back to article Western Digital: And when I pull the covers off, behold as NAND becomes virtual DRAM

Western Digital has been using software to turn SSDs into virtual memory so applications are accelerated without having to deploy DRAM or be constrained by memory capacity. The firm said the Ultrastar DC ME200 is an "optimised" Ultrastar SN200 SSD, which uses planar 15nm MLC (2bits/cell) NAND and is integrated with third-party …

  1. jms222

    All obsolete

    when the non-volatile DDR5 variants appear. Then memory volatility really will become fuzzy.

    SATA common though it has become is a terrible interface. A parallel bus protocol moved to serial forgetting that it should have become full-duplex at the same time. In contrast with SAS, PCIe and friends.

    (Yes SATA is electrically full-duplex but the protocol forgets this.)

  2. Lee D Silver badge

    Ah, the old "let's run a swap-file on a RAMDisk" trick of old.

    Honestly the two technologies really need to just merge so that you're "reserving" a space for the swap-file is really just "use the first 16Gb of the storage for main RAM" on a device where you can treat everything equally.

    We need to just replace "RAM" and "disk" with "memory" and use it for everything.

    Wanna run a server? Run several chips of memory in a RAID and thus have them verify each other (ECC, effectively).

    Upgrade the disk, you upgrade the RAM. No suspend/hibernate junk. No "I ran out of RAM so I'm just going to fall over and crash". And "persistent RAM".

    We need to get rid of hard disks, which will make everyone focus on SSDs and Flash, which means we could easily get their longevity / sacrificial sectors to the point where they compete with RAM, and they are already fast enough that you need a direct-bus connection to get the most out of them - and creeping their way towards RAM speeds all the time (I believe they are in the DDR2 ranges for speed already).

    1. h4rm0ny

      Well that's a great idea. Is it easy?

  3. binary
    Facepalm

    Is two really three?

    "Western Digital is consolidating its brand names into two classes: consumer, and enterprise and commercial."

    Now I am really confused...?

    1. MiguelC Silver badge

      Re: Is two really three?

      Why do you think there's a comma between "consumer" and "and enterprise and commercial"?

    2. Steve Knox
      Headmaster

      Re: Is two really three?

      "Western Digital is consolidating its brand names into two classes: consumer, and enterprise and commercial."

      Now I am really confused...?

      You're clearly not from Oxford.

    3. diodesign (Written by Reg staff) Silver badge

      Re: Is two really three?

      Yes, two. Consumer, and enterprise and commercial.

      C.

  4. vtcodger Silver badge

    sounds nifty but complicated

    Anybody want to bet that in about four years we'll be frantically loading patches to get around the then recently discovered MEMORYTRICK exploit that allows anybody within six blocks to extract your passwords, encryption keys, PINS and pornography from your virtual memory using nothing more than a crystal set and a sharpened swizzle stick?

  5. jon909

    Ooooh yes, the oxford comma. I believe there's a funny sketch by Mitchell and Webb on that old chesnut.

  6. Christian Berger

    So... how is this different to swap on an SSD?

    I mean it really sounds like that if you filter out all the marketing speak.

    1. Munchausen's proxy

      Re: So... how is this different to swap on an SSD?

      " So... how is this different to swap on an SSD?"

      It wasn't entirely clear to me, but it sounds like they've built in a good paging system so they can bypass Linux's house of cards, and make actual good use of the backing store.

      1. Alan Brown Silver badge

        Re: So... how is this different to swap on an SSD?

        "so they can bypass Linux's house of cards"

        The house of cards is built that way because it has to handle everything that's thrown at it. Swap has been a standardised interface on *nixen for nearly 40 years and is well-overdue for heavy maintenance.

        The reality is that "swap" as we know it has been a dead duck for a number of years (if you look at most non-choked systems you'll find they've pushed trivial amounts out to swap. Ram really is staggeringly cheap these days) and a system which "needs" more than 1-2GB of swap is a rarity, because the effect of hitting any form of drive interface in anger makes your machine leap back to the 1980s speed-wise(**), which in turn makes justifying more ram an easy decision. 32GB in a desktop is cheap. 64GB only slightly less so. Putting 512GB-2TB in a server isn't mortgage material anymore either

        There's no real secret sauce here and quite frankly the way WD & Seagate have treated buyers over the last decade-and-a-bit means they'll have to try a lot harder than this to convince me to buy their solid state products - especially when they're pushing it in a 12nm format that everyone else backed away from due to it having limited endurance and poor speeds (Hint:There are only so many electrons you can fit in a 12nm cell and it's getting towards "you can count them individually")

        (*)except for the ones which hibernate to the swap partition, but that's not quite the same thing as actively banging on it)

        (**) It's not helped by more than a few "really fast, honest guv" NVME drives being flogged by name brands who should know better *ahem*HP*ahem* turning out to have write speeds on par with (actually, worse than) 2011-era consumer sata-SSD drives and only slightly better than WD's old Raptors.

  7. EveryTime

    It really does sound as if this is swap-to-SSD, but done in a bare metal hypervisor.

    So the real product is a hidden bare metal hypervisor that exists solely to manage a hidden swap drive.

    Hypervisors look easy from the outside, but they are very tricky to do correctly. That's a lot of complexity for a single feature.

    1. eldakka Silver badge

      That's a lot of complexity for a single feature.

      But not for a marketing gimmick!

  8. Anonymous Coward
    Anonymous Coward

    Almost all computing in the abstract...

    ...is merely a matter of cache management.

  9. Cynic_999 Silver badge

    What am I missing?

    Surely having your page file located on an SSD would achieve pretty much exactly the same thing?

  10. takyon
    Paris Hilton

    hold up

    Is this like Samsung's Z-NAND SSDs but worse?

  11. Kevin McMurtrie Silver badge

    Good for multi-media too

    When you upload a photo or video to web sites, you'll notice that there's a long delay until that media is available even if the upload took a very long time. It was saved to disk them batch processed. Given enough memory, you can process most multimedia as it arrives. I did experiments with this over a decade ago where you could upload a video and, as long as it was a streaming format, it would be available in multiple formats immediately when the upload finished. The downside was that video compression eats a ton of memory. An army of midnight dialup users would easily deplete RAM before CPU usage was high enough for the load balancer to notice.

    A few GB of fast RAM and a few hundred GB of semi-fast RAM would solve it.

  12. ChiefDS

    Amdahl's law has been updated with Gustafson's Law

    Gustafson a.k.a. Gustafson–Barsis's law, updated Amdahl's law in 1988. It's a much better approximation of the theoretical speedup of latency for a given task execution when the system resources have been improved.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2020