Machine addressability: Just about every major cloud service provider now offers a set of APIs developers can use to call on some aspect of a cloud service, whether a CRM application or a disaster recovery routine. Customers can then build custom capabilities that meet their specific needs on top of the core capabilities delivered by the service provider.
The shared infrastructure aspect is still the main sticking point for potential enterprise customers of cloud services. They don't like the idea of their database tables or server images sitting cheek-by-jowl with those of other customers. Rightly or wrongly, they see it as a security risk.
When services aren't shared, however, and customers get dedicated resources, you have conventional hosting. That may mean up-front costs for hardware or software, the lack of which is one of the chief benefits of the cloud. Even if those costs are spread out, scalability won't be as transparent, and the provider's overhead of maintaining dedicated resources will be passed along to you in cost and longer turnaround times.
Is there such a thing as a private cloud?
According to IDC, the top three objections to cloud services are security, availability, and performance. Not that any of these things will necessarily be worse than what in-house IT has to offer. The problem is that all three are outside IT's control -- which is why, among the enterprise execs IDC surveyed, the so-called "private cloud" was preferred over public cloud services by four to one.
But let's be absolutely clear: The idea that you can transform an entire, existing enterprise datacenter into a centrally managed private cloud, any portion of which could be provisioned for any application on the fly by stakeholders, is nonsense. Public cloud service providers are designed from the ground up for massive scale (and often a specific purpose). Typically, they have fully virtualized infrastructures with homogenous commodity hardware running open source software. Plus, self-service provisioning, multitenanted applications, and chargeback systems do not build themselves.
So we are left with more modest ambitions. We can all agree that at least one technology that underlies public cloud services is taking the data center by storm: server virtualization. But I do not believe, as VMware seems to, that large-scale virtualization and the cloud are pretty much the same. A cloud service, inside or outside the firewall, must behave like a service.
The IBM approach to the private cloud has been quite creative: the cloud appliance. Instead of signing up for a service outside the firewall, you buy a turnkey system that meets most of the criteria of a public cloud service and set it up in-house. IBM CloudBurst appliances are fully virtualized blade servers with self-service Web interfaces and chargeback metering built in. So far, the company has rolled out a CloudBurst appliance for business analytics and a WebSphere CloudBurst appliance for dev and test.
IDC predicts that, next year, IBM will have plenty of company. In its top 10 predictions for 2010, the analyst firm anticipates that "virtually every major systems and application software vendor will introduce 'cloud appliances' as a very simple-to-adopt packaging approach for its private cloud offerings."