Seagate’s Archive HDD uses shingled recording to squeeze 1.33TB per platter

In August, Seagate revealed that it had started shipping its first 8TB hard drive. Details were scarce at the time, but the drive has finally appeared on Seagate's website along with fresh information on what makes it tick.

As we suspected this summer, the new Archive HDD family uses Shingled Magnetic Recording (SMR). Instead of laying down tracks side-by-side with space between them, SMR uses overlapping tracks that can be packed more tightly. Seagate credits this approach with a 25% increase in storage density.

The SMR promo video isn't quite as catchy the classic Get Perpendicular, but it illustrates the shingling approach nicely:

The Archive HDD family comes in 8TB, 6TB, and 5TB capacities. It packs "up to" 1.33TB per platter, suggesting the smaller sizes have lower areal densities. The 8TB and 6TB units stack six platters, while the 5TB has only four.

All of the Archive HDDs sport 6Gbps interfaces and 128MB DRAM caches. The datasheet doesn't detail the rotational speed, but this Amazon listing indicates that the platters spin at 5,900 RPM. These drives are designed for archival storage and cold data that's accessed infrequently, so performance isn't a primary concern.

The Archive HDDs are rated for 180TB of writes per year, and they're covered by a three-year warranty. That lone Amazon listing prices a 20-pack of 8TB drives at $5,336, which works out to $267 apiece. The seller behind that listing expects to have stock in early January.

Comments closed
    • Bensam123
    • 5 years ago

    Love to see some performance data on this. I don’t think the 5,900 RPM tells the whole story.

    • Generic
    • 5 years ago

    The perpendicular video will be surpassed five years after the internet stops referencing the perpendicular video.

    • PainIs4ThaWeak1
    • 5 years ago

    Forgive my ignorance but, I’m assuming these likely won’t be good candidates for media server duties? And do we imagine these drives would be compatible with existing RAID technologies (hardware/software/or otherwise)?

      • f0d
      • 5 years ago

      im assuming the opposite
      these would be perfect for media servers (or at least my use as a media server) as i pretty much only write files once to a drive on mine and it just stays there for all time and when i run out of room i buy more hard drives

      raid arrays are best used as backup for your main system or as backup for some other system that usually changes a lot so that means lots of writes
      also most raid drives have special “technology’s” that makes them good as raid and compatibility with most raid box’s, that doesnt mean these new drives cant be used in raid it just means it wont be as good as a raid specific drive

      im actually going to get some of these for my htpc if they are cheap enough

      • sjl
      • 5 years ago

      I’m going to guess that this technology will have an impact on write performance, but probably not so much read performance. A media server… depends. How much data is being written out? If it’s just acting as a dumping ground for videos that are rarely written, but often read, like an archival storage of media, it shouldn’t be a problem. A DVR, on the other hand, might have problems coping, unless you put a cache layer – a faster hard disk, or an SSD, for example – in to buffer writes.

      As for RAID, you can shove any old disk into a RAID array and it’ll work. The only question is how well; for something like RAID 5, where you have significant write amplification (one write from the OS becomes two reads and two writes at the array level – roughly; there are ways to optimise that depending on how much is being written and where), I have serious doubts; the performance hit is likely to be too severe. The devil is in the details: just what, exactly, are the performance implications of this design? Once we know that, we can answer with more than just speculation.

        • cygnus1
        • 5 years ago

        I’m curious how you’ve come to the conclusion that “one write from the OS becomes two reads and two writes at the array level – roughly”. Do you mean 2 reads and 2 writes for each member of the array? I’ve never heard of this. Please explain.

          • sjl
          • 5 years ago

          Ok. RAID 5 101.

          You have a number of drives set up in a stripe configuration; one of those drives is a parity drive: it holds redundant information such that any single disk can be rebuilt if it’s lost from the data on the other drives. Every time you write data to the RAID 5 set, that parity information has to be kept up to date.

          Suppose you have a single sector write. In order to update the parity information, you need to (1) read the original data off that sector; (2) read the parity information that corresponds to that sector; (3) calculate updated parity information (which you can do with just those two pieces of old information, plus the new data – I can explain that in further detail if you need it, or I can just handwave it); and (4) write out the new data and parity sectors.

          If you’re doing a full stripe write – same sector on all the disks – then obviously you can just calculate the new parity from that data, with no need to read anything. An almost full strip write (say, missing one or two disks) can be optimised by reading the unchanged data from disk. But a two-read-two-write amplification is the worst case scenario, and most likely in a home environment to boot. (Handwave, handwave.)

            • cygnus1
            • 5 years ago

            Well, since you want to be an ass about it. Here’s RAID5 201

            One error in the previous RAID5 101 BS was stating that there is a “parity drive” in the array. Contrary to your misconception, RAID5 does not have a dedicated parity drive. That would be, the not used in years, RAID3 or RAID4. In RAID5 parity is distributed amongst all members of the array. This is to improve performance over RAID3 or RAID4 and prevent two things during an array rebuild: a single drive bottleneck and/or over-stressing a single disk causing a 2nd failure. (http://en.wikipedia.org/wiki/Standard_RAID_levels)

            Second issue is that you misunderstand the consequence of parity write amplification. The “extra” parity writes aren’t sent to the same disk. RAID5 does not increase the number of writes going to an individual member of the array. It does increases the total number of writes vs what the OS is aware of, but they are spread amongst all the members. In a hardware RAID implementation, even parity write amplification is inconsequential as the additional writes are completed in parallel, you know, because they’re to different drives. So parity write amplification has no bearing on a SSD vs mechanical HD discussion. And it does not increase the wear on any one drive in the array. It simply means that every write will involve multiple drives. Which… is kind of the point of RAID5

            Also, parity write amplification is not the same as SSD write amplification. SSD write amplification is a symptom of how flash programming is done in large pages. To modify part of a logical sector, which is a small part of a flash page, the entire page must be rewritten, thus amplifying the number of bytes actually being written to flash for each byte sent by the host. (http://en.wikipedia.org/wiki/Write_amplification)

      • just brew it!
      • 5 years ago

      Actually, they’d probably be OK for that. All the pain is on write, and the pain is mitigated for large streaming writes (since a bunch of the stuff that would need to get rewritten is getting overwritten anyway).

      I doubt the 25% gain in capacity is worth it though, unless you’ve got one helluva huge media collection.

    • Voldenuit
    • 5 years ago

    Seagate + “Archive” = LOL.

      • chuckula
      • 5 years ago

      Haven’t you heard of WOM (write-only media) technology?

      • Thrashdog
      • 5 years ago

      Great for archiving those sensitive legal documents. Meet statute of repose requirements without ever actually being able to recover anything!

      • Waco
      • 5 years ago

      You + “Funny” = not found

      Seriously, Seagate drives are perfectly reliable compared to their peers minus a VERY small range of drives that had a high failure rate a few years ago.

        • Deanjo
        • 5 years ago

        Actually if you look at Backblaze’s data, even the newer Seagates are less reliable than any other brand. (3.8% vs 3.2% of a much older WD Green 1TB drive)

        The hard data just doesn’t back up your claim.

          • Takeshi7
          • 5 years ago

          And yet Backblaze still prefers Seagate drives over every brand, and they buy more Seagate drives than all other brands combined.

            • Deanjo
            • 5 years ago

            Because they are cheap to replace and readily available. Their preference isn’t Seagate either. And I quote

            [quote<]Hitachi Drives - If the price were right, we would be buying nothing but Hitachi drives. They have been rock solid, and have had a remarkably low failure rate."[/quote<] Their study also show that those drives from seagate that had a low failure rate started failing at a far greater rate when compared to their competitors as time goes one to the point where 1 out of every 4 will fail within 3 years.

      • stdRaichu
      • 5 years ago

      Heritage ale vocals
      Veto glacial hearse

      Voles cheat regalia
      Slather I/O cleavage

      Love a lethargic sea
      Retail salvage echo

      A cavalier log sheet
      Alleviates He cargo

      Savage ethical role
      Reallocate; give ash.

        • chuckula
        • 5 years ago

        [quote<]Slather I/O cleavage[/quote<] BEST BAND NAME EVAR!

    • albundy
    • 5 years ago

    Looks painful and itchy. Cant wait to get one!

      • Laykun
      • 5 years ago

      Yeah my thumb still hurts 7 months after my experience with shingles (was originally my whole arm).

    • Aliasundercover
    • 5 years ago

    Shingled recording sounds like a lot of pain for 25% capacity. Do you need a log structured filesystem? Or does it have special non-shingled areas for metadata? Even if your files are huge, written once and in sequence the filesystem needs directories and allocation maps.

    It takes a shocking amount of time and work before a new filesystem becomes stable. Unless Seagate has lost its last marble the drive must be intended to work with existing filesystems. It probably acts like it has an enormous sector size where changing any part 16MB means read modify write cycles. It would be slow but at least it would work.

    Or maybe it has a SSD style mapping layer inside, essentially a baby filesystem captive to the drive.

    All for 25%. You better have a sea of rarely needed data to be worth it.

      • jihadjoe
      • 5 years ago

      Rather interestingly there have been a [url=http://storageconference.us/2014/Papers/06.HiSMRfs.pdf<]couple[/url<] of [url=http://www.pdl.cmu.edu/PDL-FTP/FS/CMU-PDL-12-105.pdf<]papers[/url<] on just how to write filesystem metadata on SMR HDDs. Some call for an unshingled partition, others even for an SSD layer. Seagate has some experience making hybrid drives, so that's well within their areas of expertise. Hard to know how they're doing things with these drives without more detail, but considering it's being marketed as an "archive" drive maybe they're willing to accept the performance tradeoffs of re-writing tracks adjacent to metadata for these archive series drives. General-use, or higher performance variants will probably make use of an unshingled partition, or an SSD layer.

      • GTVic
      • 5 years ago

      Zones.

      • w76
      • 5 years ago

      [quote]All for 25%. You better have a sea of rarely needed data to be worth it.[quote]

      Datacenters for companies like Facebook and many others likely do have petabyte oceans of rarely used data. I don’t know, 25% is a pretty worthwhile iteration, I’d sure as hell like to boost my pay 25%, but at what cost? That’s the question, and there’s no solid answers yet. Might not be that bad.

      • Peldor
      • 5 years ago

      Summary of the hard drive industry for 30+ years: This new technique sounds like a lot of pain for 25% capacity.

        • meerkt
        • 5 years ago

        Yeah, but this specific one is more of a sidegrade rather than the usual upgrades.

        • just brew it!
        • 5 years ago

        At least previous advances tended to increase capacity *and* performance, since capacity increases have generally been tied to increased linear bit density. This one increases capacity, but will decrease performance by an amount that likely ranges from minimal to “OMG that’s horrible!” depending on the pattern of writes.

      • just brew it!
      • 5 years ago

      Yeah, this is clearly aimed primarily at near-line and archival storage applications. It essentially fits into the gap between normal HDDs and tape.

    • The Egg
    • 5 years ago

    Shingles, huh? Does this mean the drives are gonna explode like confetti when they crash?

    • chuckula
    • 5 years ago

    [quote<]The Archive HDDs are rated for 180TB of writes per year,[/quote<] In SSD world, the write limits are there due to flash wearing out. In the HDD world, the write limits are there because it takes a whole year to write out that much data.

      • UnfriendlyFire
      • 5 years ago

      And by that time, the motor/bearings or something else might wear out.

        • Ninjitsu
        • 5 years ago

        Yeah, I always thought the TB/year limits are mechanical tolerances.

      • Waco
      • 5 years ago

      (180 terabytes) / (365 days) =
      5.70776256 MBps

      That’s actually not a crazy estimate.

Pin It on Pinterest

Share This