Building A Sustainable Cloud (And More Predictions)

I thought that the cloud was supposed to be about the aggregation of thousands of low-cost servers and disk drives, integrated together with a web services layer and a robust object store, all accessed as a pay-as-you-go service. Wasn't that how cloud service providers were supposed to make money--by using commodity components with their own "secret sauce" software on top? High-cost legacy infrastructure products are not congruent with building a long-term sustainable cloud.

In my opinion, knowing some history is a great way to understand the present and a good base of knowledge to help guide the future. The deceased storage service providers (SSPs) of the late-90s/early-2000s--like Storage Networks, Storability and Sanrise--all tried to build infrastructures using premium-priced, high-end IT gear from companies the likes of EMC, HDS, IBM and others. They failed. Why? Sure, bandwidth was a problem back then. However, they learned a hard lesson--that you simply couldn't make money selling storage services atop multimillion-dollar machines.

Yet these days I am again starting to see a parallel to the SSP world. You see, 10 years later, we have the advent of Vblocks, Vplexes, VSPs and V Series galore--just like we used to have Lightnings, Sharks and Symmetrixes. But I'm not entirely sure what kind of service providers can afford to build out an infrastructure on these types of premium-priced IT products and still hope to make a profit. If you look at the portfolios of the big storage vendors, they are all taking expensive, high-end storage gear and putting the word "cloud" in front of it. If you are a service provider, they all want to sell you a cloud tool kit so you can deploy your own cloud and then go lease out capacity on their premium-priced gear and try to make a buck or two back from your own customers.

You have to hand it to these vendors, which do a great job of obfuscating the real requirements of cloud storage. They're not talking about usage-based billing, billions of files and objects, storage consumed as a service, the elimination of maintenance fees and forklift product upgrades. They're still selling their proven enterprise storage model--pay me now for everything and at a premium price--into the cloud.

But the whole point of the cloud is economies of scale. That means you need to have IT systems that can be utilized to help you achieve economies of scale, not high-end enterprise hardware stacks. VPlex is for cloud? It's not even storage, and it costs more than storage.Whether it's EMC Vmax, VPlex, Isilon, Atmos or NetApp, none of
these products are priced per usable gigabyte or terabyte of storage. It appears that all are priced on raw storage, the same as their software capacity licenses. In my opinion, the real cloud is a threat to their model, which is why they are all scrambling to be the "outfitter" to the cloud service providers. They'll bring your service customers, as long as you buy their hardware. They say they won't offer storage services because they don't want to compete with service providers. However, in my opinion; they're worried their margins are getting shredded.

While thinking about this scenario, I decided it would be valuable to understand just how well AT&T is doing with Synaptic Storage. After speaking with a number of people associated with Synaptic Storage--VARs, resellers and consumers of the service--it's my opinion that AT&T Synaptic Storage, which runs on EMC Atmos, has made less money on storage services to date than the cash AT&T provided to EMC for the storage farm in the first place. This is not a slam against EMC; it's opinion-based information. ATMOS has value to EMC; EMC has a product offering. However, in my opinion, the sustainable cloud requires a vastly different storage model.

I believe that service providers serious about running a sustainable cloud storage business will not repeat the mistakes of the SSP era. As an industry, we've tried the premium-priced hardware route once. Now it's time to move on to more innovative cloud infrastructures.

As I predicted, venture capitalists realized that cloud gateway providers are an inflated investment. The implosion of Cirtas proves that my analysis was correct--too much hype around the word cloud has led investors to rush money into stand-alone cloud gateway products with questionable value.

Don't get me wrong: Cloud gateways will exist as part of the portfolios of bigger players. However, I would expect more of these small players to start folding up shop much like Cirtas. As I said earlier, nobody charges for moving data to your tape library, so why would you pay for data to be moved to the cloud? Why would you pay cash up front for a cloud gateway, then pay a price per gigabyte for all the data that goes through it (uploads and downloads) on top of the price per gigabyte that you're already paying your cloud service provider? That's like running a toll road to the cloud! And that's a question that the likes of StoreSimple, Twinstrata and Nasuni will need to answer after the bloody aftermath of Cirtas.Companies will stop building their own data centers. Companies simply can't afford to continue to build data centers at 50 megawatts of power per data center--up from 2 megawatts just 10 years ago. This means that in 10 years the amount of power to run a data center went up by 2,500%. At this rate, it will take 1,250 megawatts of power to run a data center 10 years from now, and most of that will be for storage. Companies building their own data centers are going to need their own co-generation power plants just to keep their data center going. Scratch that, they'll probably need windmills--and a lot of them.

All of these explosive power costs will inevitably lead to increased cloud deployments in the next three years. The benefits of cloud economics--like keeping multiple copies of your data in high-security data centers that someone else owns, maintains and pays for--will finally put an end to the pressures and financial difficulties of individual company data center build-out.