[OpenIndiana-discuss] rpool on 4k disks?

[Reginald Beardsley]

--- On Sun, 2/3/13, Jim Klimov <jimklimov at cos.ru> wrote:

> From: Jim Klimov <jimklimov at cos.ru>
> Subject: [OpenIndiana-discuss] rpool on 4k disks?
> To: "Discussion list for OpenIndiana" <openindiana-discuss at openindiana.org>
> Date: Sunday, February 3, 2013, 6:55 AM
> Hello all,
> 
>   While thinking about Reg's setup, I began wondering:
> are there
> any known issues with rpool using ashift=12 - perhaps in
> GRUB or
> in the rpool-importing code? Is there anything hard-coded
> for the
> ashift=9 on rpools (or equivalently limited in terms of
> available
> RAM)? That is, may we recommend certain workarounds to
> create the
> aligned partitions and the properly ashifted rpool to match
> the
> larger hardware sector, or would that lead to an unbootable
> system?
> 

I'm going to work on the wiki more.  But here's a quick summary of where I got into trouble.

The disk "geometry" is bogus, however, programs are still enforcing rules based on cylinder boundaries.

In the case of the ST2000DM001, a "cylinder" is not an integer number of 4k sectors.  Trying to fix this using prtvtoc and fmthard by editing the label information failed because fmthard would not accept slices that did not align to a "cylinder".

It doesn't seem to matter to OI_151a7 that the cylinders don't align w/ 4k sectors.  The ashift=12 seems to take care of that.  In the Solaris 10 case where ashift=9, that may have been why it failed to boot after the first time. I decided not to pursue using S10 further.
The N40L is using a Radeon graphics card, so hopefully I won't see the hangs that plague my nVidia based OI_151a5 system.

There is still some confusion about USB disk I need revisit now that I have a clearer idea of what's going on.  I may be able to make my 3 TB USB disk work now.

Given how far we are from the antique geometry model embedded in the disk utilities, I suggest that they abandon that model entirely.  We have hardware lying to software lying to the user. When variable sectors per track showed up we told one small fib, but we've conflated that fib with ever more outrageous lies.  Spindle speed is meaningless for an SSD and has probably been ignored for almost 20 years.  But it's still there.

What we really have are devices which serve up blocks of fixed size with restrictions similar to the alignment restrictions for memory.  That's a pretty simple concept, at least to experienced programmers.

Short term, I'd like to urge that the text-installer be included as an option on the LiveDVD for people who have a clear notion of what they want to do.  Most of my troubles were the result of programs not letting me do reasonable things in a straightforward manner.  The rest was poor guesses at a workaround.

Have Fun!
Reg