Cloud Woo Insults the Intelligence

My latest column for the Dutch publication, Storage Magazine:

The Insulting Thing about Clouds

“Cloud computing,” and its conceptual progeny, “cloud storage,” are great ideas. However, these concepts haven’t translated into meaningful technology so far as I can tell. In fact, much of the stuff that is being cast as “cloud” today strikes me as intellectually insulting.

Early on, cloud folk were seizing upon server virtualization marketing woo to explain what clouds were all about. Server virtualization, at least the VMware version, portends to lift application and operating system software off of the underlayment of commoditized hardware technology, thereby abstracting the hardware infrastructure. That, in turn, is supposed to facilitate more efficient resource utilization, application multi-tenancy, dynamic guest machine movement across physical infrastructure, etc., which in turn is supposed to deliver a lot of cost-containment and carbon footprint reduction goodness.

The cloud folks took just this hype to the next level: why not detach the entire IT department from the corporate organization and source all IT resources across the Web? Isn’t that the ultimate expression of the virtualization value proposition?

What the marketeers aren’t saying is that the “freedom from proprietariness” promised by virtualization-qua-clouds isn’t really freedom at all. In fact, most of the vendors have been working behind the scenes to raise proprietariness to a new level.

Look at developments at Oracle, with its pre-integrated Oracle DBMS plus Sun servers plus STK storage “solution,” or Microsoft and HP with their pre-integrated rigs, or the soon to be announced server-network-storage hardware plus software bundles from Cisco Systems, NetApp and VMware, and you see what I mean. Rather than separating software from underlying commodity hardware, these vendors are working to “enable clouds” by creating new system stovepipes (I think of them as “mainframe mini-me’s”) that are designed to lock consumers into a particular vendor cadre’s combination of proprietary hardware, software and “standards.”

The idea is that a cloud service provider will then choose one of these rigs and hitch its infrastructure to the associated vendor’s star. As the service provider adds more customers, they just add more copies of the pre-integrated “building block” rig of choice. That provides a horizontal scaling model that is cost-effective and manageable, albeit using a common proprietary management interface.

The problem with this model is that clouds from different vendors, each based on different cloud rigs, will not work and play well together. A customer who sources different services from different vendors will need to use multiple consoles to monitor SLAs and will still require the services of an IT staff to “integrate” services so they map to business processes. And, of course, whatever integration the consumer is doing is subject to failures that aren’t covered by any SLA with any cloud provider.

In theory, the adoption of clouds improves on the current IT paradigm because it eliminates the investments that companies are making in on-premise hardware and software — especially the energy and administrative labor costs associated with contemporary x86 computing. But, cloud computing is actually just an outsourcing arrangement by another name, with all of the associated problems of business process-centric management and integration of services and all the same management headaches accrued to dealing with service providers.

Ask most cloud vendors about interoperability standards and common management based on open standards like W3C Web Services and REST and you will likely get either a blank stare, or an increasingly common talking point about clouds being “too new and innovative” for standards. What it is really all about is money. Hardware vendors don’t want to eliminate proprietary hardware because it enables them to pad margins on otherwise commodity components.

Interestingly, the storage vendors are in a better position than the server folks to abstract all of the value-add software heretofore vested in array controllers into an intelligent software layer, enabling both scale independently of each other. DataCore Software, FalconStor Software, Exagrid, and several others have been doing this successfully for a decade. In DataCore’s case, they have just demonstrated that they could build Petabyte-size virtual disks from commodity storage. At the same time, they can extend the value-add features that people want to use with their storage across any set of arrays, rather than isolating value-add functions to just one vendor’s proprietary rig.

This approach seems to be closer to the realization of vendor neutral clouds than fielding a bunch of application-centric preconfigured systems designed to lock out competitors. What is needed to complete the story is an open management framework, based on Web Services, that will enable the mixing and matching of virtualized storage infrastructures from different cloud services. Anyone who says otherwise is just insulting my intelligence.

Tagged as: Cisco Systems, cloud services, cloud storage, DataCore, Exagrid, FalconStor, HP, Microsoft, Oracle, STK, Sun Microsystems, virtualization, VMware