[Sussex] Sky boxen / Embeded LINUX and NT

[Geoff Teale]

Evening all,

Steve Dobson wrote:
====================
<snippage>

>>This is a different question.  I was addressing the question:
>>"How can we be sure that LINUX in embedded space isn't a lame 
>>duck just like Windows?"  This is moving into the question of:
>>"Now that `small' embedded computers are as powerful as the
>>desktop systems of a few years ago.  How do we ensure that
>>Linux wins over Windows and other commercial OSs?"
>
You're right!  I did loose the thread in the discussion someplace.

I think the answer is that LINUX will naturally adjust to the roles it 
is used in.  More features (such as low-latency kernels) will get added 
in.  My only concern  is that all of this will make the kernel more 
complex and harder to maintain.

Slightly off topic, but, from a maintenance point of view microkernels 
(the HURD excepted) are generally easier to maintain.  An analogy would 
be the maintenance of a unstructured program (i.e one where everything 
is executed in one long script) as opposed to maintenance of a well 
structured object orientated program (incidentally the company I 
currently work for tend to write programs using the first unstructured 
approach - just one of several reasons why I'm looking for a new job 
after just 6 months).

>
>I agree that "cost" is the biggest driver but your reasoning
>is skewed.  Two years ago I would have agreed with you whole
>heartedly, but experience is a good teacher.
>
>I was working on a small (postcard sized) PC to be added into
>a expensive device (~ $150,000 and up).  I had a quick solution
>that if we when to the 64Meg "Disk on Chip" from the 32Meg 
>version (a cost different of ~ $100).  The manager said that
>the device's price was determined not by the company but by 
>the market.  Therefore the cost of my time to do the extra
>work to squeeze into the smaller device was worth well over my
>year's salary.
>
>The more I have though about it the more I agree.  All products
>have price bands, VCRs, cars, laptops, washing machines.  There
>is very little choice between the price bands.  The price bands
>denote feature jumps and brand names.  If you can squeeze the
>features of the band above you have a winner.
>
>With productions runs in the embedded market being measured in
>the 100,000s a $15 saving per unit is big bucks.  What is 
>£20,000 to these savings?  Therefore if the QNZ's runtime 
>licence is less that WinCE it is more likely to be selected.
>The cost of the development licence and engineers are more
>or less insignificant over the lifetime of the product run.
>Remember most are produced (with minor tweaks) for years.  It's
>only computers that are measured in weeks (days & hours can't
>be that far away).
>
>My above arguments are for the mass produced embedded markets
>only.  For small production runs (Military, nuclear power
>stations, ...) then these argument losses big time to your.
>
>The counter argument to yours in that area is that one 
>technology is "perceived" as less risky, and therefore
>worth the extra cost.  However the technology fight has to
>done with managers that "just don't get it".  I know, it
>was one of my reasons for leaving my first company (of
>11 years)
>

A very well reasoned arguement.   I'm certainly not going to disagree 
with any of this.

- Geoff

-- 
* GJT  / tealeg
* geoff at tealeg.uklinux.net