back to article How HCI simplifies the data center

Organizations look at the cloud as an option because it was impossible to get their data centers to operate at the same efficiency. Now if you flip it over and get the data centers to behave the same way as the cloud, how would that change things? Companies are already on board with server consolidation and virtualization – …

COMMENTS

This topic is closed for new posts.
  1. Nightkiller

    There you have it in a nutshell. The cloud is based on the premise of a bank. The bank lends out depositor's money to others on the notion that not all of the depositors need access to their deposits at the same time. If there is a run on the bank (sorry we're out of capacity at the moment), the bank will have an auction to see who gets access to their own money by paying MORE.

    What a delightful idea. IT dealing with the banks.

    1. Anonymous Coward
      Anonymous Coward

      Public cloud infrastructure is more like a bank. Where you have thousands of customers who will not all need all of their resources at the same time due to different seasonal peaks, different quarter ends, etc. That allows people to add resources, take out short term resource loans, completely close their accounts, etc on the fly.

      On prem IT really isn't like that, HCI or no HCI, because you still have to scale to peak. If Black Friday requires some retailer to have 20,000 cores, for instance, they need to buy servers, network, data center space, etc for the full peak load. Even if their average load is 10,000 cores and they really only need the peak for one or two days. They can't just 'borrow' 10,000 cores for that one day or few days and then return the resources when they are no longer required and stop paying for them... as you would if you needed a short term loan from a bank.

      1. Trevor_Pott Gold badge

        "On prem IT really isn't like that, HCI or no HCI, because you still have to scale to peak"

        Bull.

        You only have to keep your critical workloads on-prem. The stuff that is actually burstable is likely not mission-critical (it will just run slower if there aren't 'enough' instances) and you can farm that out. To the public cloud or to a service provider. Use the right tool for the job. The world isn't black and white/one or the other.

        Use the public cloud only for what it's good at: providing non-mission-critical capacity when you are over peak. It's 2017. We can do this stuff now without having to throw out our ability to sweat our assets through bust times, have control over our own data, or run sensitive workloads in our own legal jurisdictions.

        Hybrid IT isn't just some buzzword. It's not even some ideal towards which we are striving in the distant future. It's a thing we do today. Some things on prem, some things in the cloud. It's not rocket surgery. It's just some bloody YAML.

  2. baspax

    I am sure you have checked out Turbonomic. It takes care of on prom and cloud overprovisioning

    1. Throatwarbler Mangrove Silver badge
      Meh

      . . . when it works. I have definitely experienced some issues with it behaving unexpectedly or unintuitively.

      1. baspax

        Interesting. Would you care to elaborate? I am genuinely interested as this is the first time I heard about it behaving weirdly

    2. Trevor_Pott Gold badge

      Sort of. Turbonomic has its limitations, and is itself limited by what the management software will allow. You cannot do proper overprovisoning of CPUs, for example, when you are forced by the cloud implementation to dedicated cores when creating a tenant.

  3. Anonymous Coward
    Anonymous Coward

    "to free up nerds to focus on application-level issues infrastructure"

    So why not just use public cloud? If the idea of HCI and related 'private cloud' tech is to automate away all the simple, low level admin... then what does it benefit you to keep owning the real estate, boxes, and giant air conditioners?... and on prem is just inherently never going to give you the flexibility of public cloud, e.g. run a thousand cores for 10 minutes and then hand them back.

    1. Trevor_Pott Gold badge

      Cost. Security. Privacy. Regulatory compliance.

      Not everyone gives no fucks about the cost of things, is bamboozled into thinking the opex model is great, or lives in the US. There are lots and lots of reason not to use the public cloud.

      There are lots of reasons you should use the public cloud.

      It's what you want to do that determines where a workload is best run. Not some sort of religious belief.

      1. Anonymous Coward
        Anonymous Coward

        I have not seen a case in which public cloud has not been less costly than on premise... when the cost compare is done properly. People often take the cost they are paying per TB for storage from NetApp or whoever, for instance, and compare it to the cost per TB for cloud storage over a three year refresh period and in some cases the on prem cost is lower (in some cases not)... but rarely do they factor in data center floor space (either rented or the amortized cost of the data center tiles and everything that goes into the data center, UPSs, generators, AC/chillers, all of the cabling, fire suppression, security systems, electricity to power the data center, etc, etc). The cost of networking. The cost of storage management software and automation software. When you add all that up, public cloud is less.... The largest cost items are things like scaling to actual utilization vs peak. Cloud allows you to scale to actual utilization, on prem requires peak scaling... scaling to peak will require a company to buy way more hardware than they need for most of the year. You can auto scale with public cloud which allows you not to pay for infrastructure until the minute you are using it... with on prem, most companies buy out infrastructure years in advance of needing it or, at the very least, months.

        Also, until recently, public cloud was dominated by AWS. AWS didn't have a huge incentive to dramatically lower costs. Now Google and Microsoft are legit competitors... and the three of them are constantly undercutting each other price. The prices will continue to drop two or three times a year for some time now that there is a competitive market.

        Security, privacy, regulatory compliance - All the major cloud providers have security and data protection measures, operations and skills that are far beyond what an average company could hire or build. It is just an advantage of operating at hyper-scale. All of the major cloud providers have every security audit and regulatory compliance certification under the sun.

        1. Trevor_Pott Gold badge

          Members of the local cult "haven't seen a case" where their forced religious indoctrination of children causes lifelong trauma either. Despite this, escapees spent the rest of their lives in therapy.

          I have yet to see more than edge cases where the public cloud actually is less expensive than on premises. Then again, I'm not a believer. I'm just someone who uses spreadsheets a lot.

          1. Anonymous Coward
            Anonymous Coward

            Well, agree to disagree on that point.

            Economics 101 lets you know that cloud will be less costly and better than on prem. Cloud providers have huge economies of scale. They buy processors, boards, cabling, etc by the thousand count, operating millions of VMs or containers. They are going to have much lower component purchasing prices that smaller companies. Eventually, as Google has done, they are going to remove the middle men and build all their own tech... no Cisco, VMware, MSFT, NetApp, Oracle, EMC, etc taxes or 500% mark ups. Likewise with data centers. It is less costly to build and operate one gigantic data center than building 100 smaller data centers, even if the sizes are the same when you add up 100 smaller data centers. When you are operating at that scale it makes sense to invest in PhDs to build automation technology that just is not accessible to the smaller operators who cannot afford to have an academic computer science department on staff. There are huge cooperative benefits too. As different businesses require different levels of resources at different times, sharing resources allows you to things like not have to scale to peak.

            Everyone knows this intuitively. If I were to ask you if it would be less costly to purchase network services from an ISP like Verizon or bury your own fiber lines underneath the ground/oceans around the world, everyone would say that obviously it is going to be a lot more cost effective to just rent lines from the ISP where many people share the cost burden than to pay billions to create your own global network.... That is exactly the same situation as cloud vs on prem data centers. The difference is that everyone has been in the "network cloud" from Verizon or Orange from the outset of the IT industry, so it doesn't seem strange... it seems like the way things obviously have to work.

            1. Trevor_Pott Gold badge

              "Economics 101 lets you know that cloud will be less costly and better than on prem"

              Let me guess, you also believe - despite mountains of evidence - that trickle down economics works.

              *sigh*

              And the earth is only 4000 years old. The eye is irreducible complexity. *thud* *thud* *thud* *thud*

              "Everyone knows this intuitively. If I were to ask you if it would be less costly to purchase network services from an ISP like Verizon or bury your own fiber lines underneath the ground/oceans around the world, everyone would say that obviously it is going to be a lot more cost effective to just rent lines from the ISP where many people share the cost burden than to pay billions to create your own global network...."

              Um...you're demonstrably wrong. Massively, demonstrably wrong.

              First off, it's cheaper to build your own global network the instant the cost of laying your own fibre gets to about 1/20th the cost of renting it. Right about there you can go lay multiple strands of cable, use whatever capacity you need and rent out the rest.

              You know, like Google does. Sure wasn't cheaper to just stand up their own datacenters and pay the rent to the ISP. Nope, they laid their own fibre. And yes, they even have a stake in oceanic cable.

              I can introduce you to municipalities that also lay fibre for everything from last mile to backhaul. I can introduce you to WISPs and even businesses as small as 10 people who would rather pay the municipal fees to dig a ditch to lay fibre between their location and the local internet exchange than to pay the ISP. Shock, horror...it turned out to be significantly cheaper.

              In some cases the ISP is cheaper. In many others it's not. Just like in some cases (you know, those very rare niches where you have "could native" burstable workloads) the cloud is cheaper. In many others (such as 24/7 workloads), it's not.

              As with everything in IT it depends. You do a needs assessment and you use the right tool - technical and economical - for the job. You don't decide on the tool and then contort all reasoning beyond logic in order to fit what you do to that tool.

              Also - and I don't understand why I have to keep repeating this to someone supposedly so smart - there are huge differences between regulated industries (like telcos) and completely unregulated ones (like public cloud services). What my governments impose on telcos here as minimum service quality, pricing caps and more keeps monopolistc behavior in check. There is absolutely nothing keeping monopolistic behavior in check amongst the cartel of public cloud providers.

              Also - and again, I can't understand why this is so hard for you to get - when you do the actual numbers on running your own workloads you don't have to be running that many workloads 24/7 before rolling your own is significantly cheaper than public cloud.

              Religion. All you're touting here is religion. It is no different than trickle down economics or praying the gay away.

  4. Stuart Dole
    Pint

    At first I thought it was "HCl", not "HCI"

    Just dissolve your troubles with a nice low pH...

    (A pint for other ways to dissolve your troubles...)

    1. J. Cook Silver badge
      Coat

      Re: At first I thought it was "HCl", not "HCI"

      That reminds me of a little ditty:

      Charle was a chemist,

      But he is no more.

      He drank what he thought was H2O

      but was H2SO4.

      Mine's the lab coat with the 'bad chemistry puns' book in the pocket.

  5. Anonymous Coward
    Anonymous Coward

    "Customers don't know what they need"

    IMHO the most important sentence in the whole piece and spot on (nice piece, BTW, Trevor). But I'd go further: sometimes they simply have no idea how much they need either.

    At the moment, we're developing something completely new. Privacy and security are EXTREMELY important in what we do, so going public cloud is quite simply not an option, but because what we do is entirely new there are also not yet any metrics we can rely on for planning. We have some idea of storage needs, but CPUs, threads, memory is something we have taken a stab at but we know that when stab meets reality they may not exactly match.

    So we've partnered up with a not-quite-public cloudy setup. We've worked with this company before (we screened them for some clients and then got friendly, which naturally buggered up any further screening revenue due to conflict of interest :) ), and they will build and run the IBM Power based infrastructure for us in which we spin up whatever VMs we need.

    Their client base means they run a nice hardware margin, so we start in their spares rack which is easy to isolate when we hit the numbers that turn what we do into a sustainable business (which is around 500 clients). They own the kit, run the hardware, comms and maintain the virtualisation and VM backups (two site model with 90km separation and three separate comms providers), we do the rest.

    We know what we need, we just don't quite have an idea of how much of it :).

This topic is closed for new posts.