Re: speed measured (Was: Re: AGFA SD card)
> Michael J. Mahon wrote:
>>>Michael J. Mahon wrote:
>>>>I think the problem is that flash memory is so trouble-free in operation...
>>>>I've never found any *bad* flash memory,
>>>>and I've never heard from anyone who wore some out!
>>>Yes ! And the housing boom will last forever ! The prices will never
>>In this case, more like, "The card will last longer than the camera!"[/color]
> No. It is more like: "In the past 5 years, my house grew significantly
> in price. Houses of everyone whom I know grew up in prices. Therefore,
> everyone's house in the US grew up in price. The growth will be
> "My memory cards never failed. Cards of everyone around me whom I know
> never failed. Therefore, nobody in the world had their memory cards
> failed. In fact, the cards will never fail".[/color]
Gee, and here I thought that we had reached a point of some
Your analogy is false. My statements were not analogies, but facts.
I have already experience my media outlasting my cameras several times.
I have personally used them heavily for the last five years and
have not experienced a single problem with data reliability. This
is a reasonable base of experience to justify lowering my level of
concern about them as a point of failure. I am much more concerned
about physically losing one containing valuable pictures than I am
about the card losing some data, and for both reasons, I never have
data *only* on a card for more than a week. BTW, I have had data
stored on these cards for years (redundantly) and have not observed
any loss of data over this period, either.
Unlike a potential real estate bubble, there is no mechanism which
could lead to a catastrophic loss of much of the flash memory data
in the world, the country, or even in my house--except physical
destruction! (Which is why I believe in off-site archives.)
In general systems design, there are four possible combinations
of system inputs and outputs:
1) Input signal is not present and is not detected.
2) Input signal is not present and *is* detected.
3) Input signal is present and is detected.
4) Input signal is present and is *not* detected.
While states 1) and 3) represent accurate behavior, all real
systems are subject to errors, represented by states 2) (misdetection)
and 3) (failure to detect). The negative consequences of error 2) can
be as crippling as those of error 4).
You accuse me of error 4--not detecting a problem that may exist
in reality. But you are subject to error 2--detecting a problem
that may not exist.
Notice that we now have several years worth of experience with
consumer-grade flash memory devices of various designs, and there
is no outcry about failures or design problems. They work for the
purposes for which they are intended. (Archival storage may not be
one of those purposes!)
I never said that they will never fail--of course, they will. But
will anyone care if they are using them to: take pictures before
transferring them to other media, store MP3s to listen to them
portably, or even (as I do) use them to simulate a hard drive for
an older, slower computer (with regular backups). If one type
lasts 8 years and another lasts 16 years, and I will have replaced
them all in less than 5 years, should I care?
I view them as convenient, reliable (in practice), and, ultimately,
disposable media that sells for less than the price of a good meal
at a restaurant.
Unless you are responsible for designing the next generation of
flash memory, you would probably be well served by regarding them
Now, if you were expressing worry about the practical life expectancy
of today's so-called "archival" media, like recordable DVDs, I'm with
BTW, have you checked your tires recently? How about your brakes?
And how do you feel about riding in elevators? ;-)
Using cumulative failure data to estimate future risk is an inductive
process, and therefore cannot be certain. But is it the best way we
have of making rational decisions, and it does not rule out hedging.
Have a nice day.
Home page: [url]http://members.aol.com/MJMahon/[/url]
"The wastebasket is our most important design
tool--and it is seriously underused."
Re: speed measured (Was: Re: AGFA SD card)
Michael J. Mahon wrote:
> Notice that we now have several years worth of experience with
> consumer-grade flash memory devices of various designs, and there
> is no outcry about failures or design problems. They work for the
> purposes for which they are intended.[/color]
I have both my first degree and doctorate in semiconductor physics (I
do not practise this occupation, though). I fabricated semiconductor
microdevices by myself. Some of them failed later. The reasons were
that e.g. I etched the semiconductor material too thin, or deposited a
too thin layer of metal, or I inadvertently misaligned the mask when
depositing contact layers. Or the wire contacts were not welded too
well to the device. The device worked well in the beginning, but failed
later, because too much current destroyed the paths which turned out to
be too thin, or electromigration of metal into the semiconductor
So that I know that the chip which I am holding in my hands may well
turn out to be a time bomb. All is going well, until one day the unit
"tires". I am a perfectionist by nature, and I want to minimize to zero
the amount of mishaps on me. Some manufacturers of semiconductor chips
adhere better to the QA processes, and their chip designs are
inherently better than those of other manufacturers (e.g. provide
thicker paths, better alignment of masks, use of materials less prone
to disintegration, incorporating less amount of elements which may fail
etc). For example, the MLC chip has much more complicated architecture
(=more elements), therefore it is more probable to fail than an SLC
chip. Some manufacturers have their metal contacts made more sloppy
(e.g. Taiwan-made Toshiba vs Japan-made Toshiba or China-made SanDisk).
The probability of failure of contacts in the chip of _that_ particular
manufacturer is higher.