A 16TB SSD: OMG

By | August 13th, 2015

Samsung 16TB SSD

It was reported today by Ars Technica UK, Slash Gear and others that Samsung is introducing a 16TB SSD. The world’s largest hard drive was unveiled at the Flash Memory Summit in Santa Clara, California. What’s even more amazing is the 2.5″ form factor – it would fit in your laptop.

Before you run out (ok, log in to Amazon) to buy one, you need to know the price is expected to be between $5,000 and $7,000 each, at least initially. That price isn’t that crazy. If you were to buy (8) eight 2TB SSD drives today at $800 each it would be $6,400 and they would not all fit in your laptop.

Could Backblaze use these SSD’s to create a 720TB Storage Pod? Sure, it would cost about $220,000 for the drives and take a bit of modification to fit 2.5″ drives versus 3.5″ drives in a Storage Pod. We’ll leave the modifications up to Backblaze Labs to figure out. The real question is “would” we build a 720TB Storage Pod made from 16TB SSD drives? Let’s do the math…

  • $5,000 divided by 16TB of storage is $312.50 per TB of storage.
  • $312.50 per TB divided by 1,000 is $0.3125 per GB of storage.

Currently we pay about $0.032 per GB for traditional hard drives, a Seagate 4TB drive for example. The 16TB SSD is nearly 10 times the cost per GB. That said each 16TB SSD Storage Pod would save (3) three Storage Pod chassis using 4TB drives – that’s roughly $12,000 ($4,000 per chassis in rough numbers). That would still make a 16TB SSD Storage Pod about 9 times more expensive. Even after we put all the numbers into our spreadsheets that factor in electricity, rack space, drive failure, maintenance, etc. the 16TB solid state drives would still be 6-7 times more expensive.

I guess we’ll have to wait a little longer for prices to drop before we jump on creating an SSD Storage Pod. Too bad, it would have been fun to write about a 720TB Storage Pod or better yet a 14.4 Petabyte Backblaze Vault. Still, a 16TB SSD is pretty cool, I wonder what’s next?

 

Andy Klein

Andy Klein

Director of Product Marketing at Backblaze
Andy has 20+ years experience in technology marketing. He has shared his expertise in computer security and data backup at the Federal Trade Commission, Rootstech, RSA and over 100 other events. His current passion is to get everyone to back up their data before it's too late.
Category:  Cloud Storage
  • I have written to Gleb at Backblaze before about the advantages of the 2.5″ drive form factor. The general response was that it was not yet cost effective. The tipping point has been passed. The cost benefits of the 2.5″ drives is now clear.

    The cost parity of 2.5″ drives has arrived. The Seagate 4 TB drives are retailing for $0.032/GB, well below the latest published $0.044 cost for POD 6.0 storage.

    The R&D in 3.5″ technology is dwindling, as the 2.5″ form factor gains popularity. The 3.5″ drives will fade from use as did the 5″, 8″ and 14″ drives before them.

    The cost per bit of SSDs continues to plummet; manufacture cost is less than rotating media and manufacturing cost delta for different densities is negligible. This means that a 16 TB drive could profitably retail for under $100 and the 60 TB drives are already in the lab. Yes, they will gouge $10K for the early production units, but that will all as production volumes increase.

    Using a COTS 2.5″ form factor the POD storage: density can be increased 270%, rack usage reduced by 25%, POD depth reduced, a single EVGA 750 power supply should be sufficient, POD mass reduced, thermal loading reduced, and power consumption reduced (providing a double power savings, as cooling would be reduced).

    There are x24, x28, X36, x40 & X48 port expanders in production. A single back plane can support many drives and provide a 12 Gbps back-haul to the motherboard. In the 6.0 POD 4 such expanders could be installed; with lots of extra room.

    The 2.5″ performance characteristics will continue improve, realistically another 4 times in the immediate future and that again in the foreseeable future.

    Conclusion: By going to a 2.5″ storage form factor, a 3U POD can be profitably made using currently available components, with multiple secondary benefits. And the increasing storage density trend will increase an order of magnitude in the foreseeable future.

    Regards,

    – Rick Peralta

    • BTW: the cost to profitably manufacture and market a 16 TB SSD is probably about $50.
      And the cost to manufacture the emerging 60 TB drives is probably around $100.
      Yes, the R&D NRE and executives with vision will need to be compensated, so don’t expect those prices too soon.

  • Jorge Schuldt

    Good post ! Speaking of which , you require a a form , my business filled a blank form here http://goo.gl/VCqupE

  • Kv F

    I am building a storage pod of my own with the 16tb ssds, not by cause its cheaper, because its cooler, more efficient and faster.
    Better is in the eye of the beholder.

  • Pavel Ni

    At quantities you use, I believe it’s cheaper to simply buy NAND flash on open market and send it to factory in China to make one huge “single board flash NAS.” One remarkable feature of new 3D nand chips is their record low power consumption per GB. Still more than a spun down HDD, but less than one running on full throttle. It means you can use cheaper PSUs that don’t need to endure the current spike during the spin-up.

    If you are very lucky, you can buy cheapest nand in batches from $0.16 per GB

    • Matching the manufacturing and quality of mass production is difficult.
      Sticking to standard COTS components allows for shifts in vendors.
      That way if one company lets production quality slip, shift to the next.
      That sort of market pressure will be a win-win as we all benefit from such competition.

      • Pavel Ni

        Nah, even cheapest Chinese SSDs with worst flash chips available are sold at 150%+ of the component cost, there is merit in making huge SSDs in house.

  • Richard Thompson

    It will be interesting if you guys get to 2.5″ form factor with SSDs in the future. As stated before, reduction in power, vibration, etc – all great things. 1G ports would have to go to 10G.. ;-)

  • James G

    Rad. Can’t wait to have a lot more storage on our personal devices for storing high resolution content like 4k videos. Streaming is nice, sometimes, but in the U.S. we can’t count on carriers to innovate fast enough to bring us Gigabit broadband.

  • Karrie P Bond

    I love seeing the behind the scenes calculations and the actual numbers of how that decision is made. For business owners like me, it’s fun to see someone else crunch theirs.

  • Dan

    $220k? Where can I buy one? (chuckles) That would be amazing though!!!

  • Based on current trends, when will we start measuring storage in exabytes?

    • Andy Klein

      The NSA is already talking yottabytes (https://www.backblaze.com/blog/nsa-might-want-some-backblaze-pods). For the rest of us, Petabytes of storage for a business in not unreal even today. A few might crack the Exabyte barrier, Google comes to mind. For consumers, Terabytes should do for the time being.

      • Right, forgot about petabytes. Yottabytes are only mega-petabytes. Star Trek: TNG measures data in giga-quads, which I think are equivalent to zettabytes. Fortunately, we have a century or two to get there.

  • tinynot

    question for everyone here, i build my own towers and, well, is it possible for the solid state tech they have now to ever replace the ram/dram, memory that is used in PC’s. heck i hope you all even understood what i am trying to ask. seems to me to be a good question. anyone? and depending on what the answer is, why is it that they have not done so yet?

    • conbal

      i guess no one wanted to chime in, so I’ll take a stab at it. My guess is that the SSDs have fast enough seek times for file system I/O, but NOT anywhere near fast enough for application memory (RAM). I’m too lazy to look it up, but that’s my guess.

      • tinynot

        thank you for taking a stab at it, now about what you stated. I seem to remember software that added some of your harddrive as needed for more space for use as MEM back when the amount of MEM on the sticks at the time wasnt quite enough. and, i know less then you do about the sticks of MEM/ram they have today, when a system loses power, whatever is only in the RAM is lost, could they now have a stick with some SSD on it or something so it would not be lost, just a thought, and i have no idea if it is at all possible, thinking out of the box i guess i am tring to do, with the limited knowledge i have. even back when i started with the 486dx66, or whatever it was. 1MB of RAM was like $100usd, and the stick never had any moving part, if RAM is not SSD, then what is it, again just asking! just seems they could do better if anyone was even working on it, is all!

        • Bryan

          Hey Tinynot. I think what you’re referring to is Microsoft’s paging file system (PFS) which allocates a portion of the hard drive to supplement the system memory which is normally used. I don’t see any technical reason why there couldn’t be a mini SSD on a stick of RAM aside from the fact that PFS already does this without bottling down the stick.

    • A bit off topic, but an Intriguing question.

      First, there are systems that use FLASH, but through a filesystem, not as primary storage. Most cell phones do this, as to many tablets.

      FLASH tends to fatigue with use. So, DRAM would still be needed for primary memory.

      Prospects of a 1 TByte tablet is interesting.

  • Screamer

    I was a bit disappointed when you stated, “The world’s largest hard drive….”. But it’s a good read, nonetheless.

    • Andy Klein

      Check out the title of the Ars Technica article linked to in the post. Who am I to argue with them. OK, the Samsung presentation shown does say “World’s Largest Drive”, which is much better, but perhaps we should say World’s Largest Storage Drive to be precise.

  • Few years ago I wrote to Gleb & Tim that switching to 2.5” form factor for drives would be feasible from the space savings point of view. You could easily put the drives in 3U form factor pods. Perhaps you’ll be able to fit not 15, but 35 drives in a row and 5 instead of 3 rows, totaling 175 drives per 3U pod. Even for that-time HDD drives, they would consume less power (both idle & spin up) than ‘classic’ v2.0 Pod.

    This time it’s a high time to start working on this seriously. Without moving parts, and with fully closed chassis, you could drop entire isolation & anti-vibration stuff, gaining even more space. You’d have to solder your own version of SATA-multiplier backplanes.

    For that capacity & speed of drives, two 1G eth would definitely be a bottleneck.

    Gosh, sometimes I wish I worked at your labs!

    • Heh, it’s still a good suggestion, but at the moment it’s simply price-prohibitive. We need to reduce costs on storage as much as possible and at the moment the easiest way to do that is to use 3.5″ HDDs, but we hope that changes soon!

      • 2.5″ drives are at cost parity with 3.5″ drives!

        A Samsung 4TB SATA 2.5-inch Hard Drive (ST4000LM016 ) is retailing for $136.95 with free shipping. That is $0.034 per Gigabyte.

  • bertie_booger

    The other MAJOR advantage is IOPS. 2,000,000 of ’em to be exact. 2 Million EACH DRIVE (*drool*) Now that might not be such an advantage for backups perhaps…but hey…who’s complaining?
    Now if you’d please excuse me while I go check my shorts.

    • Sakebomb

      That figure is in correct it’s not per drive. It was for 45 drives setup in a server configure. So 2 mill /45 is ~44,000 iops per drive. While I’m sure there is overhead in this conf. No official numbers have been posted on the drive specs yet. The interface is the bottleneck. If they come out with an NVmE version that number will go up.

      “>>Samsung showed off a server with 48 of these new SSDs, with a total storage capacity of 768 terabytes and performance rated at 2,000,000 IOPS (input/output operations per second). By comparison, the consumer-grade SSD that you have in your PC is probably capable of around 10,000-90,000 IOPS, depending on the workload.”

      • bertie_booger

        Ahhh “comment in haste, repent at leisure” Indeed good sir, you are entirely correct.

    • For Backblaze, the IOP rate is not critical… nor is bandwidth.

      Sustaining 22 MBit per second from all 45 drives is probably enough to fill a 10 Gbps NIC interface.

      That said, a high end storage server would benefit from excess bandwidth, low latency and high transaction rate.

      Looking at SSDs, they generally can sustain 32K – 200K IOPs. With an array of 45 or 200 the aggregate rate can get pretty high. And the amazing thing is that the FLASH devices themselves can go orders of magnitude faster.

      Considering raw numbers is often deceiving. For example an array of 45 16TB drives could provide 200 Gbps of data and packing the case full could provide a terabit of bandwidth. But who could be there to read it? There are not yet network interfaces or primary buses that can move data that fast.

  • disqus_j335f41OE4

    Andy, an SSD RAID is an all-new paradigm both due both to the ultimate speed, and also the constant read & write speeds regardless of which “virtual sector” each chunk of data lies — And this can be leveraged in several ways:
    A) Synchronizing spindles becomes a non-issue, which changes the fundamental assumptions used for electromechanical fault-tolerant disk arrays;
    B) Speed of RAID rebuild would be much faster: Assuming you stay with a 45 drive pod, all 45 drives could be linked together into a single RAID level 3, with perhaps 3 or 4 dedicated parity drives per pod.
    Let’s go back to the RAID definitions and ask ourselves these three fundamental questions:
    1) If we were to use (hypothetical) conventional hard drives each with a battery backed-up RAM cache the size of the drive, which RAID level would be optimum? This is essentially the model of an SSD (NVRAM) array;
    2) Because of the factors listed above, pods can have many times the 45 drives in a single RAID array, as if any drive fails, the whole array can be rebuilt in minutes, not hours. How can this be leveraged?
    3) If we were to start with a clean sheet of paper to build an entire rack, how would we do it?

    Dan Schwartz
    [email protected]

  • Chris Moore

    Can you please share the source for your port multiplier back-plane board?

    • tinynot

      what is a “port multiplier back-plane board?

      • Chris Moore

        Review the design: https://www.backblaze.com/blog/petabytes-on-a-budget-how-to-build-cheap-cloud-storage/
        They went to direct cable with version 4.0, but they have gone back to the backplane topology with the 4.5 design. According to what I have seen in this blog.

        • tinynot

          i still dont know what that is, i am stupid with tech. things, cant ya tell by now? i am awsomed by the 16TG SSD samsung though, that is so awsome, i want two of them for Xmas please?

          • Tex

            Basically they take 1 SATA port and turn it into 4 (or more) via an add on card/cable.

  • Billy

    How many 2.5″ drives could you fit in a pod? Surely more than the 45 3.5″ drives that you currently fit. You could have easily over 1PB in a pod!

    • Andy Klein

      A 1PB pod, now you’re talkin’. Not sure how many 2.5″ drives could fit into a Storage Pod, you’d need a good way to connect them to power and SATA. Might need a bigger power supply and more SATA connections. Wiring would be pretty hairy too. Still one can dream…

      • Geostyx

        But since SSD’s use way less power, maybe power wouldn’t be too bad? SATA on the other hand..

      • Chris Moore

        The data interface would be the hard part. Since there is no motor to spin, SSD is way less power intensive.

    • Reworking the 4U POD into 3U chasses, 4 rows of 2.5″ drives would fit, 36 across for a total of 144 drives, the POD could be shortened to the standard depth and dual 10 Gbe NICs could be easily saturated with traffic.

      Populating: 4 TB per drive that is 576 TB, 16 TB drives 2 PB and 60 TB drives 8 PB.

  • gadget_hero

    I can only imagine how fast a pod with those Samsung drives would be!

    • Riely Rumfort

      ~14-16GBps

  • ᅠᅠᅠ

    That’s neat, can’t wait for this to trickle down to consumer levels. Of course it takes a while for new technologies to return their research investment and drop to reasonable pricing levels, but I still feel like SSD manufacturers are milking those things to the utmost at the moment. They seem incredibly reluctant to start offering larger drives to the public, while the tiny ones still sell.

    • tinynot

      oh yes, and i may add to what you said, what the heck is with those “hybreds” a small sdd with like a tb or so of non-ssd? i remember when ssd’s started coming out a few years back, i said then what i am saying now. SSD’s are the future, well heck with that, they dragging da feet, and they milking it for all they can, heck with that. i hope this wakes them up and tells em, bring out now before you get slammed by some of your competition that dont wait….hehehe the public wants em now, i want them now. i would love to put in a 5tb into my lappies, that would be awesome. trickle, yes it seems they like to trickle on us at the consumer level, there ought to be a law, heya?

    • The cost/density of drives has driven a 4 TB laptop drive under $150, and 500 GB SSD for about the same money.