Slow write speeds on a WD 500GB drive

[Yousuf Khan]

http://www.google.com.au/search?q=wd+very+slow+writes
produces some very interesting hits.
Looks like WD might well have a problem.

Interesting! I tried a bunch of different google search keywords myself,
but couldn't find anything like this. I suppose it helps if other people
are trying the searches too so there is multiple sources perspectives.

Yousuf Khan

 

[Rod Speed]

Rod Speed wrote

Well, I found some of the old threads here (Google Groups):

I do recall seeing some of these old threads with passing interest, but
at that time it did not interest me as I wasn't aware that they affected
me too. I can now see that they may have been the very same issues
I was dealing with myself.
Yes.

HD Sentinel actually implemented a resident program
just to keep the hard disks from being put to rest, so it
must be a well-known problem.

It isnt actually. Its only the WDs that get that carried away.

 

[Rod Speed]

Rod Speed wrote

Interesting! I tried a bunch of different google search keywords myself,
but couldn't find anything like this. I suppose it helps if other people
are trying the searches too so there is multiple sources perspectives.

Yeah, its still a problem with searches that
don't have any particularly unique word to use.

I quite often find that I cant find something that
I know for sure is in groups.google somewhere.

 

[Yousuf Khan]

But you don't see too many relatively simple devices like
that suffer two completely different very serious faults.

Not so unusual when you run the device for 24/7 for months on end. When
it was being run as a PVR drive, it was never shut-off, even if the PVR
itself was off. The drive had to always be ready to record something, as
soon as the PVR woke up, so the drive itself could never sleep.

You can get delayed results like that, particularly when the power
supply seriously over voltages the system for a short time.

I don't disagree that both types of damage might have been done during
the time it was in that enclosure, but I don't agree that there was ever
any single event that was responsible for both failures. The damage to
the internal drive was likely caused by the excessive load/unload cycles
over time while it was inside the enclosure, which is basically a
mechanical failure. What caused the enclosure's interface electronics to
die was something else entirely, details not fully determined yet, but
this one could only be electronic/electrical in nature.

Harder to explain the weird symptom that the write speed
obviously varys with where the head is on the platter that
way tho and that only affecting writes, not reads.

If we ever do work out what the fault is due to, bet we
will kick ourselves for not thinking of that earlier tho.

I don't think so. I don't think a common explanation is needed for both
failures, necessarily. I'll explain why below.

BTW, I had written about this drive's history before on this newsgroup
(Google Groups):
http://is.gd/tkqeKJ

Back then, the drive was still inside its My Book enclosure, and I was
having some kind of problem with reading SMART data out of it through
its eSATA interface. It was during this thread that I was first
introduced to HD Sentinel.

The My Book contained a bridging chipset that acted as the interface
from the hard disk to the computer, offering USB, Firewire, and eSATA.
What we didn't know back then was that the chipset was manufactured by a
company called Oxford Semiconductor. This chipset contained its own ARM
processor. So it was an intelligent interface controller. If the ARM
chip failed somehow, then it would mean the interface is dead, but not
necessarily the disk inside.

http://en.wikipedia.org/wiki/Western_Digital_My_Book#Internals

Cant see how that would produce that very unusual symptom
of the write speed so dramatically affected by the location of
the head on the platter, and not the read speed.

The HD Sentinel tech believes that the excessive load/unload cycles
might have damaged the head parking area of that drive. This might have
resulted in microscopic dust particles covering the platter which made
it harder to re-magnetize the platter near the parking area. Since the
parking area is near the front of the platters, this is the area that
was first affected. He suggested that this dust is just going to keep
creeping towards the back of the drive, until the entire surface is
unusable. The fact that there are some areas which are alternately fast
and slow (i.e. the "Slow", "Medium" and "Fast" sections of the drive) is
an indication that this backward creep is already happening.

It's a theory, don't know if it's the only possible theory.

They certainly are notorious for unloading much too frequently.

Havent see anyone report that weird write speed result with them tho.

Now that I know that this Western Digital load/unload issue has come up
before on this newsgroup, I would like to know how other people
determined that this was a problem? I think Arno was one of the first to
express concern over this load/unload issue, but I don't quite know what
led him to believe that it was a problem? I realize we're getting into
historical issues here.

Yousuf Khan

 

[Rod Speed]

Rod Speed wrote

Not so unusual when you run the device for 24/7 for months on end.

Very unusual, actually. I do that with almost everything, at home and at
work.

When it was being run as a PVR drive, it was never shut-off,

My PVR isnt either except for a hardware upgrade.

even if the PVR itself was off.

I don't ever turn mine off except for a hardware upgrade.

The drive had to always be ready to record something, as soon as the PVR
woke up, so the drive itself could never sleep.

I choose to not let it sleep even tho the PVR is
quite capable of starting it up when it needs it.

I don't disagree that both types of damage might have been done during the
time it was in that enclosure, but I don't agree that there was ever any
single event that was responsible for both failures.

You don't know that with the power supply spiking it.

The damage to the internal drive was likely caused by the excessive
load/unload cycles over time

Don't believe that. There is no reason that should
produce that effect on JUST writes and not on reads.

while it was inside the enclosure,

You don't know that it wouldn't have done that when
an internal drive too.

which is basically a mechanical failure.

Have fun explaining how it can produce that very
unusual symptom you are seeing, extremely slow
writes ONLY when the head is in part of the platter.

What caused the enclosure's interface electronics to die was something
else entirely,

You don't know that either.

details not fully determined yet, but this one could only be
electronic/electrical in nature.

You don't know that either. A drop can see a failure there
just with a bad joint or a connector coming off internally etc.

I don't think so.

I've never seen one that wasn't.

I have seen some where no one bothered to
work out why the symptom was seen tho.

I don't think a common explanation is needed for both failures,
necessarily. I'll explain why below.

Didn't help.

BTW, I had written about this drive's history before on this newsgroup
(Google Groups):
http://is.gd/tkqeKJ

Back then, the drive was still inside its My Book enclosure, and I was
having some kind of problem with reading SMART data out of it through its
eSATA interface.

That isnt necessarily related to either fault.

It was during this thread that I was first introduced to HD Sentinel.

The My Book contained a bridging chipset that acted as the interface from
the hard disk to the computer, offering USB, Firewire, and eSATA. What we
didn't know back then was that the chipset was manufactured by a company
called Oxford Semiconductor. This chipset contained its own ARM processor.
So it was an intelligent interface controller. If the ARM chip failed
somehow, then it would mean the interface is dead, but not necessarily the
disk inside.

Is there any evidence that the ARM chip has actually failed ?

Even if it had, you don't know that that wasn't because the
power supply had spiked it and that it took some time for
the effect of the spike to show up with the drive itself.

http://en.wikipedia.org/wiki/Western_Digital_My_Book#Internals

The HD Sentinel tech believes that the excessive load/unload cycles might
have damaged the head parking area of that drive.

Very unlikely indeed. And doesn't explain why writes
are fine with the heads over half the platter either.

This might have resulted in microscopic dust particles covering the
platter

Yes. But that would see head crashes, not extremely
poor write perfomance over just half of the platter,
and have no effect on reads at all.

which made it harder to re-magnetize the platter near the parking area.

That 'explanation' is completely off with the ****ing fairys.

There is no way that that would affect JUST half the platter,
and that boundary never changing, or the very sharp boundary
been extremely slow writes and normal write performance and
have no effect whatever on the read speed.

Since the parking area is near the front of the platters, this is the area
that was first affected.

Its never going to affect just half the platter with such
a striking boundary between the good half and the
bad half and not affect the read performance at all.

He suggested that this dust is just going to keep creeping towards the
back of the drive, until the entire surface is unusable.

Have fun explaining why no head crashes
have been seen and just ONE bad sector.

Bet the boundary that's so striking on your
first chart doesn't move over time either.

The fact that there are some areas which are alternately fast and slow
(i.e. the "Slow", "Medium" and "Fast" sections of the drive)

That isnt actually the case. Your first chart does in fact show
that the last half of the drive is fine. The steps you do see after
that are just the areas where the sectors per track changes.

You'll find that with a brand now drive.

is an indication that this backward creep is already happening.

Fraid not.

It's a theory,

Which has holes in it you can drive an full aircraft carrier battle group
thru.

don't know if it's the only possible theory.

It's a complete dud in fact.

Now that I know that this Western Digital load/unload issue has come up
before on this newsgroup, I would like to know how other people determined
that this was a problem?

Basically by looking at the load/unload value in the SMART data.

I think Arno was one of the first to express concern over this load/unload
issue,
Yes.

but I don't quite know what led him to believe that it was a problem?

Basically by looking at the load/unload value in the SMART data.

I realize we're getting into historical issues here.

But the detail is in groups.google

 

[Yousuf Khan]

You don't know that either. A drop can see a failure there
just with a bad joint or a connector coming off internally etc.

I can confirm that it was neither a physical drop, nor an electrical
spike. Firstly, the drive was sitting in a home theatre cabinet most of
that time, so there was no physical movement done while it was powered
on, so a drop is not possible; and no earthquakes either, in case you're
going to ask that too. Secondly, all of the devices were on an expensive
surge protector.

That isnt necessarily related to either fault.

That's just the background to the thread! You don't have to argue about
everything!

Is there any evidence that the ARM chip has actually failed ?

Irrelevant, the chipset failed, the embedded ARM was part of the
chipset, so it failed along with everything else.

Even if it had, you don't know that that wasn't because the
power supply had spiked it and that it took some time for
the effect of the spike to show up with the drive itself.

See above, surge protected.

Its never going to affect just half the platter with such
a striking boundary between the good half and the
bad half and not affect the read performance at all.

Yeah, well, I'm not convinced that's the answer either, but I don't
dismiss it.

Have fun explaining why no head crashes
have been seen and just ONE bad sector.

Bet the boundary that's so striking on your
first chart doesn't move over time either.


That isnt actually the case. Your first chart does in fact show
that the last half of the drive is fine. The steps you do see after
that are just the areas where the sectors per track changes.

The "Medium" section was alternately fast and slow. The "Slow" section
was universally slow, while the "Fast" section was universally fast.

Basically by looking at the load/unload value in the SMART data.


Basically by looking at the load/unload value in the SMART data.

But there's a lot of huge numbers that can occur in SMART data that you
can get freaked out about, such as ECC Recovered, Seek Error Rate, Total
LBA written and Total LBA read. Nobody actually pays attention to those
numbers until there is an actual problem occurring.

So Arno much've noticed a problem?

 

[Rod Speed]

Rod Speed wrote

I can confirm that it was neither a physical drop, nor an electrical
spike.

No you cant.

Firstly, the drive was sitting in a home theatre cabinet most of that
time,

Most of the time is irrelevant and you don't even know
if it got dropped before it was even put in there etc.

so there was no physical movement done while it was powered on,

Doesn't need to have been powered on with a dry joint or connector loosened.

so a drop is not possible;

That's just plain wrong.

and no earthquakes either, in case you're going to ask that too. Secondly,
all of the devices were on an expensive surge protector.

Doesn't eliminate the possibility of the power supply with
an intermittent fault that spikes the output rail occasionally.

That's just the background to the thread! You don't have to argue about
everything!

You hate it when the holes are pointed out in your theorys.

Irrelevant, the chipset failed, the embedded ARM was part of the chipset,
so it failed along with everything else.

You don't know that either.

See above, surge protected.

Doesn't mean anything, see above.

Yeah, well, I'm not convinced that's the answer either,

but I don't dismiss it.

I do on that alone, let alone the lack of any head crashes.

Even just smoke from cigarette smokers produces head
crashes with modern hard drives, there is no way that any
physical wear in the landing zone could produce anything
that's not going to produce a head crash if its floating around.

He's completely off with the ****ing fairys on that claim.

The "Medium" section was alternately fast and slow.

No evidence of that with that first chart of yours.

The "Slow" section was universally slow,

Yes, the first chart does show that.

while the "Fast" section was universally fast.

No evidence of that either with that first
chart of yours that for the entire drive.

But there's a lot of huge numbers that can occur in SMART data that you
can get freaked out about, such as ECC Recovered, Seek Error Rate, Total
LBA written and Total LBA read.

Not with someone like Arno who knows what the numbers mean.

Nobody actually pays attention to those numbers until there is an actual
problem occurring.

That's just plain wrong with people like Arno, Frank and me.

So Arno much've noticed a problem?

Fraid not. He just noticed the very high load/unload count
before it had even got anywhere near the design limit for
the drive and stopped it doing that obscene level of unloads.

Its all there in groups.google.