choice of external USB drive for mount at boot

You might not get anywhere even if you manage to open it up. Some of the external drives (such as Samsung) have the USB interface on the drive board, so you can't just plug in another standard SATA drive in to it. The USB drive enclosures have separate boards to do USB to SATA.

---druck

This site has stats for the drives used in a datacenter (not comparing 2.5 to 3.5 but perhaps of some interest):

writing

Facinating reading, thanks for the link.

And such a wide range of annual failure rates, I'm surprised they stick the the Seagate Barracuda 7200.11 or .14 at all with a long term and consistent failure rate of 25% ish. They only have few hundred of each though. Another that stands out are the couple of WD Red Series at 6 ish %, isn't the Red series supposed to be the "datacentre" class of WD drive?

It confirms my experience with HGST relative to WD and Seagate, both at home and at work.

"We actually deployed over 4,000 drives, as we replaced 324 failed drives as well as 631 Seagate 3TB suspect drives." ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ You must have missed that sentence. [Snip]

I found it confusing. Ultimately all drives fail and its a bit strange to not graph failure rates with time and so on a bit more clearly.

MTBF is what you want, not how many percent of drives failed

Contrary to what most people think, MTBF is actually a way to express the percentage of drives that fail within some time interval. It is NOT an indication of how long a typical drive will last.

When a drive has an MTBF of 20 years, that DOES NOT mean it can likely be used in a system for 20 years. What it actually means is that when you have many drives and you monitor them for say 1 year, 1/20 of them will probably fail.

Yees, but....

... in fact depending on the failure modes the distribution of failures over time is seldom even.

Given 20 new drives after the first year you might have one fail from a manufacturing fault, then 4 years of no failures, then 3 in the next year and none in the following 10 years...and so on..with most of the failures concentrated towards the end when failure through actual wear takes over.

In a wear out scenario you get a typical 'early (warranty) failure from faults' then a period of little or no failure, followed by a rapidly rising failure rate towards life end.

So although MTBF id not 'how long the drive lasts' necessarily, in practical terms it's a very good guide.

To the layman, maybe. But it is not the way MTBF is defined.

MTBF is only defined for the useful life of the product. The increasing failure rate at the end of useful life is not included in it.

And again, MTBF value has no relation to the length of useful life of the product. It is really possible for a drive that has a useful life of 4 or 5 years to have an MTBF of 50 years. That only means that the percentage of drives failing prematurely is small. It does not tell you what the situation after 10 years, let alone after 50 years is.

But yes, I know, many hobbyists read it that way and expect a really long lifetime when they read the MTBF spec of 500.000 hours or more.

I know this as the 'bathtub' failure pattern of electonic components, the name coming from the shape of the curve.

AFAIK it is more typical of electronics than of mechanical ones, so its really reflecting the failure pattern of a drive's electronics rather than the mechanical bits.

Well assume we mean MTTF instead then

-- New Socialism consists essentially in being seen to have your heart in the right place whilst your head is in the clouds and your hand is in someone else's pocket.

No. its very true of mechanical systems too.

E.g. a car often has warranty faults in the first year, then becomes reliable for a long time before wear takes out one part after another.

Fair point: I was thinking of more purely mechanical stuff: no electrical bits more complex than you'd find in an electric drill, pull-start outboard motor or model aircraft engine fitted with a gloplug, where mechanical wear is almost the only source of failure.

My point is that you don't understand, and apparently still did not get, that the property is expressed in units of TIME but it bears no relation to an expected LIFETIME, it is really a calculated value that expresses "probability of failure".

So again, MTBF, or MTTF, does NOT tell you how long it will take before an average item fails. Of course, people think it does. That is why it is so popular in specs.

But designers do know what to do with this value.

same applies really.

a few DOAs or shortly after, then wear related issues towards life's end

Except that in general they DO tell you how long things will last before they fail *because* the probability of failures *is* by and large scattered near to MTTF if its expressive of linear degradation.

Unfortunately usenet posters don't seem to :-(

You keep expressing your gut feeling, but again, it is in no way related to what MTBF means.

I think this is not true. The probability of a drive failing is a function of time and is typically not constant over the lifetime of the product.

I think this may be key in what you are sayiing. How then do they calculate the MTBF?

You are free what to think, I have no problem with that.

Normally like this:

1. define a lifetime for the product (e.g. 5 years)
2. run a large batch of product for that time (e.g. 1000 pieces)
3. observe the number of failures

calculate the MTBF as (time * number of units) / number of failures.

Of course, in reality the manufacturer does not have the patience to run the test for the entire lifetime of the product, so they try to predict what will happen over the full lifetime by taking a shorter sample and operate the unit at extended stress conditions (e.g. higher temperature), and compensate the calculated MTBF according to experience with earlier batches. E.g. they know that when they sample for one year instead of 5 years, they will have 1/10 of the failures. This is the increasing failure rate over time.

However, and I cannot stress this enough, the fact that the failure rate increases dramatically late in the life of the product is NOT factored into this. So you CANNOT use the MTBF as a prediction of the lifetime of the product.