on Sun, Dec 14, 2003 at 03:35:06AM -0600, Terry Hancock (hancock@anansispaceworks.com) wrote:
> On Sunday 14 December 2003 02:16 am, Karsten M. Self wrote:
> > on Sat, Dec 13, 2003 at 11:15:07AM -0600, Kent West (westk@acu.edu) wrote:
> > > John Hasler wrote:
> > > >Terry writes:
> > > >>Now, if you are just doing this "in good faith" or making a "best
> > > >>effort", you're asking for a lawsuit the first time somebody buys your
> > > >>stuff and can't get the "backalley joe" Linux distro version 0.0.2 to
> > > >>work with it. Either that, or you're just going to be giving people
> > > >>their money-back an awful lot.
> > > >
> > > >So you guarantee that it works with a specific version of a specific
> > > >distribution. And ship it with that version installed and running.
> > > >
> > > Or to make it even simpler:
> > > "Designed for Knoppix 3.3 11-02-2003"
> > > and include that Knoppix CD without an OS installed. That'd be good
> > > enough guarantee for me.
> >
> > I've recommended this (strongly) in the past.
>
> What does "this" mean, exactly? There's more than one idea here.
The notion of HW vendors using Knoppix (or something very much like it)
as a general-purpose copmatibility testing tool.
I see bidirectional advantages: HW vendors would have better GNU/Linux
support, and Knoppix (and other distros) would gain better HW support
within GNU/Linux.
> > It would be nice to see HW vendors pick up on this.
>
> > It would also be helpful if a HW regression test fram for various
> > peripherals could be created, automated to the extent possible. There's
> > some work toward this (lshw, hdparm, my own system-info script, etc.),
> > anyone care to mention subsystems and possible tests?
>
> You know, I'm not really sure I know what "regression test framework"
> means. I've heard it before, of course, but never really heard a
> definition.
>
> But if you mean that it would be good to make testing much easier for
> the hardware guys by providing automated testing procedures, I think
> that's a great idea.
Essentially, yes.
There are a number of major subsystems:
- BIOS
- Disk, including performance (hdparm).
- Console (keyboard and text-based video)
- Graphics (X), including resolution, refresh, and framerate. I'd
think max res for 75 Hz refresh would be a good metric.
- Mouse/trackpad/pointer.
- Modem.
- Audio.
- Ethernet.
- Peripherals: Serial/Parallel (for legacy), audio out, USB, Firewire, etc.
- Features: sleep/hibernation.
Note too: there's a difference between _testing_ and _passing_. What I
want to know when I assess hardware is "how well is this going to work
for me stock, and how much additional outlay (say, PCMCIA modem for a
crap WinModem) do I need to plan for". 100% is a nice target, but it's
not the essential goal. Many of your comments indicate that it is.
> Regarding supporting "Knoppix 3.3 11-02-2003" or some such exact
> distribution, I think it's missing the point. Not enough people in
> the Linux users community is going to settle for one distribution.
> And in this case, the user is optimistically asking for a distribution
> ONLY ONE MONTH OLD! Think about what you're demanding!
An evaluation against a relatively current Knoppix release, at least as
of HW release, would be valid. I'm thinking 6-12 months, as laptop
lifecycles are very short.
> It's hard enough to stay in the business of selling Debian CDs, when
> every few months your customer base decides your inventory is all
> "stale" and isn't interested anymore. And CDs are *cheap* compared to
> computer systems.
Compatibility reports can be updated readily and in near realtime.
There's this really neat thing called "The World Wide Web" you may have
heard about that's useful for document distribution....
Knoppix, too, is released online and updated every few weeks. It can be
obtained readily.
> How many good systems must you sell to justify the cost of all the
> hardware you had to test and send back because it didn't work? How
> many Linux users will *actually* buy these hypothetical computers,
> instead of just agitating for them? The actual turn-out tends to be
> rather underwhelming, from what little I've been able to see of it.
For a number of vendors (Dell, HP, IBM, Toshiba, Sony) there's a certain
cache or volume of GNU/Linux user purchases.
We're not talking about sending stuff back, we're talking about doing a
compatibility test, and publishing the results. Need not be the
manufacturer either -- LinuxCare used to do something along these lines.
And if the framework is structured properly, a large portion of the
tests can be simply scripted and output as a standard report.
> In order to make a useful size run, the vendor's going to want more
> than 1 month lead-time! You need time to decide on a platform
> design, test the components, replace any that need replacing, and then
> fill bulk orders for enough units to make the sale profitable. I'm
> not sure Linux users as a class are that patient.
I'll be _reasonably_ satisfied with a test that's based on a reasonably
current release of Knoppix. It's the fact of the test, rather than
specific currency, which is most important. Everything else is
incremental improvement.
Peace.
--
Karsten M. Self <kmself@ix.netcom.com> http://kmself.home.netcom.com/
What Part of "Gestalt" don't you understand?
Burn all gifs! Use PNG and tell Unisys to go to hell:
http://burnallgifs.org/
Attachment:
pgp2wQ2oH97RM.pgp
Description: PGP signature