21 Apr
2012
21 Apr
'12
10:34 p.m.
On Apr 21, 2012, at 7:08 PM, Mouse wrote:
Except that your reality is a *small* sample set. When you look at data
for 1,000,000's of components you'll see different results. Until you look at
truly large data sets and do the appropriate statistical analysis all you're
really talking about is anecdotal evidence.
How 'bout MTBF?
Extrapolating
long tail effects from single samples does not yield
accurate results.
Neither does assuming that theory matches reality. I would argue that todays computer systems are
*much* more reliable
than the systems that were in general use 10-20 (or more) years ago.
On what grounds?
It contradicts tony's experience. It contradicts mine as well. Do you
have experience pointing the other way? (I would tend to assume not,
since if you had I would expect you to have mentioned it by now, but
assumptions don't make for good data either.)
Unfortunately I can't really point to it since it's proprietary to the companies
that I've work(ed) for. However, there are a number of papers available (I'll
have to dig up citations) on this topic.
If you want to argue that there are grounds for thinking they ought to
be more reliable, that's a relatively defensible stance. But reality
trumps theory, and so far it's looking as though experience is 2-0 in
the direction of your theory - or any theory that predicts higher
reliability out of today's machines - being an inaccurate description
of reality.
What I'm saying is that until you're looking at truly large populations
you can't really derive any significant fact based conclusions. You're
talking about anecdotal evidence and opinion.
TTFN - Guy