Future Clouds: Centralized Or Decentralized?

Risk concentration is one of the issues to consider as cloud computing evolves. The economies of scale that have brought cloud computing to where it is so far seem to point to further consolidation and the growth in size of a smaller number of mega data centers.

When Vivek Kundra, the outgoing federal CIO, spoke at the Federal Cloud Strategy at our CSA Summit earlier this year, my favorite slide in his deck compared the federal government to IBM in data-center consolidation. Whereas both had several hundred data centers in 1997, the federal government now has more than 2,000, while IBM has 12!

It seems as though the trend is toward putting more eggs in fewer baskets -- albeit more efficient and I believe more secure baskets. But is that truly the case? I can see Moore’s Law and management efficiencies continuing to support this trend, but I think the wild card is the cost of energy. It could very well be that this is the variable cost that upsets the apple cart, and the cost of cloud services might track the cost of energy over time.

In the U.S., many data centers have been built in eastern Washington and Oregon to take advantage of cheap hydroelectric power. It is easy to imagine a variety of events that could radically change the energy cost basis of a data center.

Greater decentralized clouds could mitigate this issue, and it is not hard to imagine more sophisticated versions of the cloud-brokering solutions of today helping customers move workloads around to lower energy cost data centers. If you take this idea to its extreme, the compute power of a few million smartphones could be pretty tremendous, and the energy costs are zero. Is it possible that the future of cloud will be a significant amount of mobile clouds?

Management costs could be higher for something like this, but there have been very good examples of well-managed distributed compute networks for years; my favorite is the botnet.

I don’t know whether this is the future, but I think we need to plan for this being a possible outcome. Clouds might include a lot of untrusted, low assurance infrastructure, and thinking of our security layers in completely virtual terms is very healthy. Building security into the applications, abstracting between the different technological layers, protecting the data wherever it might go, and instrumenting every entity (virtual machines, hypervisors, data stores, users, etc.) with identity management and nonrepudiated logging technologies is essential.

None of us really knows what the cloud might look like tomorrow, so think about implementing security in a way that allows us to take advantage of its future -- or some alternate futures.

Jim Reavis is the executive director of the Cloud Security Alliance, and president of Reavis Consulting Group.