Everybody remembers the classic fairy tale where an emperor and his people are conned in to believing he was attired in a fantastically beautiful set of clothes, when in fact he was in the buff. No one was willing to admit they did not have the refined taste and intelligence to see the spectacular cloth and splendid robes. It took the strength of innocence in a child to point out the truth. I am about as far from an innocent child as one can get, but it appears to me the cloud is parading about naked.
Every vendor has a cloud offering, every pundit “agrees” the cloud is the future, investors value every cloud company with a premium, every data center operator is “born again” as a cloud player. Every CIO has a cloud initiative and budget line. Really, I have seen this movie plot before, and it does not end well, especially for the Emperor (and the con-men vendors too).
We have worked internally on projects as well as externally with clients to implement aspects of the “cloud”. Results have been mixed and in the process gathered some hard won experience which I will condense here (while protecting both the clothed and the naked).
First, Software as a Service (SaaS) will work if adopted with minimal software modification and maximum adoption of it’s native business process. It is very cost effective if it precludes investment in internal IT infrastructure and personnel, not bad if it slows the growth of same. Outsourcing well-defined rote functions to the SaaS route works well (such as Email). Adopting SaaS for new non-strategic functions tends to be successful where there are few users and a high degree of specialization. Data backup into the cloud is an excellent example regarding highly specialized solutions that take advantage of economies of scale provided in hardware.
SaaS fails in terms of cost or functionality when it is subject to customization and extension. Service costs tend to swamp the effort from initial modification through long-term maintenance (humans=$$$$). Costs will especially spiral when you combine many users and many customizations. Remember the “Keep It Simple, Stupid” (KISS) principle saves money and points to success.
Buying virtual machines in the cloud works well if the configuration is simple; few software products, few users, straightforward integration. Development and early deployment is particularly attractive, as is usage by start-up companies and software proofs, tests, and trials. Again, the KISS principle reigns supreme. Remember hardware continues to drop in price and increase in capacity. Package software costs are stable. Understand the billing algorithms of the key “clouds”. Each has its cost advantages and drawbacks, and they change rapidly under increasing competition and hype. Always benchmark medium to long-term cloud virtual machines against native hardware virtual machine implementations, the results may surprise you (I have been surprised over and over).
The Emperor’s story is an old one and so is the cloud concept in principle; remember its first turn on the karmic wheel of optimizing the highest cost component was time-sharing. This strategy optimized the high cost of proprietary hardware/software (remember IBM and the Seven Dwarfs, but I digress into another fairy tale). As minicomputers (Digital, Data General, Wang) dropped the price of hardware through competition with IBM, software packages became the gating factor. Workstations continued the trend by another factor of 10 reduction in cost of hardware and package software (human service costs are rising). Wintel and the Internet have driven the marginal cost of raw computing to almost zero compared to the service component. As hardware has followed Moore’s law and software package economies of scale moved to millions of copies, the human costs have skyrocketed in both relational and absolute terms.
If we can keep history as our lens and focus on our cost pressure points, we can maintain our child-like innocence and see others prancing naked while we keep our kilts and heads about us.