A useful starting point is to set some boundaries on the notion of using on-network IT resources. I've talked to hundreds of enterprises, and while there's not a lot of agreement at this point on just what they expect IT hosting to mean for them, there's pretty good consensus on what it's not going to be.
No enterprise I've surveyed or worked with has ever indicated that they believed they would host their entire IT process in the cloud, eliminating their enterprise IT organizations. Similarly, none of them indicated they didn't expect to do any outsourcing of IT at all. From this, we can say with certainty that the IT organization of the future will be balancing stuff that's hosted with stuff that's maintained on in-house resources. Planning for, and executing on, that balance may then be the key strategic mission for IT, and supporting it may be the key mission for enterprise networking.
Hosting IT resources on networks seems to be a good idea if any of three things are true. First, the enterprise has applications that are generic in nature and don't interface much with the rest of the business. The CRM tasks of Salesforce.com are examples of this sort of thing. Second, the enterprise has the need for computing/storage resources linked to expanding or sustaining Web presence. The Amazon Elastic Computing Cloud (EC2) is designed in part to serve this mission, and IBM's cloud computing initiative seems targeted there as well. Third, there are applications where special conditions make hosted support of an application more economical than in-house support.
For all of these missions, it's important to realize that the decision to host any IT process is a decision that changes the risk profile of the application considerably. Application performance is a clear issue. Senior management, used to calling the CIO and yelling if some report or another isn't ready on time, may find they have little influence on third-party hosting providers. In an era where data theft seems a regular theme in technical and business publications, the fact that some of your data resides on someone else's computer, where you have little ability to control or even understand the security that protects it, may be a tad disheartening. Finally, any form of IT hosting makes the enterprise more vulnerable to network problems.
The latter may seem counter-intuitive. After all, a branch office or a remote worker is network-connected to applications today, so what does it matter if the network is hosting the whole application? The answer is that even today most enterprise networks are private, based on leased resources and not on Internet VPNs. A transition to a hosted IT strategy for even a single application component may mean that the entire application is now subject to the variability of the Internet in terms of QoS and security.
The application performance issue is two-dimensional. First, it is clear that a hosted application uses resources an enterprise doesn't manage exclusively and so might become congested through the behavior of others. A more insidious issue is the impact of Web Services on application performance. Many of the elastic computing and hosted software approaches today are based on the use of Web Services, and while Web Services are a powerful tool in application development and deployment, it's possible to build applications in a way that creates tremendous performance bottlenecks. Many times development teams, focusing on a mandate to use "state-of-the-art SOA techniques" and "create Web Services to increase modularity and portability" will forget that too much SOA can slow applications by an order of magnitude.
The issue of special cost/benefit justifications is the most thorny of all, because it's always difficult to totally quantify either side of the cost/benefit ratio in the early phases of considering a hosting strategy. One of the best examples is the move by some enterprises to consider hosted office tools rather than purchasing Microsoft Office or using Open Office. The justification is that the continued installation support of the software on a whole series of PCs is expensive. Possibly true, but how much of that expense is really offset? It's true that software installation and updates may be eliminated using hosted office tools, but it's also true that relying on a fully hosted approach means that if you lose the connection to your hosting provider for any reason, the workers out of touch are unable to use their basic word processor or spreadsheet tools. That's a big downside risk, and countering it may re-introduce some of or all of the cost you'd hoped to save.
A quick poll of 30 enterprise CIOs showed me that all but one believed that special planning was required to support any hosted strategy, all but five said that their organization had not completed the planning required, and nearly half indicated that they believed their current planning was "seriously deficient". Twenty-two said that they were being pressured by vendors or suppliers to host some IT resources, but that the sources of the pressure were not providing any insight or guidance on planning for the hosting decision.
Hosting is very likely to be in the long-term future for everyone, but it's also clearly something that will demand more planning maturity and supplier support in order to be truly successful.
