While I'd have no need for anywhere near 100+ TB, I'll be really happy when SSDs are cheaper than spinning disks. If they can make a 200TB drive for business, a 20TB drive for home users should be pretty reasonable.
My experience with SSDs and rotational drives (which may be somewhat dated and also biased by the specific models I was exposed to) is SSD failure rate is significantly lower than spinning drives, although I don't think I got much experience at 5+ years with SSDs, maybe there's a cliff. The downside is that SSD failures had no known prefailure indicators and tended to result in a completely inaccessible device with no chance of a last minute copy, whereas rotational drives often give signs (SMART metrics/warnings, failed reads of specific sectors, etc) and a last minute copy can be pretty effective if the signs are noticed in time. Sure, you should have backup protocols, but many people don't... so probability of an incident goes down, but probability of full data loss given an incident goes way up. In theory, SSDs are supposed to detect when they wear out and go to read only mode, but I've never seen that happen, and I've not run write heavy workloads that got them close to wearing out anyway.
I will say, SSD pricing is pretty frustrating. There's a lot of focus on faster and faster nvme that's barely noticeable while actual capacity has really stagnated at around 2-4tb. As someone who is very into photography, it takes not very long to fill 2tb of space. If I could buy a cheap 16tb ssd with fast reads and slow writes I'd be a happy camper but it doesn't exist unless I want to buy spinning disks (I don't)
Yeah, I'd gladly take a cheap large capacity SSD that can only do 500MB/s for storing data. It's still gonna be much faster, quieter and cooler than a HDD.
Looking at the prices, they have barely gone down at all in the last 3 years (for the same specs). And the best price per capacity is still at 1TB.
The prices kinda look like they are going down a tiny bit, and thats because companies keep introducing ssd’s without dram cache. One those are halfway full their writes are maybe 3/4 fast as a hdd and they might last for 2-3 years
I had big hopes for Samsung’s QVO line but they’ve barely done anything with it. They’re often not much cheaper than EVO. And the 8TB drive is almost £200 more expensive than 20TB of spinning rust. I could get 30TB+ storage for the same money.
Have you looked at something like Intel 660p? They're pretty crappy as SSDs go, but they're quite cheap (by about 40% compared to what I would put into my desktop, at least where I live).
The Intel 660p is a higher-tier product than Samsung QVO drives. Both use QLC, but the Intel drives are NVMe and thus have much higher peak performance (when operating out of the SLC cache). But Intel wasn't interested in pursuing the niche of users who wanted more than 2TB in a consumer drive.
I wouldnt touch Intel SSD. They ship documentation stating read-only failure mode in case of reaching end of life/defects. In a big endurance test some time ago Intel drive simply died and vanished from the system on reboot, followed up by Intel mouthpiece claiming _this is by design_.
"After a reboot, the SSD disappeared completely from the Intel software. It was still detected by the storage driver, but only as an inaccessible, 0GB SATA device.
According to Intel, this end-of-life behavior generally matches what’s supposed to happen. The write errors suggest the 335 Series had entered read-only mode. When the power is cycled in this state, a sort of self-destruct mechanism is triggered, rendering the drive unresponsive. "
Intel bribe scandal was 20 years ago, and they still do it after tweaking (clients receive free hardware instead of straight $) to not be so blatantly illegal. "Cripple AMD" ICC/MKL was 13 years ago yet "Arc 380 drivers do not support ReBar on AMD's Ryzen CPUs, the functionality only works with this technology on Intel's own processors."
That makes sense considering that QLC (2^4 charge levels) only stores 33% more data per cell than TLC (2^3 charge levels). Therefore at best it can only be 33% cheaper/bigger.
Quad bits to triple bits is 4/3=1.33.
Taking your way of counting to the extreme: if I have a cell that can handle 1 million bits, I can't just add an extra bit of storage for another 1 million bits of storage. I have to actually add the full megabit of storage.
Well, for me it's noise, power/heat and shock resistance (in laptops). And these have also indirect negative consequences. For example, an external 3.5" HDD needs an additional power supply and cable in addition to the data cable, which is cumbersome and reduces portability.
Not the op, but I've just switched away from spinning disks recently. Kinda.
I have a NAS of 4x 8TB WD Red Pro; and a a desktop with 4x1 TB of nvME and 1TB of (grandfathered) SATA SSD.
For batch photo processing (i.e. lightroom) spinning disk is really sub-optimal. I'd rather have a 256GB or 512GB nvme and process batch by batch (and then offload to NAS), then a 8TB 7200 RPM HDD with all my photos.
To OPs point - depending on shooting style, you might easily do 300GB+ over a single wedding (primary + secondary shooter, 2 cameras each, 46MP, 14-bit uncompressed, spray-and-pray style).
I've heard but don't know if its true that SSD storage can constantly suffer data corruption if its not powered. It needs power to keep refreshing stored data.
Something that maybe spinning disks don't suffer from.
I don't believe SSD loses data over arbitrarily short periods of time such as days/weeks.
On the other hand, I would not trust either medium over extended period of time (years) personally - either dump to tape/optical, or keep making copies :).
Either way, I currently do not have a use case where either my SSDs or HDDs would be off for year or more.
(I do have a set of "off site" HDDs with my family which do stay unpowered for months before being exchanged)
I suspect the reason is due to high base costs. Throwing on more dram chips and a better controller probably don’t cost much, but drastically effect performance. So, the cost savings of a simplified drive may only be a few dollars.
There's a good chunk of price to performance tradeoff. Taking samsung as an example, currently their lower end is $95/TB and their higher end is $125/TB. WD currently has a cheap QLC for $75/TB and a gen4 drive for $140/TB.
This is looking mostly around the 2TB range, and mostly M.2
It's hard to physically fit large amounts onto an M.2, but a 2.5 inch SATA drive would hold a big pile of QLC flash just fine.
I suppose the SATA bottleneck just looks bad. Maybe if we had better consumer support for 2.5 inch drives with NVMe then companies would be more willing to make that size.
Silly question - what makes you dislike buying four 4tb SSDs compared to buying one 16tb SSD? I could see it being less portable to have multiple drives, but I would presume that 16 TB isn’t really something that needs to be brought around portably unless you’re going somewhere for extended period of time without internet (preventing the use of a NAS/cloud storage). Multiple SSDs also allows the use of RAID for increasing speed and/or susceptibility of failure.
if you are moving that data around, nas/cloud storage can be super frustrating. copying 16tb of data over usb3/c is much faster than 100mbps or 200mbps internet.
If one SSD has a failure rate of 5%, 4 SSDs will have a failure rate of 18.6%. If the 4 devices are in a RAID0 for convenience (don't have to figure out where to put files), there is an 18.6% chance of losing all data vs 5% with a 16TB drive. You can use RAID5/6 to mitigate the risk but then lose capacity.
IMO, higher SSD speeds are not a target for improvement by manufacturers, but a side-effect of using higher-density chips, which is more cost-effective. They probably can't make a slow, high-capacity SSD economically.
So we desperately need a parity drive, which will bring the odds of data loss down to 2.3% for 4+1. (Though there are other factors that significantly complicate these odds.)
The point is to not lose any data. In the example, if you need 4 drives to store 16 TB of data instead of 1, your probability of data loss is 18.6% vs 5%.
Not the person you replied to, but bulk exports from something like Lightroom or Capture One benefit from faster speeds. If you have a fast enough CPU and enough memory, the main bottleneck is I/O. It doesn't matter for a few photos, but when you want to (for example) convert a whole trip worth of processed raws to JPG, every MB/sec counts.
I have a Canon R5, and i ended up with almost 800gb of storage in 8k raw, for a concert I did. I didn't even take that many videos as the camera overheated.
But a 4:20 video was 77gb. I have a 8bay(5 disks in raid 5) nas, but the performance isn't so good. Not good enough to edit, I use the 1tb nvme I have, but lots of swapping when I need to edit.
AFAICT this has been documented quite well. Use a SSD as cache. Easy to do with ZFS. The one I use on my NAS is one TB, it cost 200 EUR. It has the benefit of not causing data loss at power loss.
Most technology seems to chase weird benchmark numbers as it seems the market decides as such. It was mhz to ghz through 2010, then it was cores, then these SSD numbers, or the refresh rates.
Most of it seems absurd to the majority of users, even if you could subtract the placebo effect.
There aren't a lot of use cases for putting that much flash storage behind the bottleneck of a single SAS or SATA port. Most of the time when you're paying tens of thousands for flash, it's worth paying for enough IO bandwidth to actually be able to use all the performance that mountain of flash can offer.
I have basic understanding of a generic process. I mean specifically 200+ layers. Do they “just” deposit layer after layer after layer, 200-something times?
I am not going to pretend to be a process engineer (I am chemical engineer by education) but I can't imagine any other way to do it than as you described. Would really love for someone more knowledgeable to chime in though.
They deposit half the layers, etch holes through them and form vertical strings of memory cells, then repeat all that to build a second deck of layers on top. The interface between decks is a potential source of problems, but is more manageable than trying to drastically increase the aspect ratio of the holes that need to be etched through the stack.
Etching through lots of layers at once is why 3D NAND flash memory is actually economical, compared to basically every other kind 3D chips where layers have to be built one at a time.
Pure is basically an enterprise flash storage company. Great products, but FlashBlade specifically has issues with maximum single stream performance due to their architecture.
I think it's less of an ad and more an in depth explanation about the technology that goes into high capacity SSDs. Today, Pure just happens to be one of the only companies with a product on the market. That will no doubt change in a few months.
I am not familiar with this. I just watched the video yesterday and found it interesting to know that enterprises are considering alternative controllers, and also caches which would enable SSD with larger sizes.
Capacity is not the major problem. It is the $/TB price points. SSDs are still 5x more expensive than HDD for bulk storage. It doesn't matter if you can buy a 200TB SSD if it is 200x more expensive than a 1TB SSD.
HAMR technology is going to give us much bigger HDD. If I can buy a 50TB HDD or a 50TB SSD in a couple years, which one will I buy if the SSD is 5x more expensive?
SSDs make total sense for storing the files you need on a regular basis (e.g. your operating system and most used applications). They are just too expensive right now to store your massive video collection on them.
You could say the same thing about CPUs. Do you buy the $400 CPU that has good performance? Or do you splurge for the top of the line Threadripper that costs a few thousand? It is always a trade-off between price and performance. If the situation is right and your job justifies saving a few minutes for lots of money, then go for it.
If I needed a ton of disk space and I could buy a 50TB HDD for $1000 or a 50TB SSD for $5000 right now, I would opt for the HDD. But that is just me.
resilience too, especially if you're just mostly storing the data and not overwriting it constantly which is the actual death of SSDs whereas it's is mechanical wear with HDDs
For 50Tb? I'm not one to question anyone's data hoarding, but I just can't imagine needing silence or random access on 50Tb, at least not enough to justify 5x prices. By all means, give me 5Tb in my laptop at lightning speed, I'll pay through the nose for it. But in bulk? I can't imagine wanting to spend thousands when hundreds will do.
price isn't the only thing to consider. speed and density matter too. if I have 1pb of storage across 10 servers on hdds, the networking cost will be much higher than 1pb across 2 servers on ssds, and will the retrieval times. when speed and density are considering factors, ssds make sense.
when I needed 1.2pb of data fully searchable with retrieval times under 3s (across the entire data set, or huge portions.like half or quarter of the data), hdd wasn't even on the table.
but we did use hdd for off-site backups of the data in case we hit a huge shtf scenario.
If I'm buying for my home 6 bay NAS, I'll take the savings and buy the HDDs. As it stands now, network performance is my bottleneck, not the hard drives.
I'll only consider SSD for that system when the price is competitive. And I do spend extra on NAS HDDs.
79 comments
[ 2.6 ms ] story [ 118 ms ] threadBut I would love to have a pure SSD NAS.
Looking at the prices, they have barely gone down at all in the last 3 years (for the same specs). And the best price per capacity is still at 1TB.
https://techreport.com/review/26523/the-ssd-endurance-experi...
"After a reboot, the SSD disappeared completely from the Intel software. It was still detected by the storage driver, but only as an inaccessible, 0GB SATA device.
According to Intel, this end-of-life behavior generally matches what’s supposed to happen. The write errors suggest the 335 Series had entered read-only mode. When the power is cycled in this state, a sort of self-destruct mechanism is triggered, rendering the drive unresponsive. "
I have a NAS of 4x 8TB WD Red Pro; and a a desktop with 4x1 TB of nvME and 1TB of (grandfathered) SATA SSD.
For batch photo processing (i.e. lightroom) spinning disk is really sub-optimal. I'd rather have a 256GB or 512GB nvme and process batch by batch (and then offload to NAS), then a 8TB 7200 RPM HDD with all my photos.
To OPs point - depending on shooting style, you might easily do 300GB+ over a single wedding (primary + secondary shooter, 2 cameras each, 46MP, 14-bit uncompressed, spray-and-pray style).
Something that maybe spinning disks don't suffer from.
On the other hand, I would not trust either medium over extended period of time (years) personally - either dump to tape/optical, or keep making copies :).
Either way, I currently do not have a use case where either my SSDs or HDDs would be off for year or more.
(I do have a set of "off site" HDDs with my family which do stay unpowered for months before being exchanged)
This is looking mostly around the 2TB range, and mostly M.2
It's hard to physically fit large amounts onto an M.2, but a 2.5 inch SATA drive would hold a big pile of QLC flash just fine.
I suppose the SATA bottleneck just looks bad. Maybe if we had better consumer support for 2.5 inch drives with NVMe then companies would be more willing to make that size.
IMO, higher SSD speeds are not a target for improvement by manufacturers, but a side-effect of using higher-density chips, which is more cost-effective. They probably can't make a slow, high-capacity SSD economically.
That doesn't make sense, the probability that all 4 SSDs will simultaneously fail at any given moment, assuming a uniform distribution, is p(0.05)^4.
So we desperately need a parity drive, which will bring the odds of data loss down to 2.3% for 4+1. (Though there are other factors that significantly complicate these odds.)
Disclaimer: I got a D in Probability & Statistics
Opening a single picture is a sequential operation, as well as dumping pictures from camera to disk.
Am I missing something obvious?
Exactly what i was thinking!
But a 4:20 video was 77gb. I have a 8bay(5 disks in raid 5) nas, but the performance isn't so good. Not good enough to edit, I use the 1tb nvme I have, but lots of swapping when I need to edit.
Most of it seems absurd to the majority of users, even if you could subtract the placebo effect.
How do they manufacture this monster?
Etching through lots of layers at once is why 3D NAND flash memory is actually economical, compared to basically every other kind 3D chips where layers have to be built one at a time.
They introduce in the program PureStorage's FlashBlade, which uses Intel Xeon CPUs as the controller for SSDs. https://www.purestorage.com/products/file-and-object/flashbl...
But why are you posting an ad for Pure?
HAMR technology is going to give us much bigger HDD. If I can buy a 50TB HDD or a 50TB SSD in a couple years, which one will I buy if the SSD is 5x more expensive?
SSDs make total sense for storing the files you need on a regular basis (e.g. your operating system and most used applications). They are just too expensive right now to store your massive video collection on them.
Only if you don't care about access time or transfer speed.
If sitting there waiting 15 minutes costs you real money then having a bunch of fast but smaller SSDs quickly starts to makes sense.
Napkin math:
If 15 min/day costs you $15 then in a month $15*5d*4w=$300, or north of $3000 a year.
And usually if something costs money per time it's more than a $15.
SSD, because silence and random-access performance.
If I needed a ton of disk space and I could buy a 50TB HDD for $1000 or a 50TB SSD for $5000 right now, I would opt for the HDD. But that is just me.
when I needed 1.2pb of data fully searchable with retrieval times under 3s (across the entire data set, or huge portions.like half or quarter of the data), hdd wasn't even on the table.
but we did use hdd for off-site backups of the data in case we hit a huge shtf scenario.
I'll only consider SSD for that system when the price is competitive. And I do spend extra on NAS HDDs.
Currently prices are:
256GB $0
512GB $200
1TB $380
2TB $740
So the equation seems to be Price(n) = 2 * Price(n - 1) - 20 where n is the number of upgrades from base, and Price(1) = $200.
Let's assume each upgrade is a doubling of capacity, so extrapolating...
128TB $46100
256TB $92180
So 200TB should be somewhere in between, right around $70K.
> By 2018, they all expect the gains to accelerate with predictions of 128 TB and 256 TB, 2.5-inch, read-optimized flash SSDs
https://www.techtarget.com/searchstorage/feature/Almost-time...
256 GB 512 GB 10 TB 100 TB