In just two memos, Navy Chief Information Officer Terry Halvorsen defined computing for the Navy and Marine Corps into the foreseeable future. Every server will be virtualized by 2017, and 7,500 desktops will no longer live on someone’s desk, rather that data will reside in Navy data centers.
The goals are ambitious, but they obscure a fundamental change for which virtualization is just a first step.
In one respect, nothing could be less controversial than virtualization. Over the last five years, virtualization has become the standard method of deployment across government. As servers grow in capacity, it makes less sense to continue with the one-workload-per-server habit that has developed over the last 20 years of client-server computing. Halvorsen is simply formalizing, and maybe hastening, a transition that has already started.
The immediate goal, of course, is cost reduction. His measure of success: the number of servers converted. I’m sure this is informed by the Navy’s positive experience so far. In July 2012, Halvorsen announced that the Navy had eliminated 2,000 of their 6,000 shore-side servers using virtualization, with a commensurate reduction in capital expenses. But in exchange for a reduction in capital expenses, virtualization has non-obvious effects on operational expenses. Software often is more expensive than hardware, including some of the software that runs on these virtualized machines. Some vendors force you to pay to license the maximum possible capacity, rather than what you are actually using. This is particularly true for desktops, where the licensing for office productivity software can be many times the cost of the machine itself. Also, virtualization reduces the friction of creating new servers, which can balloon operational costs: those servers still need to be fed and clothed, even if they don’t reside on physical machines.
Halvorsen no doubt understands all this and decided to virtualize anyway. He’s playing a long game and it has little to do with up-front reductions in capital expenses. By virtualizing, he’s building for the future.
Halvorsen has every reason to expect continued budget cuts over the next few years, and he has to prepare for that. Virtualization is an easy choice: he gets the quick reductions to capital expenses he needs, and moves the burden to operational expenses, which are more easily managed and controlled. More importantly, his datacenters and desktop deployments can respond much more easily to changing demands and budgets. They can be moved to new hardware platforms without downtime. They can be easily duplicated. They can be allocated more or less hardware as missions change. They can be better managed and accounted for — a server cannot hide under a desk or in a closet when it is virtualized. Virtualization is also a precondition to moving to clouds, which everyone expects to be the norm in the very near future. When Halvorsen is confronted with an unmanageable uncertainty in budget and mission, these benefits of virtualization are not pleasant side effects. They’re mandatory for the datacenter he will need in 10 years.
This move, though, has a risk. At the moment, a single company is doing the lion’s share of enterprise virtualization: VMware. There’s good reason for this: they have an excellent product, and practically invented virtualization as we know it. But that doesn’t mean that one company should have effective control over an entire layer of the Navy’s architecture. That kind of control gives strong incentives to extract as much rent as possible from its Navy customers, undoing Halvorsen’s heroic cost-cutting measures. Worse, the Navy may find that the flexibility and agility of virtualization is actually undone by a vendor who has no reason whatever to ease migrations to other, cheaper platforms.
This is not hypothetical. Pat Gelsinger, VMware’s CEO, said at a partner conference earlier this year, “We want to own corporate workload…we all lose if they end up in these commodity public clouds. We want to extend our franchise from the private cloud into the public cloud and uniquely enable our customers with the benefits of both. Own the corporate workload now and forever.” Gelsinger’s statements are something to keep in mind as the Navy moves much of its unclassified work to public clouds.
We have seen this movie before, on the mainframe, proprietary UNIX systems and relational databases: when you have one supplier for a piece of critical infrastructure, things get expensive quickly and it is very difficult to leave.
Fortunately, alternatives exist. The trick for Halvorsen is ensuring that these alternatives are practical. That means ensuring a procurement environment that encourages competition among virtualization vendors. It means actively training staff on multiple platforms. It means creating a level playing field by employing open standards wherever possible. But most importantly, it means promulgating the policies and procedures the Navy needs for an orderly exit from any virtualization vendor they provide.
If Halvorsen is able to take the same amount of courage and resourcefulness that went into these memos on the acquisition of a technology, and applied them to the exit from that technology, he will be extraordinarily well positioned for whatever the future brings. That is what leadership looks like.
Gunnar Hellekson is chief technology strategist of Red Hat U.S. Public Sector.