Will Dyson on 8 Oct 2006 01:12:26 -0000


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: [PLUG] RAID cards

On 10/5/06, gabriel rosenkoetter <gr@eclipsed.net> wrote: 
On Mon, Oct 02, 2006 at 12:54:59PM -0400, Will Dyson wrote:
> A while back, I remember seeing a proposal for a standardized on-disk
> format for hardware and software raid. I had thought that the major
> vendors were on-board with the idea, but I haven't heard anything
> since (from card vendors or the Linux MD/DM maintainers).

Care to reference a url on that? 

Hi Gabe,

The standard is called DDF. Sorry for being too lazy to look it up before.

http://www.snia.org/standards/home

http://www.intel.com/technology/magazine/standards/RAID-spec-1105.htm

I am inclined to find any such proposition laughable at best. The
idea that you could do this and have the format be bootable on even
two of the hardware architectures where the card was useable is a
support nightmare, if even technically feasible.


Given that a few minutes of searching have not located me any cards
that even claim to support DDF, others seem to share your opinion on
the need for such a thing.

But I don't see what the ability of one card to work in multiple types
of host systems has to do with it. DDF is a data format for describing
the striping layout of a raid group. If a card's firmware has bugs
that cause it to only work on one host architecture, then that
firmware has problems. I don't see how using a card-specific raid
group description format would prevent that situation.

Granted, my 3ware 9000-series RAID controllers are in an i386 machine
running FreeBSD with drivers and management interface provided by
the (card's) vendor, but if I were willing to go off in the rough
and write my own drivers and management interfaces to their (published)
API, I would fully expect that they didn't do anything assinine
with bit order (or anything else) that would make them non-functional
(or even just non-bootable) on a system with a different hardware
architecture than they'd anticipated.

I find it highly dubious that anybody would get all vendors
producing computers with PCI busses to sign on to such a
standardization unless it were codified as part of PCI, where it
really, really doesn't belong. Never mind that even if the choices
made at the time seemed like a good idea for an on-disk format,
locking into that prevents people from devising more efficient
methods.


I am confused by your references to the PCI spec.

In a full-hardware-raid card, the on-disk description of array groups
a matter for the card's firmware alone. If such a card were to support
DDF, then I would expect its management interface to work just like it
does with the vendor's current proprietary format.

If the DDF specification failed to specify the byte sex (or whatever)
of the on-disk data, then that would be a grave flaw in the standard.
But it would not be an indictment of the general concept. If the
standard was not flexible enough to deal with future striping layouts,
then I suppose a new revision of the standard would be needed (along
with a new firmware to understand the revised standard).

Are we even talking about the same thing?

--
Will Dyson
___________________________________________________________________________
Philadelphia Linux Users Group         --        http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion  --   http://lists.phillylinux.org/mailman/listinfo/plug