I'm skeptical about lilo, though. I thought grub's
claim to fame was that it could read a real file system even without the
kernel's help. The bootloader needs a lot more code than what can easily
be stuffed into the MBR before it can access a raid array, no?

> 2. You will also need to install a MBR on each drive.

How does the BIOS know which MBR to load? In this case grub won't help
because the MBR has to be loaded first. Are there BIOS'es now which will
fail-over to another physical device if the primary boot device is dead?

And, it the BIOS *is* that smart, would it not also be possible to keep
a boot CD in the CDR drive while booting primarily off the non-redundant
/boot partition? That keeps things simpler but still provides fairly good
redundant boot capability in the case of the sudden death of the primary
boot device. It's probably easier to convince the machine to boot from a
CD than from a flash.

> 4. Boot partitions are mostly read-only so good flash memory should have
> a greater MTBF

That's a very good point. No moving parts. But it's still a single-point
failure that *could* make the machine unbootable. Then again, the CPU or
power supply is probably just as likely to go out, eh?

---
Joseph L (Joe) Larabell            Never fight with a dragon
http://larabell.org                     for thou art crunchy
                                  and goest well with cheese.