14 Comments

Summary:

If you successfully launch a number of web firms, at a certain point the economies of scale of others’ clouds starts fall away and you may as well run your own. But is it always a good idea to build your own cloud when you get big enough to do so?

New York-based cloud computing startup 10gen launched today with backing from CEO Kevin Ryan’s startup network, Alleycorp. It makes sense, since with several ventures already under his belt, Ryan probably has enough customers to both justify the buildout and break even right away. And the founders know scaling, having built out ad network DoubleClick.

But is it always a good idea to build your own cloud when you get big enough to do so?

Yesterday, for example, I had a great chat with Lana Holmes, a Bay area startup maven, about product management and how to focus on doing the one thing that matters to your company. “The example I use is Amazon,” she said. “They just focused on selling books. And look at them now.”

At their root, Amazon’s EC2 and S3 offerings are the result of excess capacity from sales. The offerings have paved the way for an online world in which compute power is a commodity. The company has subsequently built, on top of those offerings, a layer of billing, services and support for them.

The motivation behind the creation of 10gen is similar: If you successfully launch a number of web firms, at a certain point the economies of scale of others’ clouds starts fall away and you may as well run your own.

It’s easier than ever to launch your own cloud. You’ve got grid deployment tools from folks like 3Tera and Enomaly. Virtualization management can be had from the likes of Fortisphere, Cirba and ManageIQ, to name just a few. And license management (built into cluster deployment from companies like Elastra) is knocking down some of the final barriers to building a cloud that you can offer to third parties as well.

But imagine a world in which there are hundreds of clouds to choose from. Moving a virtual machine is supposed to be as easy as dragging and dropping, and cloud operators will hate that. They’ll resist, putting in proprietary APIs and function calls. Applications and data won’t be portable. You’ll be locked in to a cloud provider, who will then be free to charge for every service. Sound familiar?

My guess is that as the cloud computing market grows and matures, one (or more) of three things will happen:

  • Standardization and portability, in which consortia of cloud vendors agree to a standard set of APIs and coding constraints that guarantee interoperability. This isn’t just about the virtual machines; they’re fairly standard already. It’s about the data storage systems and the control APIs that let cloud users manage their applications. This is the mobile phone model, where number portability is guaranteed and there are well-known services like voice mail and call forwarding.
  • Shared grid computing, in which smaller clouds sell their excess capacity to bigger clouds. This would let the big cloud dominate while paying the smaller cloud just enough to stop it from launching an offering of its own. Think of this as the electric company model, selling computing between clouds the way a solar-powered household can pump excess electricity into the power grid.
  • Specialization, where clouds are good at certain things. You’ll get OS-specific clouds (Heroku is already providing optimized Rails deployment atop EC2.) It’s only a matter of time before we see clouds tailored for specific industries or the services the offer — anything from media to microtransactions. Sort of like the cable channel model, with specialized programming that allows niche channels too survive.

Whatever happens, it’s clear that good old-fashioned branding, plus a healthy dose of experience, will be key to winning as a cloud provider.

During a panel at Interop last week that I sat on with folks from Amazon, Opsource, Napera, Syntenic and Kaazing, I asked the audience how many of them would entrust Microsoft to run a cloud with Microsoft applications, and how many would prefer to see Amazon running a Microsoft kernel on EC2. Roughly 75 percent said they’d trust Amazon to run Microsoft’s own apps rather than Microsoft.

So when’s the right time to launch a cloud computing offering of your own? Unless you have the branding and reputation to support that launch — or you can re-sell excess capacity to partners or specialize — maybe never.

You’re subscribed! If you like, you can update your settings

  1. Should You Launch Your Own Storage Cloud? | MSPmentor Thursday, May 8, 2008

    [...] spotted a very interesting article on GigaOm today. It explores the factors you should weigh before launching your own storage cloud. It’s a particularly timely read for managed service [...]

  2. I asked the audience how many of them would entrust Microsoft to run a cloud with Microsoft applications, and how many would prefer to see Amazon running a Microsoft kernel on EC2. Roughly 75 percent said they’d trust Amazon to run Microsoft’s own apps rather than Microsoft.

    Did you ask the follow up question of why they’d trust amazon more than msft to run the apps?

  3. Uhmm, I’d like to see a couple of major ‘clouds’ condense and challenge Amazon but I don’t see the point of having proprietary platforms putting the clamps on developers and their software.

    Let the clouds take on the character of the open and collaborative web and not the other way around. Open sourcing the grid and its API’s would make the most sense especially for ISV’s who are careful to thread the waters because of Lock-in but do have a really interesting app. Doing anything other than that is like feeding them to sharks.

    Best..
    alain
    morpheXchange.com

  4. Amazon, Google, Sun, and MS probably pay no license fees for their cloud infrastructure and they will also have large economies of scale. Given these facts, I doubt it is possible to be cost-competitive with the big guys while using proprietary management software.

  5. Wes, they may not be buying in licensed software, but all four you mention will be building their own to some extent – and seeking to recoup those costs from their customers. The economies of scale you mention will bring the costs down, but the need to support different customers with different scenarios will inflate them again; if you have simple or special requirements, particularly if you have a big system anyway, you may well come out ahead by doing the same thing for yourself. Obviously Amazon, Google, Sun and MS all thought it worthwhile to build their own, and I’m sure many others have as well: why did Google build their own rather than buy into Amazon’s?

  6. Alistair Croll Friday, May 9, 2008

    @vijay: Yes, as best I could with a show of hands. They seemed to feel that Amazon has 2 years’ experience with the problem of running things, which gave them the upper hand. Not sure why they ignored Live and Hotmail though…

    @friarminor: I’ve actually been chatting with friends about the idea of “open service” versus open source — a loose federation of services that keep the for-pay ones honest. Openoffice limits the maximum amount Microsoft can charge for a word processor; why shouldn’t a set of RESTful APIs and hosted platforms, built on space computing cycles from willing participants, do the same for hosting?

    @Wes: Software costs are IMHO immaterial when you consider that most of the platforms are wrought from open stuff anyway. Not sure license fees are entirely relevant here, unless you’re talking about enterprise applications that have a per-CPU cost (in which case, you may want to consider a new license.) As you point out, this isn’t the case for for-profit licenses like MS. AMZN not only has custom kernel support, they’re also getting lots of pressure to put MS kernels on EC2 — something that the audience felt was the top feature needed.

  7. Wes Felter Friday, May 9, 2008

    Alstair, this article suggests that people should buy software like 3Tera to manage their own clouds, and I’m saying that once your cloud gets large you’ll be paying a *lot* of money to 3Tera – maybe more than it would cost to do it yourself.

  8. IMHO, most effective strategy to be adopted by small players is specialization. To the end user, clouds give infinitely scalable bandwidth, computing power and virtualized operating system. But spcialisation plays a key role in differentiating your service. Someone building a cloud entirely for java based applications can have edge over plain vanilla cloud providers running VMWare or XEN. Some may even resell smaller portion of larger cloud as their own, just like resellers in webhosting business.

  9. “It was twenty years ago today…”

    Has everyone forgotten the Unix Wars already? Go Google them, and then brace yourselves for round 2.0, where once again the difference in semantics between the English word “standards” and the technology industry’s interpretation of same gimps thousands of innocent developers.

    Anyone who even *dares* to mention POSIX will feel the sharp end of my boot…

  10. Ideas: Social Compile Cloud (cont.) : Geek Building The Bridge Part 2 Saturday, May 10, 2008

    [...] mengawali penjelajahan saya atas Web 2.0. Hari ini GigaOm is back in my heart. Post-nya tentang tren layanan cloud menggelitik otak saya dan akhirnya ide SCC mendapat bahan [...]

Comments have been disabled for this post