RPR said:
Just to clarify what Arno wrote:
MTBF and average life time are unrelated.
Failure rates of disk drives (or any mechanical devices) follow
a "bathtub" pattern. The initially high failure rate (DOA, infant
mortality) drops quickly (within hours/days) to a baseline, which
stays more or less constant (generally for years under proper
usage and conditions), then starts to rise slowly due to wear.
Now the MTBF is the inverse of the failure rate in steady state,
i.e. at the baseline of the bathtub curve. It is only the height of the
bathtub bottom, not the length of the bathtub that determines MTBF.
It isnt even that. MTBFs are purely calculated, not measured
and bear no real relationship to actual average life at all.
The length is the lifetime, which is rarely specified (because
manufacturers would get into too much hot water if they did.)
It isnt even possible to specify it. The best that can be done
is to see what the average life turns out to be, and thats quite
difficult to quantify, and that data is useless by then anyway
because you cant buy the drive new anymore anyway.
Generally the warranty period is a
good indication of the expected lifetime
Like hell it is. 3 years was close to universal with mass market
commodity drives. Then most manufacturers decided that the
inevitable cost of the extra 2 years of warranty prevented them
from being as aggressive on the price of the drive, so most of
them changed to 1 year instead for that reason.
Now we have seen Seagate decide to offer 5 years in the hope that
that would increase their sales volume enough to come out ahead.
since manufacturers like to brag about their warranty but don't
want to spend much on warranty replacements of course.
Its much more complicated than that.
You can have a very high MTBF, i.e. a very reliable drive, but it may
wear out within a year (deep but short bathtub.) Conversely, you can
have a population that dies slowly but constantly over 20 years (long
but shallow bathtub - low MTBF but high average life.)
And you can get some duds like the IBM 75GXPs and the
60GXPs and the Fujitsu MPGs which died like flys and the
MTBF had absolutely no predictive value on that what so ever.
By definition, when the devices start to wear out, you're out of
the steady state and those failures don't count against the MTBF.
You clearly dont have a clue about MTBF at all.
MTBF is usually monitored by ORT (Ongoing Reliability Test)
processes that take a sample of new drives (after burn in to
get rid of infant mortality) and run them for a couple of weeks
or months, so reliability is only proven for that long.
MTBFs arent even measured at all like that.
Experience, component data and field
data are used to extrapolate from there on.
Wrong again.
If some new component is introduced (which is the case for
practically any new product), there may be surprises if the
component doesn't meet what the supplier promised, or if
the designers specified a component that's not quite right
for the intended function, or if there are interactions that nobody foresaw.
And some drives like the IBM 75GXPs and the 60GXPs can turn
out to have a fundamental design problem that doesnt show up
for quite a while in the field and the MTBF is completely irrelevant.
Doesnt actually happen that often with hard drives.
Just remember that wear generally rises
with the exponential of the temperature.
Thats utterly mangled as well.
Switching a drive on generates a lot
of heat from friction and inrush current.
And that is why the power cycles are
specified separately in the datasheets.
I concur with Rod. The only drives that died on me over the last
couple of years were Notebook drives. No surprise there (bleeding
edge + power cycles + abuse). But I still have a couple 2 and 4 G
SCSI drives running in my Linux boxes. They must be mid-90's vintage.
I've got some IDEs of that size in dinosaurs that dont warrant
anything bigger, tho I did bin the 2G because its irritatingly slow.
The 4G is a bit marginal in that area, but since it only affects the
boot time and I dont normally boot, just return from hibernation,
I havent bothered to replace it with something faster.
