From: Pavel A. on
I have a strange issue with a HP Mydrive "World edition" connected to
ethernet.
Files copied to this drive sometimes get corrupted, in a very specific
pattern.

We cannot reproduce this corruption with smaller files, less than 2 MB, but
with
files > 20 MB it occurs in 80% of copies.
I copy the files by drag & drop from Explorer, or by copy / v command,
results are same. More over, copy /v does not find any error.
Then I run fc /b <copy> <original> and it finds differences.
In a ~ 20 MB file, there are few bytes (1 to 6) with one bit (same in all
cases) stuck in "1".

The drive sits in a conditioned server closet.

What can cause this? Should I look for electrical interference, magnetic
field and so on?

Can you recommend a good disk test for revealing data corruptions?

Regards,
-- pa


From: John Wunderlich on
"Pavel A." <pavel_a(a)12fastmail34.fm> wrote in
news:8716D7B5-71BB-42E9-AF51-75D991227A5A(a)microsoft.com:

> I have a strange issue with a HP Mydrive "World edition" connected
> to ethernet.
> Files copied to this drive sometimes get corrupted, in a very
> specific pattern.
>
> We cannot reproduce this corruption with smaller files, less than
> 2 MB, but with
>files > 20 MB it occurs in 80% of copies.
> I copy the files by drag & drop from Explorer, or by copy / v
> command, results are same. More over, copy /v does not find any
> error. Then I run fc /b <copy> <original> and it finds
> differences. In a ~ 20 MB file, there are few bytes (1 to 6)
> with one bit (same in all cases) stuck in "1".
>
> The drive sits in a conditioned server closet.
>
> What can cause this? Should I look for electrical interference,
> magnetic field and so on?
>
> Can you recommend a good disk test for revealing data corruptions?
>
> Regards,
> -- pa

Ethernet is a serial medium that is well checksumed, so it is unlikely
that the network is causing this problem. My bet would be a loose
data bit connection somewhere between the NIC and the hard drive inside
the NAS. It sounds like you need to replace your NAS.

HTH,
John

From: Paul on
John Wunderlich wrote:
> "Pavel A." <pavel_a(a)12fastmail34.fm> wrote in
> news:8716D7B5-71BB-42E9-AF51-75D991227A5A(a)microsoft.com:
>
>> I have a strange issue with a HP Mydrive "World edition" connected
>> to ethernet.
>> Files copied to this drive sometimes get corrupted, in a very
>> specific pattern.
>>
>> We cannot reproduce this corruption with smaller files, less than
>> 2 MB, but with
>> files > 20 MB it occurs in 80% of copies.
>> I copy the files by drag & drop from Explorer, or by copy / v
>> command, results are same. More over, copy /v does not find any
>> error. Then I run fc /b <copy> <original> and it finds
>> differences. In a ~ 20 MB file, there are few bytes (1 to 6)
>> with one bit (same in all cases) stuck in "1".
>>
>> The drive sits in a conditioned server closet.
>>
>> What can cause this? Should I look for electrical interference,
>> magnetic field and so on?
>>
>> Can you recommend a good disk test for revealing data corruptions?
>>
>> Regards,
>> -- pa
>
> Ethernet is a serial medium that is well checksumed, so it is unlikely
> that the network is causing this problem. My bet would be a loose
> data bit connection somewhere between the NIC and the hard drive inside
> the NAS. It sounds like you need to replace your NAS.
>
> HTH,
> John
>

It could be RAM somewhere in the path to the platters. Say, the cache
RAM on the hard drive inside the NAS itself could be bad. Perhaps a
hard drive diagnostic, and slaving the drive to an existing computer,
would make it possible to test for that. The NAS itself may be a
small computer, with its own RAM, which would give another place
for problems to occur. Robust designs would do self-test at
startup and not start if a problem was detected. Or, to give better
run-time protection, having ECC on RAM would help detect things
such as stuck bits in any RAM being used.

Paul
From: Pavel A. on
Thanks. This kind of confirms my thoughts. The errors occurs on the path
from RAM (cache) to the disk drive, otherwise copy /v would fail earlier.
But OTOH, we consumers like our CE stuff to be dirt cheap... so we get what
we paid for.
--pa


"Paul" <nospam(a)needed.com> wrote in message
news:hj5hut$4qv$1(a)speranza.aioe.org...
> John Wunderlich wrote:
>> "Pavel A." <pavel_a(a)12fastmail34.fm> wrote in
>> news:8716D7B5-71BB-42E9-AF51-75D991227A5A(a)microsoft.com:
>>> I have a strange issue with a HP Mydrive "World edition" connected
>>> to ethernet.
>>> Files copied to this drive sometimes get corrupted, in a very
>>> specific pattern.
>>>
>>> We cannot reproduce this corruption with smaller files, less than
>>> 2 MB, but with
>>> files > 20 MB it occurs in 80% of copies.
>>> I copy the files by drag & drop from Explorer, or by copy / v
>>> command, results are same. More over, copy /v does not find any
>>> error. Then I run fc /b <copy> <original> and it finds
>>> differences. In a ~ 20 MB file, there are few bytes (1 to 6) with one
>>> bit (same in all cases) stuck in "1".
>>>
>>> The drive sits in a conditioned server closet.
>>>
>>> What can cause this? Should I look for electrical interference,
>>> magnetic field and so on?
>>>
>>> Can you recommend a good disk test for revealing data corruptions?
>>>
>>> Regards,
>>> -- pa
>>
>> Ethernet is a serial medium that is well checksumed, so it is unlikely
>> that the network is causing this problem. My bet would be a loose data
>> bit connection somewhere between the NIC and the hard drive inside the
>> NAS. It sounds like you need to replace your NAS.
>>
>> HTH,
>> John
>>
>
> It could be RAM somewhere in the path to the platters. Say, the cache
> RAM on the hard drive inside the NAS itself could be bad. Perhaps a
> hard drive diagnostic, and slaving the drive to an existing computer,
> would make it possible to test for that. The NAS itself may be a
> small computer, with its own RAM, which would give another place
> for problems to occur. Robust designs would do self-test at
> startup and not start if a problem was detected. Or, to give better
> run-time protection, having ECC on RAM would help detect things
> such as stuck bits in any RAM being used.
>
> Paul