Cloud: Infrastructure Complement, Not Replacement

The cloud is the future of IT, but will it be an all-encompassing future or is there still room for traditional infrastructure?

To hear it from some thought leaders, and the rising cloud industry in general, it is only a matter of time before all enterprise activity winds up in the cloud. After all, how can anyone justify limited scale, higher costs and lesser overall performance by building and maintaining local IT?

Yet it seems clear that, despite its many advantages, the cloud is not the perfect fit for all circumstances, and many business operations are in fact more suited to the local servers, storage and networking elements that are roundly criticized as “the old way of doing things.”

The chief advantage of the cloud, after all, is scalability. But the fact remains that scale comes in many forms. As analytics firm Sociocast learned recently, the cloud excels at scale-out applications like ecommerce and email processing, but Big Data analysis requires a scale-up approach, which can quickly overload even the most robust cloud service. When you need to run more than 300 billion transactions per month, simply adding more virtual servers to the mix produces nothing but a steady stream of diminishing returns. Ultimately, the firm moved its processing loads from the cloud to a managed service architecture that provided access to real, bare-metal servers.

Fundamentally, this is a bandwidth problem, according to tech consultant Daniel Golding. Part of the reason the cloud is so cheap compared to traditional infrastructure is that it relies on the public Internet to relay data to and from the enterprise. This may suffice for normal workloads, but once data hits a critical mass, latency rears its ugly head and thousands of users start to wonder why the lightning-fast responses they’ve grown accustomed to have disappeared at the most critical stage of their operation. Fixes for this problem include dedicated links to the cloud provider or even manual delivery of disk drives to the cloud provider, but these come at a cost.

Then again, some applications just don’t lend themselves to distributed architectures at all, says Cloudability’s Mat Ellis. High-speed trading, for example, probably won’t gel with anything greater than a sub-millisecond response time, which even the most advanced clouds will have trouble providing in the foreseeable future. As well, many legacy applications are geared specifically for local infrastructure — if you want to put them in the cloud, they’ll need an extensive rewrite. And if you happen to occupy a heavily regulated industry, HIPAA and FISMA are still working out compliance details for the cloud.

Ultimately, it’s the cloud’s ability to deliver data on the long-sought-after utility model that is driving much of the buzz. But as Arizona State University’s Dan Gillmor points out, data and electricity are not the same. Data is customized to the individual users, so it can’t be distributed on a commodity basis as efficiently. For highly generalized applications, then, the cloud offers many advantages, but once specific user needs kick in, local infrastructure starts to shine.

Given the history of technological development, none of this should come as a surprise. Few advancements have completely wiped out the old way of doing things despite the cheerleading that accompanies the new paradigm. Even in today’s high-speed, connected world, goods are still shipped by sea and rail.

The cloud will prove exceptional at a great many things, but the wise CIO will keep all options on the table for a while longer.