Haai,

On Wed, June 6, 2007 16:31, Jason Thorpe wrote:
>
>[snip]
>
> Well, sure, 20 years ago they were an improvement over what had
> existed previously (which is to say -- "nothing" ... did you ever use
> 4.3BSD on an hp300 and suffer through the hard-coded partition scheme
> in the disk drivers?)

Nope, that happened quite a few ticks before my life started...

>
> But by today's standards, they stink.

Guess I'm lucky I've got a cold...

> They're not extensible, contain
> too much useless data in them that has no meaning on modern disks,
> have limited data ranges that they can describe,

Like indicated earlier in the discussion, a new, simplified disklabel
format would solve all those problems on a whim.

> and have no standard
> layout even within NetBSD.

Why bother standardizing that? It's sysadmin-specific and you don't need
to bother with it once it's set up properly. Surely you wouldn't want to
dictate a fixed FS hierarchy and disk layout to everyone?

>>[snip]
> On the contrary.  GPT is an extensible partition map format that
> supports extremely large disks and an arbitrary number of partitions
> without requiring nesting or other nasty gunk.  And due to its use of
> GUIDs, it can uniquely identify both file systems as they move from
> one machine to another (for e.g. hot-plug external disks)

See other half of the discussion for an explanation why I don't think it
will be that useful.

> as well as
> uniquely identify file system formats without risk of numbering
> collision.

GUIDs are a myth. They are theoretically impossible to generate with
common hardware. If you insist, better just add a description field to the
second sectors of both the disklabel and partitions, and blatantly refuse
to set a default.

Then again, I absolutely see no reason for disks (or other devices, for
that matter) to be roaming without support for networking...

...again, an updated disklabel format will solve all the remaining of
those issues.

>
>> ...please, let's just let MBR die as nature intended :)
>
> The EFI standard that describes GPT includes a provision for a
> "protective MBR" that is essentially there for backward compatibility
> with non-GPT aware systems.

So we'll be stuck with MBR for the rest of the days of computing. Are the
trinary, analog, and isolinear versions standardized already?

> Its intended usage is to mark the entire
> disk as "reserved -- don't futz with it!".

*sigh* The fact Windoze and some other OSes are ill-behaved are the
problem of those OSes, _NOT_ others. The MBR we have to contain the
disklabel on some systems is bad enough already.

Or am I expected to dig up my copy of Norton Utilities 1.0 for Slowaris
every time I nuke a disk in a test system for some reason?

> The fact that it can be
> used in other clever ways is a useful side-effect in some situations,
> but is orthogonal to GPT's usefulness on its own.

Hm -- I haven't seen _any_ clever way as of yet.

Baai,

De Zeurkous
-----------

Friggin' Machines!

>
> -- thorpej
>
>