Sign in / up

back to article Yes, you heard me – the storage infrastructure WARS are over

Though there are still a great many players on the field fighting savagely for the right to dominate the industry for the next decade, I believe the storage infrastructure wars are already largely over. With so many startups still entering the storage space, and so much money flowing around it seems like I'd be mad to do so, but …

COMMENTS

Post your comment
House rules Send corrections
Add to 'My topics' unstar *
  1. dan1980

    "Over the past year I have personally talked with either the virtualisation teams, the storage teams or both at over four hundred of them."

    That's impressive. Amazing that you find time to write or indeed engage in gainful employment with all conferences you must attend.

    If only you put half as much effort into your school work as you do socialising with your friends then maybe you wouldn't be failing History!

    Oh wait, that was me. No, sorry; carry on.

    5 0 Reply
  2. Platypus

    Server SANs aren't going to be the answer unless/until they deal with the issue of server-resident storage being lost on server failure. As soon as you start replicating to avoid that, you're in the same territory as the existing scale-out and "hyper-converged" vendors which can implement the exact same data flow to/from the exact same devices. (Disclaimer: I work on GlusterFS, which is in this category.) If all you need is the speed of local storage without availability, you don't need a server SAN; you just need plain old local storage managed however you see fit. If all you need is availability without the speed, you're back to traditional SAN or NAS. The whole point is that sometimes people need both, and server SANs are hardly alone at that intersection. In fact they're the new arrivals struggling to piece together a real story.

    TBH, I think "server SAN" is just a marketing term for something that was already possible (and often done) technically. Maybe that marketing allows the virt team to take ownership instead of working with people who actually understand storage, but I guarantee that will end in tears when data gets lost. Server SANs are the "peace in our time" of the storage-infrastructure wars.

    4 1 Reply
    1. Trevor_Pott Gold badge
      Reply Icon

      Server SANs are hyperconverged. Lash together storage from multiple servers, present it to the cluster as shared storage. Or, more to the point, hyperconvergence is one of the possible means by which a serverSAN can manifest.

      Server SANs can be done without running a compute workload on the same node. Then it's not hyperconverged. Run a compute workload om the node and it is hyperconverged.

      Marketing terminology. It are the dumbs.

      0 0 Reply
      1. Trevor_Pott Gold badge
        Reply Icon

        Compute workload om? On you miserable, non-tactile touchscreen craptasm! Why isn't there an edit button in the mobile UI?

        :(

        0 0 Reply
      2. Platypus
        Reply Icon

        Yes, marketing terminology is dumb ... and that's all "server SAN" is. "Lash together storage from multiple servers and present it to the cluster" is a concept that existed for quite a while. Why claim victory for "server SAN" instead of the broader category, except as a marketing move?

        0 0 Reply
        1. Trevor_Pott Gold badge
          Reply Icon

          "Why claim victory for "server SAN" instead of the broader category, except as a marketing move?"

          A) "Server SAN" was coined by Stuart Miniman. An analyst, not a marketing bloke, because we needed something to call "lashing together storage from multiple servers and present it to the cluster" that was shorter than "lashing together storage from multiple servers and present it to the cluster"

          B) Because not all "clusters of storage lashed together" are the same. Object storage is, for example, going it's own say, despite being something we could reasonably call a "server SAN" as a technicality. The big money isn't in object storage. You can't charge the big margins for it. It's things like VSAN, Nutanix, etc that are wining out and going to form the "default" for enterprise workloads.

          C) I calls it like I sees it. Arrays, clusters of arrays and "lashed together clusters of storage that don't run workloads on the same nodes as the storage" are simply not winning out over more modern "hyperconverged" (and by, do I loathe that term) setups.

          And it's enterprise workloads that matter, mate. They're where the money is. They're where you get the margin.

          1 0 Reply
          1. dan1980
            Reply Icon

            @Trevor_Pott

            "I calls it like I sees it."

            Right, so now you are dividing your time between attending industry events and your side project as a whale biologist.

            I have added nothing useful to this conversation.

            0 0 Reply
            1. Trevor_Pott Gold badge
              Reply Icon

              Did you know that the closest land-based relative to cetaceans (a family that includes whales) is the hippopotamus?

              1 0 Reply
              1. dan1980
                Reply Icon

                The hippopotamus is a semi-aquatic, noble, yet unattractive mammal that tends to have a rather detrimental effect of the environment. Just like Christopher Monckton.

                0 0 Reply
          2. Terafirma-NZ
            Reply Icon

            I don't see how this model really fits the enterprise world very well. Cloud providers that spin up 1000's of the same 3 templates sure but some SQL server with 60 - 100 TB of database files behind it equates to many millions of $ of Nutanix/vSAN nodes to run a single VM when a AFA would be far better suited and lower cost.

            A simple design of Pods would make more sense and this can be seen when you look at the new Facebook design they call them clusters. Each cluster can be made up differently be it blade with SAN or hyper-converged what is still to be defined in my view is applications being extended to make use of the different types. (Think something like SharePoint supporting moving content form old sites to object storage for long term retention but still searchable.)

            As well as some form of abstraction over them to keep it all together and this is where I see Openstack pushing hard.

            Far too often the human part of us wants every rack in every row to be the same but its just not economic in technology with such a high change rate. Even if you did how will new and old Server SAN storage cope with new and old models Nutanix/vSAN has not yet been around long enough to show what happens when you have 40% of your cluster that is 10 x faster than the rest, does it all slow down to the lowest denominator thus making expansion pointless and rip and replace becomes attractive again or do you get data segmentation.

            1 0 Reply
            1. Trevor_Pott Gold badge
              Reply Icon

              I...I don't think you understand how these work? Especially how they're installed in enterprises.

              As a general rule, enterprises used to buy hardware dedicated to a specific project or workload. Each DB had it's own SAN, it's own servers, etc. But then we found that this was spectacularly inefficient and led to massive underutilisation of resources. Private clouds - or at least virtualisation setups that were closeish - began to become the order of the day.

              Resources began to be purchased and pooled based on cumulative predicted need, not based on the individual project or workload. Now the question has become "how best to maintain these sorts of environments."

              Something like a Nutanix or VSAN cluster rarely goes beyond 16 nodes, sometimes to 32. You get multiple clusters in a virtual datacenter. You are highly unlikely to have nodes in the cluster that are different speeds/capabilities because clusters tend to live and die as a group. We've seen that even in non-VSAN clusters thus far. Clusters are born, they live and they die as one.

              But in the rare instance where clustered are mixed - I run a mixed cluster myself - sysadmins can simply tell workloads to keep the copies on "like" nodes. If you have PCI-E storage on nodes A-D and only SAS storage on nodes E-H, then you can "segregate" the cluster into two.

              In theory, you could end up with a workload split along the storage plane, but only if you'd lost enough of one type of node that rebuilding would cause it to put the second copy on the other class of node. As soon as you've repaired the server sin question, policies would take over and make sure your workloads go where they are supposed to.

              If your assertion is somehow that server SANs are unable to support SQL or OLTP workloads, well...you're just wrong. You're wronger than wrong.

              Believe it or not, server SANs have been around long enough to evolve to handle the concept of diversity in workloads...and to handle workloads that are as demanding as anything you could throw at a traditional SAN. Indeed, I'd challenge a traditional SAN to keep up with the all-flash server SANs. The MCS setups in particular are utterly spectacular.

              0 0 Reply
    2. jedangerous
      Reply Icon

      Preach on!

      Completely agreed. I wrote a small article discussing a similar topic ;) http://jedangerous.blogspot.ca/2014/08/thoughts-on-storage-in-enterprise.html

      0 0 Reply
  3. pyite

    Half hour boot times

    You said "we wouldn't tolerate a system that took a half an hour to boot in today's world."

    I wish this were the case, but server POST times are now the biggest headache in all of IT, with practically every server taking 5 minutes and many are over 10 when loaded with memory. I have to deal with new hardware from all major vendors, sometimes requiring many reboots to get things just right. On my last big project, a majority of my "working" time was spent staring at POST screens. SuperMicro is about the only vendor who seems to be resisting this disturbing trend, but only because they suck a little less than the others.

    It is time for IT customers of the world to unite against this nightmare. The problem actually comes down to the fact that people working on the servers are typically not the same people making the purchasing decisions.

    Somewhat of a tangent, I'll admit, but this particular premise is clearly wrong.

    3 0 Reply
    1. Trevor_Pott Gold badge
      Reply Icon

      Re: Half hour boot times

      5-10 minutes is a long way from a half hour. Also: for the record, you don't tend to need such long boot times in server SAN boxes, because you need RAID cards so complicated that they need to load an OS from the future when they init.

      Of course, you could also be seeing extended boot times because you're using 1.5TB of RAM and doing extended memory tests on each boot. Dell in particular seems to like really indulgent mem tests.

      0 0 Reply
  4. Riku

    Mainframe zombies?

    Forgive me, but it sure does feel like virtualisation, cloud apps/storage are merely LPARs, thin-clients and bureau computing dressed up in PFY app-isim-ness.

    From the old greybeards to the new beardy hipsters. Everything old is new again.

    0 0 Reply
    1. Trevor_Pott Gold badge
      Reply Icon

      Re: Mainframe zombies?

      The mainframe is ultimately what everyone wants, but IBM refuses to price it reasonably.

      2 0 Reply
      1. dan1980
        Reply Icon

        Re: Mainframe zombies?

        @Trevor

        "IBM refuses to price it reasonably."

        The word does not even exist in their corporate culture. They sold their of their presence in the everyday world to Lenovo, I think, just to be rid of any association with rank-and-file.

        Hang on, didn't they recently sell their Watson architecture along with their foundry?

        0 0 Reply
  5. Anonymous Coward
    Anonymous Coward

    Storage or data governance

    Whilst I do love getting boxes of shiny new kit to stack and rack, and then reconfigure it about 5 times as the project priorities change, is the question not more of should the data governance be better so that we don't end up with sprawling kit? If I could get users to classify and store their data with the correct retention period (and no holiday photos, mp3s and Christmas menus), then I'd get roughly 30% of storage freed up. The only critical unstructured data is in the realms of finance, legal and hr - anything else doesn't have a shelf life greater than 12 months. If it does then it needs to be in a structured app so it has some relevance.

    It always seems that when IT looks at the data growth problem, it's more faster cheaper storage with the latest whizz bangs is the answer given and not what is the data and what relevance does it have in the company... guess that means engaging with other teams and getting Records, Compliance and legal involved to have a structured approach...

    Back to the cellars with me to rant at the rats!

    0 0 Reply
    1. Trevor_Pott Gold badge
      Reply Icon

      Re: Storage or data governance

      Exactly why data governance is the new hotness, and new ways to get disks into the datacenter are not. :)

      0 0 Reply
  6. Mark #255
    Stop

    Compass points

    Erm, what is east-west traffic (and north-south traffic)?

    (and where's my "dear lazywebs" icon when I need it, eh?)

    0 0 Reply
    1. Trevor_Pott Gold badge
      Reply Icon

      Re: Compass points

      Hey, sorry about that. Maybe this story will help some?

      1 0 Reply
      1. Javapapa
        Reply Icon

        Re: Compass points

        Trevor, as an old AS/400 application developer, this kind of story helps tremendously understand aspects of computing I never see up close and personal.

        Please clarify one point that is confusing me. When you describe arrays vs SAN, are you referring to NAS arrays? Link below uses term array for both NAS and three subtypes of SAN. And of course there are still legacy Direct Attached Storage (DAS) grinding away.

        http://en.wikipedia.org/wiki/Disk_array

        0 0 Reply
        1. Trevor_Pott Gold badge
          Reply Icon

          Re: Compass points

          By "array" I typically mean SAN and NAS. DAS stuff is usually not called an array. It's just called DAS. (Or JBOD). It's a separate thing. It's usually many more disks than you'll find in a server SAN, but it's not shared across multiple systems...or at least not enough systems to make more than a two or three node cluster.

          DAS is a very Microsoft thing, at least where virtualization is involved. I know it's still a thing for those few running workloads on metal, but only Microsoft really thinks it's remotely viable for hosts with hypervisors on them. But, hey, it's Microsoft. Being trapped 15 years in the past is quite an advancement for them.

          1 0 Reply

POST COMMENT House rules

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

Forgotten password ?

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like