Intel Unveils One-Petabyte Storage Servers For Data Centers (theinquirer.net)

Slashdot reader #9,219 Guy Smiley shared this report on a new breed of high-density flash storage. The Inquirer reports: Intel has unveiled a brand new form factor for solid state disc drives (SSDs)... Intel Optane's new "ruler" format will allow up to a petabyte of storage on a single 1U server rack... By using 3D-NAND, the ruler crams in even more data and will provide more stability with less chance of catastrophic failure with data loss. The company has promised that the Ruler will have more bandwidth, input/output operations per second and lower latency than SAS... As part of the announcement, Intel also announced a range of "hard drive replacement" SSDs -- the S4500 and S4600 0 which are said to have the highest density 32-layer 3D NAND on the market, and are specifically aimed at data centres that want to move to solid state simply and if necessary, in stages.

Comments

Layers?

[DontBeAMoran]

32-layer 3D NAND

What do you mean? Like the T-800 or onions/parfaits?

Re: Layers?

[DontBeAMoran]

I think you're being a little too anal about this.

Re: Layers?

[LifesABeach]

Intel finally understands storage requirements for my Porn collection. But one question lingers, Intel has a "Rack"? I'm into Racks; does anyone have an image of them?

Re:Layers?

[unixisc]

Yeah, what exactly does 'layer' mean here? Are they trying to say 5 voltage levels, that would give 32 variables? Or do they actually mean 32 levels, which would be hairy to maintain? Somehow, I've never had much confidence in multi-level cell flash, where one needs more than 2 voltage levels to increase the 'bits/cell' and thereby the density. Spansion had an interesting idea in ORNAND, where they had reversible source-drains to store 2 bits in a cell, but from what I recall, that had performance issues due to lack of a real ground.

So how has the reliability story been on this, when compared to the first SSDs that we saw enter the market?

Re:Layers?

[mlyle]

It means 32 layers of cells in the chip, which themselves are probably TLC. 3D-NAND/VNAND are not new, and there's products with up to 64 layers in the market.

Re:Layers?

[ChumpusRex2003]

It means layers of transistors on each chip.

Traditionally, silicon ICs consist of a single layer of transistors, with several layers of metal wire interconnects placed on top. In other words, the transistors are arranged in a 2D array over the surface of the die.

The major breakthrough in flash memory a few years ago, was the development of "stacked transistors". In other words, multiple independent transistors could be stacked one on top of the other. This now typically goes by the name "3D", with current products now being 32 or 64 layers stacked. The net result is that larger lithography can be used for the individual flash memory cells (leakage due to quantum tunnelling becomes unacceptable once flash cells become too small) improving SNR/endurance, while simultaneously increasing areal density.

will they use SAS/SATA or pci-e or some intel only

[Joe_Dragon]

will they use SAS/SATA or pci-e or some intel only thing??

But I may need to get an AMD EPYC system to get the PCI-e lanes to make the most of it unless you can get 4 cpus into an 1U box.

Re: Old Hat

[crow]

No. This is in 1U of rack space. That's incredible density.

Re:Old Hat

[MiliusXP]

In 1U form factor ?

Competition?

[crow]

So is Intel also selling the raw components? In the past, they've been a neutral vendor. With this move, they could be making a huge jump into the storage industry, competing directly with HPE, Hitachi, and DellEMC.

Re:Competition?

[unixisc]

It would make sense only if Intel has problems filling fab capacity, but I fail to see that happening. If they (re-)enter that market, they'll drive down the market prices of flash, making it a loss for everyone! They were in both the NOR flash business as well as the NAND flash business years ago as Numonyx, which they then sold to Micron, and then they formed a partnership w/ Micron on this.

If they do that, I'd conclude that things are really bad at Intel

Re:Old Hat

[Nutria]

I shudder to think of the cost of a PB of flash RAM...

Re:Finally

[Anonymous Coward]

I can store my porn collection in 'the cloud'!

Where will you store it, in petafiles?

Re:Old Hat

[Jason1729]

Around 50 cents/gigabyte. So $500,000ish.

Re: Old Hat

[Jason1729]

At 50 cents per gig, the flash is worth about $500,000 for this PB. Spinning platters are under 5 cents per gig. $50,000 for this PB.

For most datacenter storage purposes, the spinning platters are not a bottleneck. The new product is cool and will be very useful in a few specific applications, but it is not going to change much for datacenters. The flash SSD will also save power as well as space, but not enough to justify 10x the upfront cost. Maybe next generation.

Re: Old Hat

[darkain]

This isn't about the "A or B" situation, it is about having a hybrid of both. Use magnetic storage for bulk data like photos and videos, and use this new SSD for database itself. This has pretty much always been the case of SSD vs HDD in the datacenter. Now we can grow databases exponentially larger.

Re: Old Hat

[Jason1729]

Read the post I was replying to to understand the context of what I said.

This is *not* going to lead to "Cumulative storage capacity of a datacenter has now the potential to grow enormously", which is the entire meaning of my previous post. There certainly will be uses for this product, but the higher density is not going to lead to higher capacity datacenters.

Daisy Daisy....

[lobiusmoop]

This image of the new drives loaded in a rack makes me think of this scene from 2001.

Ramming storage into a 42U rack is fine but...

[wwalker]

single 1U server rack.

That's a really small rack. Why would anyone create a server rack with just a single 1U capacity? /s

Re: Old Hat

[bill_mcgonigle]

Maybe the petabyte of xpoint flash is just the cache drive for your real data store.

Heck, you'd probably need eight of these just to cache an exabyte of backend.

Still not reliable yet for server use

[Billly Gates]

Many industrial controls with SSDs seem to always fail and one of the projects I worked on was replacing the SSDs that come with them with a hard disk as they never seem to have problems like the SSDs do.

I have seen other slashdotters on here who work in the enterprise who have loads of failed ssds on their desks, but their hard drives while slower are always less in quantity in comparison. It doesn't matter the brand. They all fail and when they do they go hard.

Re:Still not reliable yet for server use

[Anonymous Coward]

AC here due to NDA, I am long time Data Center fellow (15+ years,) I do hardware engineering for the Data Center space for a big company you all hate and love. One of my biggest projects was designing a cabinet that had the highest storage capacity in the world.

One thing I can speak for is SSDs. They are reliable, but they do have issues. Often times when you think an SSD has died it hasn't really died, usually it is due to a firmware bug. Last year I was going through a thousand SSD's from three manufactures that had been determined had failed by software or hardware troubleshooting by techs. In 60% of the failures, we were able to recover them back to regular use, the end causes being two fold, a firmware glitch that could be corrected, the other problem being the SMART data was not being read properly or improperly reported by the SSD.

We were able to implement a software solution that detected the problems that were found in the investigation and provide corrective action. This resulted in a huge drop in failures. Further interface with manufacturers has caused them to correct the various issues, which eventually rolls out to other enterprises.

Re:Still not reliable yet for server use

[thegarbz]

Pointless anecdote. Extraordinary high failures rates in SSDs point towards problems in your power supply or heating more than anything else. The fact you can't keep them running in industrial applications especially is a sign of this as solid state electronics don't magically degrade outside of the known read/write failure mechanism.

My own anecdote goes the opposite way. We have fun playing frisbie with the left over drive platters from the endless string of failures, and I'm sure every enterprise does as well given that HDD failure is just a given and everyone treats them like a normal part of the business maintenance.

I'll put my name to saying this doesn't just include SSDs, but some 90% of the hardware failures I've seen in industrial systems are due to external problems. 20 year old UPSes, noisy as shit power, overheated power-supplies, earthing designed by the spaghetti monster himself, lack of surge protection, inadequate ventilation, poor dust control, the list just goes on and on. SSDs are sensitive to power issues, that is know and given. Design accordingly and you won't have an issue.

Re: Old Hat

[Billly Gates]

If your application is Rainbow Tables (or similar) then it's totally worth it.

Or the new minimum requirement for Java 9 JRE

Re:OTOH, failue modes and rate?

[stabiesoft]

For me, I have had a usb drive plugged into a small atom box I used as a mail server/web server/ftp server. It ran literally for years as the /home mount which serviced some log files, the ftp and www files. I've since moved the servers, but the little box keeps on humming. So I'm not claiming reliable or unreliable, just YMMV!

Re: Old Hat

[DontBeAMoran]

The flash SSD will also save power as well as space, but not enough to justify 10x the upfront cost. Maybe next generation.

And if not the next generation, then maybe deep space nine, voyager or enterprise.

Re: Old Hat

[Neuronwelder]

Thank you Jason!!

Re:OTOH, failue modes and rate?

[Gondola]

Enterprises use backup power and power conditioning.

Hot-pluggable NVRAM - yeah!

[illtud]

OK, it costs a fortune at the moment, but can I just give a cheer for a hot-pluggable format for NVRAM? I've used the Samsung 960 pro M2 2TB, and it's blinding fast, but my Ops guys won't touch it for production as it means downtime on a failure (and it looks like we've got a failure after 4 months - thank goodness for warranties)

I'm looking forward to 1PB in 1U, but my prediction for that being realistic is 2022 @ $120k.

Re:OTOH, failue modes and rate?

[wolrahnaes]

High-end SSDs these days, including anything deserving of the description "enterprise grade", have supercapacitors for the specific purpose of allowing writes to complete in the event of power loss.

USB drives generally don't have something like that. I'm sure some might, but most don't.

Re:OTOH, failue modes and rate?

[drinkypoo]

Brand and model? I have had varying luck with different stuff. My OCZ ATV died and they replaced it with a RALLY2 which has been solid for years now. Out of a half dozen Sandisk SD cards I've got only one is bad, but out of three Sandisk USB drives, three are bad.

Re:Old Hat

[mikael]

Yes, there's a picture. It's 1U in height (1.752 inches, 44.5 millimeters). The SSD drive is like a thick white school ruler. Then these can be jammed in together 42 vertically and possibly the same horizontally.

https://en.wikipedia.org/wiki/...

Re:OTOH, failue modes and rate?

[HiThere]

OK, thanks. That finally sounds like a valid answer.