On 8/9/2018 8:56 PM, Paul wrote:
mike wrote:
On 8/9/2018 4:28 AM, Paul wrote:
Bill wrote:
mike wrote:

How's the reliability?
I'm still reading that they fail catastrophically without warning.

I've heard that the reliability of SSDs far exceeds that of the
mechanical hard drives (for, in fact, an obvious reason--no moving
parts). The "trim" software for my Intel SSD even provides an
indication of the drive's reliability (I'm not sure how well that
works). I do regular backups too.

You're the perfect customer for an SSD.

You're mixing up reliability and wear life.

I beg to differ. When the drive QUITS WORKING, that's end of life.
Doesn't matter what caused it.

We have to be careful to separate the component parts though.

No, WE don't. WE...meaning consumers...care about RESULTS.

The stuff you mention below is relevant to manufacturers, not consumers.
When I buy a car designed to run on city streets and I drive that car on
city streets, I care about one thing. Did I get to my destination?

If my alternator quits, I see a red light. I may decide to reduce power
consumption as much as I can. I will likely get to my destination.
I will almost always get to someone who can help. It will get fixed
and life goes on.

If my car just decides, to STOP DEAD and require that I buy a new one,
simply because the alternator exceeded some life prediction, sorry, you
have to abandon your cargo too,
I'm gonna be ****ed!

I'm not bitching about wearout. I'm bitching about a vendor decision
that might make the collateral damage MUCH worse than it needs to be.

If it's simply a matter of data retention ability of some number of cells,
the device should function 'till the last one dies, then give me easy
ways to recover what isn't dead.

In the case of a SSD, it's not at all clear who uses what strategy
to what degree. "Trust me, we got good stuff...secret stuff...but
it's good stuff."


Some aspects of the electronics you use, can never be
all that good. They could never last "forever". The
best power converter they could make at work, was
around 10 million hours MTBF. The guys who did that,
were pretty product of their tiny gadget.

MTBF as a number is easily misunderstood. It's been almost 30 years
since I did a calculation, and I don't want to start now,
but...as I recall, some relatively large number of devices
are expected to fail within the MTBF. IIRC, it's like
30% or so?

I figgered someone would challenge, so I looked it up. I got it
backwards:

Once the MTBF of a system is known, the probability that any one
particular system will be operational at time equal to the MTBF can be
estimated.[1] Under the assumption of a constant failure rate, any one
particular system will survive to its calculated MTBF with a probability
of 36.8% (i.e., it will fail before with a probability of 63.2%).[1] The
same applies to the MTTF of a system working within this time period.[5]

The bathtub shaped curve, assumes failures are random,
and are a function of the "quality" or the "architecture"
of the item.

Having an actual wearout behavior, that is not
a random phenomenon. That's predictable.

Say I start writing my brand new M.2 motherboard SSD at
2.5GB/sec. And I do that all day long, day after day.
These devices don't have enough wear life, to make it
all the way to the end of the warranty period while
doing that. If you do that to the drive, grind it into
the ground, all the retailer has to do, is check the
wear life and point out to the customer "you wore it out,
it didn't fail as such". You would be denied warranty
relief at retail level or at factory level.

The warranty covers the "random stuff". The stuff
that went into the MTBF calculation.

I disagree.
Warranty is a sales tool.
Sometimes, it's mandated by law.
Sometimes it IMPLIES that your product is better
than the competition.
Sometimes, the consequences of failure far outweigh
the cost of the device. A tire that falls apart may
cost the vendor $50 to remedy, but the lawsuits
surrounding the 50-car pileup it caused can be
astronomical.
Vendors make tradeoffs between cost of doing it
right vs cost of failure. It's all part of the balance
sheet...the cost of doing business.

A few vendors just fix what broke.
Most find a way to blame it on you so they don't have
to fix it. The length of the warranty is largely
irrelevant.

That's just the way the world works.

Another problem with MTBF is the lack of useful data.
Back in the day, we could get numbers for individual devices.
But what about the vendor six levels deep in the manufacturing
process who mis-calibrated the reflow oven by a few degrees
and built in a failure mode for the 100 solder connections?
There's too much statistical data from too many places.

Don't get me started on the fact that devices don't exist
in a vacuum. The probability of breaking your phone by sitting
on it is far higher than the probability of a manufacturing defect
caused crack. It's not a warranty issue, but it still don't work
no more. Length of the warranty or MTBF number is irrelevant.

The fast pace of new technology is also a big issue. By the time
you could determine the failure rate by testing, the product has
been replaced by a newer version several times. Accelerated
life testing is good, but still seriously flawed for new
technologies that we barely understand.


So...MTBF calculation == good.
Counting on that to save your ass is inadequate.
Warranty length is mostly independent of that number.

Everyone knows
that like toilet paper, the number of write cycles
is limited. If a customer burns them up, they won't
be providing a shiny new drive for the customer
to burn up a third, fourth, or fifth time.

If it's determined a customer abused a product,
then there's no warranty. Maybe your credit card
provides more extensive product insurance, but
the normal retail relationship mainly covers
product defects. And wear is not a defect.
It's an expected parameter a customer
can control, by not doing too many writes
per day...

If I buy a gallon of paint at Home Depot,
apply it to the walls, when the can is empty
I can't run to the store and go "Defect! Defect!
This can is empty. I blame the hole on top."
The warranty might cover a failure of the
chemical composition (maybe the drying
accelerator agent is missing and the paint
never dries for you). But simply using up
the quantity of materials in the can, doesn't
entitle you to a "infinite refill". I can't
paint the Brooklyn Bridge with one can of
paint.

Paul

Are we having fun yet?