The Register Home Page

back to article If hardware price squeezes make you sad, VMware says it will all end in tiers

VMware is working to find new roles for its memory tiering technology, in the expectation that memory and hardware prices will remain high for years and users will therefore look for ways to put off server purchases. Memory tiering moves data out of DRAM and onto another storage medium that’s fast enough that CPUs can still …

  1. AtomicDog

    Irrelevant?

    ... because nobody's dumb enough to further tie themselves into Broadcom's VMware shakedown racket.

    1. Roj Blake

      Re: Irrelevant?

      Given the choice between inflated hardware prices and Broadcom's egregious price-gouging I know which I'd choose.

      Of course, Broadcom won't be the only people with this sort of product out there...

      1. Sandtitz Silver badge

        Re: Irrelevant?

        "Of course, Broadcom won't be the only people with this sort of product out there..."

        Which other hypervisors have this kind of memory tiering?

        1. AtomicDog

          Re: Irrelevant?

          Even if there aren't any now, there will be within a matter of months...

          Also, you buy memory, that's a capital expense, it sits on a company's balance sheet, it's an asset, you write nominal depreciation off against tax over time, you never pay any more for it.

          You buy VMware products in this day and age, you pay every year, more and more every year, at Broadcom's whim. You might as well write them a blank cheque. And it can't possibly perform as well as if you'd just bought the extra RAM in the first place.

          If you're going down the scary opex route you might as well chuck your stuff in the cloud.

          Or, god forbid, optimise your code so you don't need so much memory in the first place. We've all become so lazy and wasteful...

          1. Sandtitz Silver badge

            Re: Irrelevant?

            "Even if there aren't any now, there will be within a matter of months..."

            Nonsense.

            "You buy VMware products in this day and age, you pay every year, more and more every year, at Broadcom's whim."

            Now you are just lying and spreading FUD. You can pay for 3- and 5-year terms.

            Look, I know everyone hates Broadcom, but there's no need to invent stuff.

            "And it can't possibly perform as well as if you'd just bought the extra RAM in the first place."

            No-one is claiming so.

            Memory prices have gone up 10-fold during last year - 1TB server memory can cost $50k or more today. And that's a single server. With this memory tiering you could go for half the memory and top if with a 512GB NVMe drive which costs a fraction of the memory price.

            If you don't have to worry about budgets, go for RAM only. Many others do not have that luxury.

            1. AtomicDog

              Re: Irrelevant?

              "Nonsense"

              If it's that a useful feature why on earth wouldn't it be added to KVM/QEMU sharpish? Who are you to dismiss something so out-of-hand?

              "3/5 year terms"

              Oh goody, more time for you to get thoroughly addicted/over-reliant on the technology so that Broadcom can REALLY make you squeal when it comes to renewals. Like they already have done with many of their customers. I'm so sorry for not quoting Broadcom's terms verbatim, but the bloody principle still stands.

              "1TB server memory"

              Seriously what the hell are you actually doing that needs you to CapEx 1TB of RAM, yet you can't afford it?

              I've worked for plenty of clients and companies that have ended up with more than that amount of RAM in play, but it's always been through incremental expansion. Anything more "spiky" has been done in the cloud to prove the concept first before justifying massive CapEx.

              1. Sandtitz Silver badge

                Re: Irrelevant?

                "If it's that a useful feature why on earth wouldn't it be added to KVM/QEMU sharpish? Who are you to dismiss something so out-of-hand?"

                I don't see any talk about devs implementing this feature on KVM or QEMU. Implementing this isn't just adding an enable_nvme_tiering setting somewhere but will take a lot of development and testing resources.

                If the KVM developers find this a useful feature, they might implement it. But talking about months is pure fantasy.

                "Oh goody, more time for you to get thoroughly addicted/over-reliant on the technology so that Broadcom can REALLY make you squeal when it comes to renewals.

                People have been reliant on VMware for couple of decades because it has been the best-in-class. And it still is. Moving from VMware to any other is a downgrade in features, ecosystem, maturity.

                If you rely on Nutanix, HPE, IBM/RedHat or other paid virtualisation tech, they are equally looking to extract all loose money from their clients as well. Sorry to tell you this.

                Moving to any other hypervisor environment is always a chore, why would it be harder in the future? Some of VMware competitions is making the move easier these days.

                "Seriously what the hell are you actually doing that needs you to CapEx 1TB of RAM, yet you can't afford it?"

                That was for a single server. Times 10 or 100 for some companies. Even the bigger companies balk at some point. If you can save thousands or millions with memory tiering - just like you can/could with storage tiering - with minor performance loss, why would you NOT do it?

                "I've worked for plenty of clients and companies that have ended up with more than that amount of RAM in play, but it's always been through incremental expansion."

                1TB is peanuts where I work. That was an example of how hardware refreshes these days are very expensive compared to year ago.

                "Anything more "spiky" has been done in the cloud to prove the concept first before justifying massive CapEx."

                Can't work with cloud because of latencies, regulation and other reasons.

        2. K555

          Re: Irrelevant?

          Not such a precise science, but you get this behaviour on KVM with vm.swappiness set high. One ZRAM device with LZ4 tier1 and ZSTD tier2 mounted as swap at highest priority and then NVMe mounted as swap next on the list.

        3. Roland6 Silver badge

          Re: Irrelevant?

          IBM mainframe…

          Whilst VMware was very good, it was (circa 2002) inferior to mainframe hypervisors, although its UI was significantly better…

          Which begs the question why did it take until 2025 for VMware to implement those obvious memory management functionality.

          1. Sandtitz Silver badge

            Re: Irrelevant?

            "Which begs the question why did it take until 2025 for VMware to implement those obvious memory management functionality."

            You can always ask the same question why the VMware competition has been so slow to gain features already in VMware.

  2. DrewPH Silver badge
    Facepalm

    So when I run out of memory (which is now unaffordable) it will hive data off to NVMe drives, which are now also unaffordable. Got it.

    1. Anonymous Coward
      Anonymous Coward

      I was surprised when I found out that the AI boom had caused the price of even old-school "spinning rust" hard drives to spike, as I'd have thought they'd have been too slow to be of direct relevance there.

      At this rate, I'll be expecting to retire on what OpenAI pay me for those old data cassette recorders and tapes stored alongside my Atari 800XL...

  3. Fido

    What's old is new

    Since a long time one can overprovision RAM between Linux KVMs and it will swap cold pages as needed. I expect any virtualization solution built on Linux would include this same capability and if it didn't, could easily be modified to do so in the future.

    Long before that IBM discovered what happens when virtual machines with virtual memory start thrashing.

  4. Anonymous Coward
    Anonymous Coward

    I see BroadCON have now pulled (without any word) public access to their VDDK!!!!! Which just about every migration tool uses to migrate from vSphere, Nutanix MOVE as an example. Utter housery thank FECK we completed our migration from VMWare over a year ago. This nonsense is going to be a BIG issue for lots of businesses! https://platform9.com/blog/vddk-no-longer-available/

  5. Anonymous Coward
    Anonymous Coward

    I’m pretty sure…

    …that’s a description of swap memory.

    I must be getting old.

    1. Roland6 Silver badge

      Re: I’m pretty sure…

      Remember the US patent rule:

      Swap memory in a hypervisor

      A hypervisor is not an OS and VMware isn’t for IBM mainframes (prior art) but x86 platforms and thus this is a new and patentable invention.

    2. munnoch Silver badge

      Re: I’m pretty sure…

      That's what I thought, silly me...

      Although as I understood it, swapping is when entire processes are kicked out to secondary storage and doesn't require virtual memory support. Paging is when individual pages are moved in and out which does.

  6. Greedy_Soap

    Anyone who is not actively migrating away deserves the worst, no one should ever trust Broadcom stuff, these scumbags should be jailed!

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