Interesting article on the "costs of supporting legacy hardware"

[Christopher Browne]

On Tue, Feb 21, 2012 at 5:05 PM, D. Hugh Redelmeier <hugh-pmF8o41NoarQT0dZR+AlfA at public.gmane.org> wrote:
> | From: Christopher Browne <cbbrowne-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org>
>
> | I think that the thesis presented explains a lot about why mobile
> | platform projects have a hard time staying viable.  You likely can't
> | buy a cell phone from 6 years ago, and it would likely have *so*
> | vastly less capacity than new hardware that few will be terribly
> | interested in working on having modern software accommodate the
> | elderly hardware.
>
> Cell phone hardware is always the minimum to run todays software.
> When you buy one, you cannot upsize it (and DIMMs and hard drives,
> choose a better video card) nor can you do so later.  There is a real
> force pushing towards minimal hardware: trying to keep the power
> budget, the size, and the weight to the minimum, and of course the
> price.
>
> On top of that, the marketing system provides no incentive for
> vendors to provide upgrades.

Laptops have a somewhat similar profile, though there's sometimes
*some* capability to upgrade memory and/or hard drive.  You certainly
have no choice for video save for whatever is soldered into place.

> | And it seems to me that the "costs of supporting legacy hardware" has
> | a fair bit of explanatory power for how difficult it is to support
> | anything other than the very most popular OS.
>
> Legacy is only one dimension of diversity.  There really is a lot of
> other kinds of diversity.  Your SCSI card is a perfect example.
>
> In fact, the opposite of legacy is also a problem: new hardware comes
> out faster than support can flow from the kernel to the distros to the
> releases.  As you point out, Hurd and Plan/9 would probably be
> livelocked if they tried to support new hardware when it comes out.

The presence of VM software is somewhat helpful, except in the sense
that there's something somewhat futile about running two operating
systems simultaneously, the host and the hosted system.

> | Linux (the kernel) has so many developers surrounding it that the
> | community can support a pretty broad set of hardware.
>
> Not broad enough.  Mostly due to trade secrecy.  And not quickly
> enough.
>
> The world that seems to be the "growing tip" of hardware is quite
> closed to us.  Tablets and mobile phones have completely closed
> "peripherals".

You have a point, but in terms of there being a Huge Volume of more or
less ancient hardware supported, well, Linux certainly has that.  The
problems Linux faces are less about having enough developers to cope
with the workload and more about fighting with trade secrecy, which is
a different problem.

> | The BSDs have fewer developers, but still have enough that they can
> | keep up by "cherry picking" from what Linux gets supported.
>
> I didn't think that they are copying code from Linux drivers.  I
> assume that they gain information about hardware from code in Linux
> drivers.

Oh, I wasn't intending to imply copying of code.  I think it has
happened once or twice, but the point is rather about gaining
information via what the Linux driver developers have discovered.  The
BSD guys don't have to dig quite as hard.

> |  The hardware compatibility list
> | (<http://www.plan9.bell-labs.com/wiki/plan9/Supported_PC_hardware/index.html>)
> | is fairly large; the last time I attempted an install I ran afoul of
> | the "oops, it doesn't support my SCSI controller" problem.
>
> The correct solution was to buy a different SCSI controller.  Or give
> up (which you and I did).
>
> I suspect now most plan9 die hards (AKA users) run on virtual
> hardware.  But I don't know.

That seems almost certainly true for Hurd.

> | Entertainingly, the list of supported SCSI controllers hasn't changed
> | much since then.
>
> Isn't SCSI legacy now?  I mean the physical layer.  The only SCSI
> hardware that gets used in my world is old scanners with unique
> capabilities (35mm film scanner; flatbed scanner that handles legal
> size paper).

I find it interesting that Plan 9's hardware list has burgeoned in
almost every area *other* than SCSI.

There were some developments of interest after the last cards that
they list.  I had an Adaptec 2930, I think, and there were a few more
Symbios/NCR generations of chips.  Though with the Adaptec boards,
there were pretty serious problems with the "BIOS of the week"
situation, where the vendor was so keen to spew out new board versions
that it was mighty difficult to keep *any* driver compatible with it.
It certainly is NOT something that would encourage Plan 9 folk with
limited time on their hands to work hard on drivers only to expect to
fall behind anyways.
-- 
When confronted by a difficult problem, solve it by reducing it to the
question, "How would the Lone Ranger handle this?"
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists