I am currently using an old HDD to record video from x4 cctv cameras,
although not continuous .

I used to use a Samsung SSD, but that just suddenly failed after about
18 months / 2 years.

After doing a bit of googling on this subject, I haven't found anything
yet which goes against an ssd.
I have a Samsung EVO 860 250Gb which i could use, as I would rather not
spend money on a HDD if I don't have to.

So what do you suggest would be the best : HDD or SSD.

On 03/03/2024 12:35, RobH wrote:
> I am currently using an old HDD to record video from x4 cctv cameras,
> although not continuous .
>
> I used to use a Samsung SSD, but that just suddenly failed after about
> 18 months / 2 years.
>
> After doing a bit of googling on this subject, I haven't found anything
> yet which goes against an ssd.
> I have a Samsung EVO 860 250Gb which i could use, as I would rather not
> spend money on a HDD if I don't have to.
>
> So what do you suggest would be the best : HDD or SSD.

I did a fair amount of research on this as SSDS are now very available
and not too dear.

Two things appeared to be more or less true.

1. A small percentage of SSDS fail unexpectedly totally and for no
discernible reasons and you are just as fucked as you would be using a
hard drive.
2. Conversely the wear levelling algorithms that spread write operations
across all the NVRAM cells more or less evenly actually result in
*longer* life than a hard drive that is being thrashed.

Certainly all my SSDS (one failed almost immediately) are in fine fettle
and showing no sign of wear at all

And, as with hard drives, use of SMART interrogations reveal when things
are *starting* to go bad.

The one area where they are possibly inferior to spinning rust is data
retention in an unpowered state.

So an archive needs to be powered up from time to time.

Another point is that while a spinning rust disk has a mechanical
lifetime after which bearing wobble and so on limits it no matter how
much its written to, you can achieve extremely long SSD life by having a
very large capacity most of which is *not used*.
The wear levelling will ensure that the writes are spread out over all
the blocks, such that any given block will have many less writes than
the total number.

In your application, I would unhesitatingly go with SSD.

And as big as you can afford.

--
"The most difficult subjects can be explained to the most slow witted
man if he has not formed any idea of them already; but the simplest
thing cannot be made clear to the most intelligent man if he is firmly
persuaded that he knows already, without a shadow of doubt, what is laid
before him."

- Leo Tolstoy

On 03/03/2024 13:44, The Natural Philosopher wrote:

> The one area where they are possibly inferior to spinning rust is data
> retention in an unpowered state.

I just connected up an SSD that I haven't used for 3 years, and it just
'worked'. I read a few files, and all seemed good.

On 03/03/2024 13:44, The Natural Philosopher wrote:
> On 03/03/2024 12:35, RobH wrote:
>> I am currently using an old HDD to record video from x4 cctv cameras,
>> although not continuous .
>>
>> I used to use a Samsung SSD, but that just suddenly failed after about
>> 18 months / 2 years.
>>
>> After doing a bit of googling on this subject, I haven't found
>> anything yet which goes against an ssd.
>> I have a Samsung EVO 860 250Gb which i could use, as I would rather
>> not spend money on a HDD if I don't have to.
>>
>> So what do you suggest would be the best : HDD or SSD.
>
> I did a fair amount of research on this as SSDS are now very available
> and not too dear.
>
> Two things appeared to be more or less true.
>
> 1. A small percentage of SSDS fail unexpectedly totally and for no
> discernible reasons and you are just as fucked as you would be using a
> hard drive.
> 2. Conversely the wear levelling algorithms that spread write operations
> across all the NVRAM cells more or less evenly actually result in
> *longer* life than a hard drive that is being thrashed.
>
> Certainly all my SSDS (one failed almost immediately) are in fine fettle
> and showing no sign of wear at all
>
> And, as with hard drives, use of SMART interrogations reveal when things
> are *starting* to go bad.
>
> The one area where they are possibly inferior to spinning rust is data
> retention in an unpowered state.
>
> So an archive needs to be powered up from time to time.
>
> Another point is that while a spinning rust disk has a mechanical
> lifetime after which bearing wobble and so on limits it no matter how
> much its written to, you can achieve extremely long SSD life by having a
> very large capacity most of which is *not used*.
> The wear levelling will ensure that the writes are spread out over all
> the blocks, such that any given block will have many less writes than
> the total number.
>
> In your application, I would unhesitatingly go with SSD.
>
> And as big as you can afford.
>

Thanks for that, and as I mentioned, I have a 250GB EVO ssd .
In the machine I plan to use the ssd, there is a 120Gb spinning rust
disk , of which there is almost 90gb free space.
So with a 250 GB disk there would be about 220 Gb free space, which is
not used by anything

On 03/03/2024 16:35, GB wrote:
> On 03/03/2024 13:44, The Natural Philosopher wrote:
>
>> The one area where they are possibly inferior to spinning rust is data
>> retention in an unpowered state.
>
> I just connected up an SSD that I haven't used for 3 years, and it just
> 'worked'. I read a few files, and all seemed good.
>
That is useful info

Most people are saying 'be worried after ten years'
>
>

--
“Ideas are inherently conservative. They yield not to the attack of
other ideas but to the massive onslaught of circumstance"

- John K Galbraith

On 03/03/2024 17:36, RobH wrote:
> On 03/03/2024 13:44, The Natural Philosopher wrote:
>> On 03/03/2024 12:35, RobH wrote:
>>> I am currently using an old HDD to record video from x4 cctv cameras,
>>> although not continuous .
>>>
>>> I used to use a Samsung SSD, but that just suddenly failed after
>>> about 18 months / 2 years.
>>>
>>> After doing a bit of googling on this subject, I haven't found
>>> anything yet which goes against an ssd.
>>> I have a Samsung EVO 860 250Gb which i could use, as I would rather
>>> not spend money on a HDD if I don't have to.
>>>
>>> So what do you suggest would be the best : HDD or SSD.
>>
>> I did a fair amount of research on this as SSDS are now very available
>> and not too dear.
>>
>> Two things appeared to be more or less true.
>>
>> 1. A small percentage of SSDS fail unexpectedly totally and for no
>> discernible reasons and you are just as fucked as you would be using a
>> hard drive.
>> 2. Conversely the wear levelling algorithms that spread write
>> operations across all the NVRAM cells more or less evenly actually
>> result in *longer* life than a hard drive that is being thrashed.
>>
>> Certainly all my SSDS (one failed almost immediately) are in fine
>> fettle and showing no sign of wear at all
>>
>> And, as with hard drives, use of SMART interrogations reveal when
>> things are *starting* to go bad.
>>
>> The one area where they are possibly inferior to spinning rust is data
>> retention in an unpowered state.
>>
>> So an archive needs to be powered up from time to time.
>>
>> Another point is that while a spinning rust disk has a mechanical
>> lifetime after which bearing wobble and so on limits it no matter how
>> much its written to, you can achieve extremely long SSD life by having
>> a very large capacity most of which is *not used*.
>> The wear levelling will ensure that the writes are spread out over all
>> the blocks, such that any given block will have many less writes than
>> the total number.
>>
>> In your application, I would unhesitatingly go with SSD.
>>
>> And as big as you can afford.
>>
>
> Thanks for that, and as I mentioned, I have a 250GB EVO ssd .
> In the machine I plan to use the ssd, there is a 120Gb spinning rust
> disk , of which there is almost 90gb free space.
> So with a 250 GB disk there would be about 220 Gb free space, which is
> not used by anything

Sounds like a plan, then.

--
“Ideas are inherently conservative. They yield not to the attack of
other ideas but to the massive onslaught of circumstance"

- John K Galbraith

On 03/03/2024 17:36, RobH wrote:
> On 03/03/2024 13:44, The Natural Philosopher wrote:
>> On 03/03/2024 12:35, RobH wrote:
>>> I am currently using an old HDD to record video from x4 cctv cameras,
>>> although not continuous .
>>>
>>> I used to use a Samsung SSD, but that just suddenly failed after
>>> about 18 months / 2 years.
>>>
>>> After doing a bit of googling on this subject, I haven't found
>>> anything yet which goes against an ssd.
>>> I have a Samsung EVO 860 250Gb which i could use, as I would rather
>>> not spend money on a HDD if I don't have to.
>>>
>>> So what do you suggest would be the best : HDD or SSD.
>>
>> I did a fair amount of research on this as SSDS are now very available
>> and not too dear.
>>
>> Two things appeared to be more or less true.
>>
>> 1. A small percentage of SSDS fail unexpectedly totally and for no
>> discernible reasons and you are just as fucked as you would be using a
>> hard drive.
>> 2. Conversely the wear levelling algorithms that spread write
>> operations across all the NVRAM cells more or less evenly actually
>> result in *longer* life than a hard drive that is being thrashed.
>>
>> Certainly all my SSDS (one failed almost immediately) are in fine
>> fettle and showing no sign of wear at all
>>
>> And, as with hard drives, use of SMART interrogations reveal when
>> things are *starting* to go bad.
>>
>> The one area where they are possibly inferior to spinning rust is data
>> retention in an unpowered state.
>>
>> So an archive needs to be powered up from time to time.
>>
>> Another point is that while a spinning rust disk has a mechanical
>> lifetime after which bearing wobble and so on limits it no matter how
>> much its written to, you can achieve extremely long SSD life by having
>> a very large capacity most of which is *not used*.
>> The wear levelling will ensure that the writes are spread out over all
>> the blocks, such that any given block will have many less writes than
>> the total number.
>>
>> In your application, I would unhesitatingly go with SSD.
>>
>> And as big as you can afford.
>>
>
> Thanks for that, and as I mentioned, I have a 250GB EVO ssd .
> In the machine I plan to use the ssd, there is a 120Gb spinning rust
> disk , of which there is almost 90gb free space.
> So with a 250 GB disk there would be about 220 Gb free space, which is
> not used by anything

that 250 GB HDD o SDD can't hold that much footage off 4 CCTV cameras?

I have 8 cameras and 20 TB of storage and that gives me 83 days.....

So my 8 cams are producing almost 250 GB of new data a day so 4 cams
would give you just 2 days before overwriting?

Hope you check your CCTV every day while you're next on a week's
holiday! :-)

On 03/03/2024 18:46, The Natural Philosopher wrote:

>>> The one area where they are possibly inferior to spinning rust is
>>> data retention in an unpowered state.
>>
>> I just connected up an SSD that I haven't used for 3 years, and it
>> just 'worked'. I read a few files, and all seemed good.
>>
> That is useful info
>
> Most  people are saying 'be worried after ten years'

Ah, I see. So, SSDs are unsuitable for very long term archival storage,
say. Is the issue that the charge simply leaks away over time? So, maybe
the SSD will still work, but the data won't be readable?

On 04/03/2024 18:03, GB wrote:
> On 03/03/2024 18:46, The Natural Philosopher wrote:
>
>>>> The one area where they are possibly inferior to spinning rust is
>>>> data retention in an unpowered state.
>>>
>>> I just connected up an SSD that I haven't used for 3 years, and it
>>> just 'worked'. I read a few files, and all seemed good.
>>>
>> That is useful info
>>
>> Most  people are saying 'be worried after ten years'
>
> Ah, I see. So, SSDs are unsuitable for very long term archival storage,
> say. Is the issue that the charge simply leaks away over time? So, maybe
> the SSD will still work, but the data won't be readable?
>
>
Yes, but remember as soon as it gets power, my *understanding* is that
it will 'refresh' and be good for another ten years. I might be wrong on
that, though
>
>
>
>

--
There’s a mighty big difference between good, sound reasons and reasons
that sound good.

Burton Hillis (William Vaughn, American columnist)

The Natural Philosopher wrote:

>> Ah, I see. So, SSDs are unsuitable for very long term archival
>> storage, say. Is the issue that the charge simply leaks away over
>> time? So, maybe the SSD will still work, but the data won't be readable?
>
> Yes, but remember as soon as it gets power, my understanding is that it
> will 'refresh' and be good for another ten years. I might be wrong on
> that, though

I wondered about that, if any bits are lost, your filesystem is not
trustworthy, maybe you could reformat it and bring it back to life, but
what if the drive itself stores internal metadata or firmware inside its
own flash? It could be bricked ...

On 04/03/2024 19:24, Andy Burns wrote:
> The Natural Philosopher wrote:
>
>>> Ah, I see. So, SSDs are unsuitable for very long term archival
>>> storage, say. Is the issue that the charge simply leaks away over
>>> time? So, maybe the SSD will still work, but the data won't be readable?
>>
>> Yes, but remember as soon as it gets power, my understanding is that
>> it will 'refresh' and be good for another ten years. I might be wrong
>> on that, though
>
> I wondered about that, if any bits are lost, your filesystem is not
> trustworthy, maybe you could reformat it and bring it back to life, but
> what if the drive itself stores internal metadata or firmware inside its
> own flash?  It could be bricked ...
>
I suspect its not organised like that. But I don't have time to find out
right now.
But most filesystems will recover from the odd lost bit.,

>

--
Labour - a bunch of rich people convincing poor people to vote for rich
people by telling poor people that "other" rich people are the reason
they are poor.

Peter Thompson

On 3/3/2024 1:46 PM, The Natural Philosopher wrote:
> On 03/03/2024 16:35, GB wrote:
>> On 03/03/2024 13:44, The Natural Philosopher wrote:
>>
>>> The one area where they are possibly inferior to spinning rust is data retention in an unpowered state.
>>
>> I just connected up an SSD that I haven't used for 3 years, and it just 'worked'. I read a few files, and all seemed good.
>>
> That is useful info
>
> Most  people are saying 'be worried after ten years'

I think that's a reasonable assessment.

One person at least, took a 16 year old NAND flash-based device,
and could not read it. Which means some critical data table
likely failed (like the mapper table).

SSDs are NOT engineered as time capsules. The gate-drain
behavior at ten years, is the basic physics of the thing.

TLC cells can become mushy in as little as three months,
and if you see the read speed drop, that means an ARM core
is doing error correction on every sector being read out.

You can see this, even on "brand new" TLC drives. If you
HDTune bench the thing, before doing a "freshen, write from
end to end" procedure, the read-rate seen could be 1/4 of the
speed rate on the tin. Don't panic when that happens, and
use the drive a bit or write it, to improve the benchmark.

I've never heard any "squeaking" about gate-leak behavior
having been defeated or extended in any way. It is possible
to anneal a NAND flash cell, and "make it last forever"
in terms of write life. That would take my current 4TB drive
from 2400TBW, to infinity. However, we don't know how to implement
annealing at an atomic level, which is why wear life is not
going to get extended that way. But the "leaky-gate" problem,
that's more or less a constant of the thing. That's why they're
not expected to be data capsules. We even get hints at this,
with NOR flash in motherboard BIOS, where a byte or two will
corrupt in the chip and prevent POST. That could happen
past the ten year mark.

An Intel employee "hint" mentioned here, indicates that
Optane has even-higher wear life than NAND (duh), but the
time capsule behavior is no good, and he allows that a NAND
based device might be better. The reason we would check and
look up Optane, is because the storage mechanism doesn't
use a floating gate. At $3000 per drive, these are not items
we will be burying in the ground anyway.

https://techreport.com/forums/viewtopic.php?t=121371

The upper limit on "convenient" storage, is likely MDISC
and the estimated lifespan of carefully preserved polycarbonate
for the discs. The chemical doing the recording might have
an extremely good life, but the polycarbonate survival
needs to be taken into account. So when they say "a thousand years",
they mean "if the polycarbonate actually lasts that long".

*******

There are rotating drives, intended for video recording.
And they will have a cache behavior suited to a fixed pattern
of access. That's a "purple" drive, versus the "red" drive
used for a NAS. Both drives stay spinning and don't park.
The NAS assumes there is no pattern to the application,
while the "purple" drive has probably been studied in
recorder equipment, to handle the pattern. Maybe it means
the file system can't have journaling, to simplify where
the writes go.

A four camera setup, should be easy for a "Purple" drive
to handle, even if you put the recording on the wrong
file system. Then you have to decide how big of a drive
to buy. An 8TB purple would be Helium, a 6TB purple
might be an air breather. What's the room air quality like ?
If high humidity, I might go with the Helium. A Helium drive
is "guaranteed" <cough> to hold the gas for five years. Some
Helium drives actually have a pressure sensor, but the details
of the SMART entry seem to have only been obtained by
observation and reverse engineering. And if you drilled a hole
to "let the gas out", well, now the drive is dead, because
the flying behavior is fouled up (wrong flying height).

If the Helium drives had a "gas refill port", I might
give them the benefit of the doubt.

I have one 500GB drive here, which has lasted over 50,000 hours,
and little sign of degradation. And brothers of that drive,
which are showing issues at 5000 to 10000 hours. There's no
predicting which drive will be a champ.

Within the last year, I purchased three 1TB WD Black drives
("Made in china"), and one of them would not spin when plugged
in new. That's my *first* infant mortality on a hard drive here!
So rather than "shite in the year 2000", that's shite from 2023.
As a data point. That's not supposed to happen, as the
drive has to be commissioned before it can be shipped, and the
motor has to work. The boxes the drives ship in, are of
good quality. We can't blame dropping the box, since the
drives have a 300G rating, and the box materials are soft
enough, no box-drop can achieve 300G.

Paul

On 3/4/2024 2:37 PM, The Natural Philosopher wrote:
> On 04/03/2024 19:24, Andy Burns wrote:
>> The Natural Philosopher wrote:
>>
>>>> Ah, I see. So, SSDs are unsuitable for very long term archival storage, say. Is the issue that the charge simply leaks away over time? So, maybe the SSD will still work, but the data won't be readable?
>>>
>>> Yes, but remember as soon as it gets power, my understanding is that it will 'refresh' and be good for another ten years. I might be wrong on that, though
>>
>> I wondered about that, if any bits are lost, your filesystem is not trustworthy, maybe you could reformat it and bring it back to life, but what if the drive itself stores internal metadata or firmware inside its own flash?  It could be bricked ...
>>
> I suspect its not organised like that. But I don't have time to find out right now.
> But most filesystems will recover from the odd lost bit.,

SSD drives have critical tables, just as HDD do.

The external LBA presented to the drive, is the virtual address.
The mapper table, gives a physical address.

Sector 0 is stored at 0x12345678 and
Sector 1 is stored at 0x3579ABCD.

For wear leveling, "fresh" blocks come from the free pool, and
the drive "fragments" as time passes. Thus, after a period
of time, the mapper table almost looks random. You might say
to me, why isn't the table like this ?

Sector 0 is stored at 0x00000000 and
Sector 1 is stored at 0x00000001.

It might be at first, but if any sort of non-uniform or
random write behavior comes along, that helps randomize the mapping.

Losing the mapping table is deadly for storage, and is
one of the reasons scoping a NAND flash chip and "trying to
suck the data out of it", is going nowhere fast. You need the
mapping table, to even think about harvesting data from raw chips.

A hard drive has mapping tables too, but they're used
for reallocations. At one time, an IBM document claimed
1MB of their 8MB cache RAM chip, was used for a remapping table,
for fast access to remap info.

Some hard drives used to die (predictably) roughly
30 days after purchase. This was due to a firmware
bug in a critical data table. Which might be prevented
with a firmware flash before that happens. It turns out
that SSDs have not been the only devices to suffer
at the hands of "funky" firmware.

The reason SSDs recover from mushy sectors, is the size
of the error corrector syndrome, compared to the size
of the data. A 512 byte sector now has 51 bytes error syndrome.
That's a 10% overhead. Maybe there's a Reed Solomon being
used there. Whatever the algo is, the corrector is implemented
in firmware, and there seems to be no interest in a wire-speed
dedicated error corrector logic block. This is how we can
tell "something is wrong", when it reads at 530MB/sec new and
reads at 300MB/sec three months from now. Turn off the power
for three months, do an HDTune bench, and see. Modern drives
are TLC or QLC, and the "mush" behavior was first observed
on TLC.

Paul

The Natural Philosopher <tnp@invalid.invalid> wrote:
> On 04/03/2024 18:03, GB wrote:
> > On 03/03/2024 18:46, The Natural Philosopher wrote:
> >
> >>>> The one area where they are possibly inferior to spinning rust is
> >>>> data retention in an unpowered state.
> >>>
> >>> I just connected up an SSD that I haven't used for 3 years, and it
> >>> just 'worked'. I read a few files, and all seemed good.
> >>>
> >> That is useful info
> >>
> >> Most  people are saying 'be worried after ten years'
> >
> > Ah, I see. So, SSDs are unsuitable for very long term archival storage,
> > say. Is the issue that the charge simply leaks away over time? So, maybe
> > the SSD will still work, but the data won't be readable?
> >
> >
> Yes, but remember as soon as it gets power, my *understanding* is that
> it will 'refresh' and be good for another ten years. I might be wrong on
> that, though

I suspect it may need go through and rewrite all the data. ie it's not just
a case of powering it up for 5 seconds then powering it down again, it needs
to be left powered. Then it's a question of whether the flash cells will
refresh themselves, whether the controller will tell them to do it as part
of regular management tasks it does in the background, or whether they need
external hints to do that. Maybe you need to do a full drive read so that
it fetches every block and then rewrites it in the process? Or maybe you
need to explicitly ask it to rewrite the block?

I'd want to understand this better before relying on it.

Theo

On 06/03/2024 14:46, Theo wrote:
> The Natural Philosopher <tnp@invalid.invalid> wrote:
>> On 04/03/2024 18:03, GB wrote:
>>> On 03/03/2024 18:46, The Natural Philosopher wrote:
>>>
>>>>>> The one area where they are possibly inferior to spinning rust is
>>>>>> data retention in an unpowered state.
>>>>>
>>>>> I just connected up an SSD that I haven't used for 3 years, and it
>>>>> just 'worked'. I read a few files, and all seemed good.
>>>>>
>>>> That is useful info
>>>>
>>>> Most  people are saying 'be worried after ten years'
>>>
>>> Ah, I see. So, SSDs are unsuitable for very long term archival storage,
>>> say. Is the issue that the charge simply leaks away over time? So, maybe
>>> the SSD will still work, but the data won't be readable?
>>>
>>>
>> Yes, but remember as soon as it gets power, my *understanding* is that
>> it will 'refresh' and be good for another ten years. I might be wrong on
>> that, though
>
> I suspect it may need go through and rewrite all the data. ie it's not just
> a case of powering it up for 5 seconds then powering it down again, it needs
> to be left powered. Then it's a question of whether the flash cells will
> refresh themselves, whether the controller will tell them to do it as part
> of regular management tasks it does in the background, or whether they need
> external hints to do that. Maybe you need to do a full drive read so that
> it fetches every block and then rewrites it in the process? Or maybe you
> need to explicitly ask it to rewrite the block?
>
> I'd want to understand this better before relying on it.
>
> Theo

I said it was only my *understanding*, not guaranteed fact!

There may well be a way that flash RAM re-charges itself without a write
cycle
Remember it's the erase cycle that is on a block basis and therefore slow.

It may be that a background process rewrites data to the cells that have
'1's' ...

--
“Those who can make you believe absurdities, can make you commit
atrocities.”

― Voltaire, Questions sur les Miracles à M. Claparede, Professeur de
Théologie à Genève, par un Proposant: Ou Extrait de Diverses Lettres de
M. de Voltaire