Chen Xiao Fan wrote:
> It is reported in C/Net (news.com) that People in some wealthy
> countries have thrown out Windows PC (keep the LCD) plagued
> with virus/spyware and buy a new PC instead since it is cheaper
> to get a new low end PC (US$300-500) than to get someone
> to fix it. I am not so sure whether it is true or not.

It's certainly true here in New York.  Two factors come to mind: (1) the 
going rate for professional computer repair work seems to be >= 
$60/hour, and (2) the average person is still pretty uncertain what the 
difference is between hardware and software, much less what exactly a 
virus does.  So a lot of people think if their computer is "infected," 
that's that.  They don't really understand that if you reinstall the OS 
it will be just the same as it was new.

> Maybe it is better to donate them to some charity and install
> Linux on it.

I would say so.  On the other hand, even the charities here want to set 
up machines to run Windows, because that's what most their recipients 
know if they know anything.

To get back on topic, my take on this is that it simply costs less to 
diagnose and fix or replace things that are observed to break, at a high 
level, than it does to do rigorous testing to find low-level problems in 
all constituent parts.

Here's a little numeric model off the top of my head:

- fraction of boards that have a fault = f

- cost of the bare board = b

- cost of testing a board before assembly = t

- cost of everything else that goes into a unit, assembled to the point 
where it can be functionally tested = e

This assumes that we're always going to functionally test units once 
they're assembled.

If your strategy is S1 = "pre-test all boards, and use only boards that 
work" then you will incur cost (b + t + e) on all good boards, but just 
(b + t) on bad boards.  So average unit cost is:

	f(b + t) + (1 - f)(b + t + e)

If your strategy is S2 = "functionally test all units after assembly, 
and throw away the ones that fail," you will incur cost (b + e) on good 
boards, and also (b + e) on bad boards.  So average unit cost is:

	b + e

In order for strategy S1 to be cheaper than S2, we must have

	f(b + t) + (1 - f)(b + t + e) < b + e

Simplifying:

	fb + ft + b + t + e - fb - ft - fe < b + e

	b + t + e - fe < b + e

	t - fe < 0

	t < fe	

So let's say f, the board failure rate, is 1%.  Then pre-testing makes 
sense if, for instance, your parts and assembly together cost > $100 and 
pre-testing adds < $1 to your board cost.  This seems fairly unusual 
given what I'm hearing about testing costs.

And note that if you decide that repair is cheaper than discarding 
defective units, or if you allow for failure in components other than 
the board, pre-testing gets even less attractive.

I better get back to work.
-- 
Timothy J. Weber
http://www.lightlink.com/tjweber
-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist