[time-nuts] the logic versus cpu equation
walter shawlee 2
walter2 at sphere.bc.ca
Sat Jan 16 13:37:24 EST 2016
I always marvel at the argument that a $1 CPU can out-perform a bag of logic
which will be "more expensive". this is such utter hogwash, I really felt I had
to comment.
this view assigns zero cost to the software, development (IDE), programming, and
support of the code, a cost that dwarfs the hardware cost by many orders of
magnitude. furthermore, testing and validation of the code is not trivial, and
it is quite common for uncorrectable timing errors to surface due to device
archtecture.
further, this design cannot be readily duplicated, making sharing of the effort
very awkward. the world is littered with abandoned software designs which fall
into disuse for this very reason. not to mention 18 month obsolescence cycles
for microprocessor parts. it aways sounds so simple to "just email the code".
but what happens 5 years later?
in addition, the integrity of the design is fundamentally uncertain, as ALL
programmed parts fail, sometimes within a short window. data retention is often
brief (5 years, almost never over 20 years, and easily upset by static
discharge, read cycles, elevated temperatures, radiation and other factors)
there is *no such a thing as durable code*, it is always a transient design
element. those who believe otherwise have simply failed to read the giant stack
of device datasheets that clearly spell this out. as NAND flash users in server
farms are quickly learning, drives based on this technology have much higher
failure rates than mechanical magnetic storage, and the extensive published data
is quite compelling.
a design based on hard wired logic always works as intended, and is easily
maintained in the event of device failure. this is simply not true with any
microprocessor, gate array or pal based solution. never forget that the
programmed part IS NOT an off the shelf item. the amount of high tech gear
currently being trashed because of data corruption is staggering.
this problem has surfaced in so many areas of technology from cell phones, test
equipment to environmental controls, it simply makes my blood run cold to hear
the fiction of "software designs work so much better than hardware" repeated
without at least a word from the opposing view.
I have worked all my adult life in the design and support of advanced systems
and I can assure you that nothing is more troublesome in the field than a
software based design, which invariably also cost far more than the supposedly
expensive hardware it replaced. I feel confident the actual cost to make the
1pps software based divider, for example will be far in excess of the cost of a
few 74HC390's.
I have no argument with computer based anything, it is merely critical to
understand the staggering volatility of that platform, the true scope of costs,
and the consequences of failure. I have watched many companies suffer under the
burden of software development gone off the rails, and know how much in
retrospect they wished for a different path. and yes, this was written on a
computer, I own 8 of them,
so please look for the real message here, don't go for the sarcastic response.
all the best,
walter
--
Walter Shawlee 2, President
Sphere Research Corporation
3394 Sunnyside Rd., West Kelowna, BC
V1Z 2V4 CANADA Phone: (250) 769-1834
walter2 at sphere.bc.ca
WS2: We're all in one boat, no matter how it looks to you.
Love is all you need. (John Lennon)
But, that doesn't mean other things don't come in handy. (WS2)
More information about the time-nuts
mailing list