Sigh, is there a reason to bring this up again, particularly when you
don't really seem to be adding anything? Comments in-line:
I have seen various comments like "we leave our servers running 24/7
because powering up a hard drive causes more wear than leaving it
running." I think it's mostly laziness and apathy about conserving
energy.
It has nothing to do with conserving power *or* laziness. Servers run
24/7 because they are *used* 24/7. I am not going to tell users that
their data is unavailable because the "servers are conserving power
right now". It *is* possible to have the drives spin down when not in
use, but a busy server is rarely going to do that due to the amt of
actual use *and* background tasks ( a lot of servers do significant
maintenance tasks like backup and optimization when idle).
The TiVo forums discuss it a lot because a standard TiVo HD
runs all the time, buffering 30 minutes of whatever channel it's left
on.
It shows up on these forums because people freak out that Tivo's normal
mode of operation (always running) will shorten the life of the HDD.
They usually freak out after a hard drive fails (which is a bit late!).
The truth is that one way or the other it doesn't make *that* much
difference either way if a hard drive runs all the time or not. Hard
drives fail, and we just have to deal with it until a reasonable
alternative is available (like solid state memory) and affordable.
Light bulbs fail all the time, and people don't freak about those! (well
most don't anyway).
Do IT people who leave servers running 24/7 ever have much choice of
NOT leaving them on 24/7? If not, how can they make scientific
comparisons of drive-bearing life? As long as the head isn't moving,
bearing life seems to be the main concern. On a home PC left on all day
it's far less likely that the drive will be doing anything but spinning
at high RPM for no real reason.
Why is it up to the IT people to do the scientific comparisons, isn't
that up to the manufacturer? As an IT person who *also* used to design
bearings (ball, journal *and* roller) there *are* both wear and startup
issues. Most catastrophic failures occur at startup unless there is an
unrelated cause of failure (like an overheat condition, overloading or
oil starvation).
I've heard similar claims that the "shock" of turning on a light bulb
is worse than leaving it on all the time. Usually those comments came
as a way to excuse energy consumption after a debate on the merits of
waste. In reality, bulbs have a finite hours rating and will burn out
faster the longer they are left on, as long as they aren't flipped on
and off as torture. CFL bulbs (w/ballast) don't like to be switched on
and off quickly, but I can't imagine them burning out faster if you
only cycle on/off once in 10 minutes or so.
Comparing light bulb mechanics to bearing mechanics is a completely
inaccurate comparison. Light bulbs relay on electrical resistance to
operate, there are no moving parts, lubrication, centripetal force
issues, etc. About the only things the two share in common are heat
issues (for vastly different reasons). I often compare hard drive
failure (which I will not necessarily agree is mostly caused by bearing
failure, but that is the most predictable cause) to light bulb failure
only because the failure probability curve versus time is similar. This
is not because the failure mechanism is the same, but because both oftem
fail due to very minor variances in tolerances and manufacture, probably
variances so small they wouldn't be measurable. The end result is that
a large element of failure seems to be random.
Would anyone claim that car wheel bearings get as much wear when you
pull out of the driveway vs. a 500 mile nonstop trip? In that case, the
"spin up" would be when you first move the car after sitting.
I have no doubt that a car wheel bearing *does* receive more wear in the
time it takes to "spin" up then in the same amount of time spinning
normally. But this is a low stress application, and the amount of
additional wear would be extremely small. When designing bearings for
cars and most land (or sea) based applications, factors of safety are
pretty high, particularly in a low heat application such as a car wheel
bearing. In highly weight-sensitive applications (like jet-engines)
those factors of safety are much slimmer and start-up wear is a much
bigger issue.
What
exactly causes the "big shock" when a hard drive spins up? The heat
generated from constant spinning would seem to far outweigh it.
Careful, you're showing your inexperience. If you are really asking the
question, don't draw incorrect conclusions from assuming an answer.
Startup wear is large for four reasons:
- Lubricant is not distributed evenly. If it's a pressurized
lubrication system (like in a car), the lubricant system usually isn't
at pressure yet (since it's usually powered by the engine itself). If
it's a passive system (like a wheel bearing or HDD that's simply packed
with grease) then the lubricant has likely "settled" to the bottom and
has to be "spun" up and distributed evenly by centripetal force.
- Lubricant is not yet at operating temperature. Most lubricants
(including most passively used greases and car oil) is designed to be
used at engine operating temperature. At startup it is cold and
therefore does not work as well as it is designed to, and it may be too
viscous to even flow to all the places that need to be lubricated.
- Lubricant is not at pressure. Lubrication in bearings works by
forming a "pressure wedge" between two moving parts. The pressure is
formed by the centripetal force between the bearing element (or inner
race in the case of a journal bearing) against the outer race of the
bearing. Normally the lubricant would just be pushed out of the way,
and this is what happens in a non- or slow moving part, but at speed it
cannot flow fast enough, so it becomes pressurized (in a "wedge shape"
because it is trying to flow "away" from the bearing element) and
seperates the two parts (because liquids only compress minimally). Non-
or slow moving parts will not be able to form an effective pressure wedge.
- Part tolerences are designed at operating temperature, not at resting
temperature. Different metals and materials expand at different rates
due to temperature, so their fits will differ depending on the thermal
enviroments. Tolerences are typically designed to operate optimally at
the prevalant operating temperature, and since lubricant films are
highly dependent on those tolerences for getting up to pressure,
variations in fit can affect them greatly.
For all of the above reasons, bearings (or at least a portion of them)
will be in actual unlubricated contact at startup, and a lot of wear can
occur. Once the bearing is spun up, the bearings will only be in
contact with lubricant.
Why
does Windows have a "Turn off hard disks" feature in Power options if
not to reduce bearing wear?
OMG! How about to reduce power? Also, noise is a factor.
If anyone has thorough technical articles on hard drive wear, please
post. Specifically, what is so torturous about spinning up the drive,
and how can that brief cycle be quantified, damage-wise against
constant spinning with higher heat levels?
How about just someone with a technical knowledge of how bearings work?
I believe I've demonstrated how bearing startup is an extreme event in
bearing life, and bearings in drives are no different than any others
(most use ball bearings I believe). Many drives are now using
hydrostatic bearings, which will eliminate *some* of the low pressure
problem at startup depending on how the pressure is supplied.
How can it be quantified? That's not that simple since bearing failure
is sudden and not a progressive process. Bearings fail when the
lubricant film fails. When does the film fail? It can happen at
various times for various reasons, including lubricant starvation, loss
of viscosity (or even boiling!) due to overheating, manufacturing
defects causing protrusions that project *through* or disrupt the film,
overloading, or shock (which is just a form of overload). But startup
is the only time in a bearings life that it is *designed* to be
unprotected by a lubricant film. Balanced against that, it's also
usually the lowest load point.
Heat issues are a red herring as long as they are at design levels.
Overheat is a different problem, but they shouldn't be operated that way
anyway, either intermittantly or long-term.
But this is mostly an academic exercise. HDD bearings are not under
high load or low factors of safety. They can handle high numbers of
start/stop events. But neither does a running drive use a lot of power
(a good argument can be made that they use a lot in *aggregate* so
should be minimized), so manufacturers typically use a drive in whatever
a manner works best for them. Tivo doesn't run their drives
continuously to minimize start/stop cycles, they run them that way
because they're always *using* them. I would never tell someone not to
shut their computer off to save their hard drive (unless it was already
failing, of course), it just doesn't make enough difference. If it was
significant, we would likely already know the answer as to which one was
better.
Randy S.
