7200RPM Hard Disks - Are they still Power Hogs?

[Pew]

No, enterprise stuff are pretty boring with just model numbers and typically white labels with specs like write once read many, write many read once, and are sold typically through OEM channels instead of consumer facing retailers. You are correct that Red is more of a consumer or prosumer grade stuff. This doesn't mean they are not good but they just don't trade off as extreme as enterprise stuff who would check their stats like hawk and complain if they don't meet spec, or accept a whole series or disqualify them if the overall spec is not met (like percentage of failure every month above 45C).

That makes me said, I literally just tossed out a pair of HPE 4TB SAS SSDs; I should have checked their lifespan first.

 

[PandaBear]

SAS SSDs can be very durable, fast for the right kind of work, and very power hungry. It also requires the right card to run as well as SAS isn't just regular SATA.

Also the "writes" in enterprise SSD may not be a real write but rather just write to RAM until you 1) power down or 2) were sent a flush command to flush the data. This helps reduce the amount of PE cycles at the cost of a super capacitor holding 20ms of electricity to flush anytime. The entire process is very inefficient for home user and you may end up constantly using up to 25W to keep it running idle, then a fan to cool it.

 

[meep]

Surveillance really beats on hard drives. I’d use a low cost triple-layer or even quad layer nand drive (ssd) since it’s not carrying valuable data.

Spinning drives have their place. Western digital 5400 rpm green drives have been excellent in my use, with well over 5 years of home use … one started throwing smart warnings around 8 years old, running 24x7. It’s not full-time powered any more but still works. These are nice in a RAID arrangement, but slow on their own. RAID naturally increases their summarized throughput. You might consider using those on the backend and using a small SSD as front side cache?

SSDs in the name brands have become very reliable, I like Samsung and Hynix drives, zero failures and I’ve had about a dozen in my care. The prices on them could be absurd right now.

 

[PandaBear]

Surveillance really beats on hard drives. I’d use a low cost triple-layer or even quad layer nand drive (ssd) since it’s not carrying valuable data.

Typical program and erase cycles of different types of nand:

SLC: around 1M cycles
MLC (2LC): around 15K cycles
TLC (3LC): around 1500-5000 cycles
QLC (4LC): at most 500 cycles.

Since video recording is continuous write, I would never use QLC. If you really want to there are "industrial" SSD that is more expensive and use either MLC or SLC.

Still though, if you are not on battery a mechanical HDD like WD Purple and Seagate Skyhawk are the better choice. I have a few that ran for at least 7 years already and still alive. There are custom HDD commands for surveillance system that these HDD take to align data write so they can minimize seek, and make their shingled recording work reliably.

 

[meep]

Typical program and erase cycles of different types of nand:

SLC: around 1M cycles
MLC (2LC): around 15K cycles
TLC (3LC): around 1500-5000 cycles
QLC (4LC): at most 500 cycles.

Since video recording is continuous write, I would never use QLC. If you really want to there are "industrial" SSD that is more expensive and use either MLC or SLC.

Still though, if you are not on battery a mechanical HDD like WD Purple and Seagate Skyhawk are the better choice. I have a few that ran for at least 7 years already and still alive. There are custom HDD commands for surveillance system that these HDD take to align data write so they can minimize seek, and make their shingled recording work reliably.

Is QLC really that low? Thats nuts! (Never bought one, but that’s systems robbery if really that low!)

 

[PandaBear]

Is QLC really that low? Thats nuts! (Never bought one, but that’s systems robbery if really that low!)

I have seen the worst of the worst memory being used for USB drive because study has shown people only write the whole drive's capacity 10 times in its lifetime, so they design it to last 50 total drive writes. The way it works inside the memory cell is a voltage stored inside that can be read between multiple voltages, let's say between 0V to 20V.

If you have a SLC that means anything between 0-10 is a 0 and 10-20 is a 1.
If you have a MLC that means anything between 0-5 is a 00, 5-10 is a 01, 10-15 is 10, 15-20 is 11, etc.
If you have a TLC that means 0-2.5 is 000, 2.5-5 is 001, ........ , 17.5-20 is 111, etc.
If you have a QLC that means 0-1.25 is 0000, 1.25-2.5 is 0001, ......, 18.75-20 is 1111, etc.

Simple enough, so now let's add some random noise to it.

Let's add a +/- 0.2V to all the number above, and see how many of those numbers would once in a while become "wrong" when new. Let's add another +/- 0.2V to all the number above every 500 cycles, then another +/- 0.2V again after another 500 cycles.

Yes you can keep doing that if you keep improving your error correction engine on your controller. As a matter of fact the biggest part of your flash controller today is the ECC engine, and they have gone from deterministic one to statistical one like LDPC, and rely on raid to correct those odd ball worst case scenario out there that can take maybe 500ms to fix. You will not lose data but your drive would likely be so slow it is unbearable.

Every time you add 1 bit you make it half as good and worse, and then your improvement in data density just get slightly better. Half the cost from SLC to MLC, 50% cheaper from MLC to TLC, 33% cheap from TLC to QLC....

I'm sure someday people will invent better error correction that uses physical world's behavior like how each key on your keyboard has only limited keys next to it, or a photo can only have certain color next to each other, to create specific drives for specific data (photo only ssd, or audio only ssd, etc), and keep going in that direction.

 

[meep]

I have seen the worst of the worst memory being used for USB drive because study has shown people only write the whole drive's capacity 10 times in its lifetime, so they design it to last 50 total drive writes. The way it works inside the memory cell is a voltage stored inside that can be read between multiple voltages, let's say between 0V to 20V.

If you have a SLC that means anything between 0-10 is a 0 and 10-20 is a 1.
If you have a MLC that means anything between 0-5 is a 00, 5-10 is a 01, 10-15 is 10, 15-20 is 11, etc.
If you have a TLC that means 0-2.5 is 000, 2.5-5 is 001, ........ , 17.5-20 is 111, etc.
If you have a QLC that means 0-1.25 is 0000, 1.25-2.5 is 0001, ......, 18.75-20 is 1111, etc.

Simple enough, so now let's add some random noise to it.

Let's add a +/- 0.2V to all the number above, and see how many of those numbers would once in a while become "wrong" when new. Let's add another +/- 0.2V to all the number above every 500 cycles, then another +/- 0.2V again after another 500 cycles.

Yes you can keep doing that if you keep improving your error correction engine on your controller. As a matter of fact the biggest part of your flash controller today is the ECC engine, and they have gone from deterministic one to statistical one like LDPC, and rely on raid to correct those odd ball worst case scenario out there that can take maybe 500ms to fix. You will not lose data but your drive would likely be so slow it is unbearable.

Every time you add 1 bit you make it half as good and worse, and then your improvement in data density just get slightly better. Half the cost from SLC to MLC, 50% cheaper from MLC to TLC, 33% cheap from TLC to QLC....

I'm sure someday people will invent better error correction that uses physical world's behavior like how each key on your keyboard has only limited keys next to it, or a photo can only have certain color next to each other, to create specific drives for specific data (photo only ssd, or audio only ssd, etc), and keep going in that direction.

That makes a lot of sense and also explains what would be some kind of logarithmic curve to failure based on the number of stored voltages within one cell. Thank you for taking the time to explain it.

For home use I’ve been very particular to only buy dual level, which has been pretty specific to older Samsung pro offerings. I’ve also got a few TLC Samsungs running and haven’t had any failures at all.

In fact, the only SSD failures I’ve seen either at home or work has been
- early intel 2.5” offerings
- a couple of the middle-grade consumer brands that our IT folks have asked for

Where do you currently sit on the side of commercial storage - spinning vs solid state for non-high-performance / general file storage use? I’m assuming spinning? Reliability out of spinning media on what I’m seeing is still excellent and perhaps improving?

 

[PandaBear]

I am all for using the right tool for the job. I use some older SATA SSD in boot, some cheap hand me down SSD for fast storage, HDD for archive, and some really old HDD for off site cold storage backup in case my house burn down. Data center still has a ton of HDD for non performance write that people may look for once in a while. Things like old record from 2 years ago, backup, disaster recovery backup in a different region, etc. For performance stuff people are using SSD to eliminate seek.