Commentary

Hybrid is not a strategy, it’s a symptom

Gunnar Hellekson, chief strategist for Red Hat's U.S. public sector, writes that uniting developers and integrators will help agencies find success in the cloud...

Gunnar Hellekson

June 30, 2015 7:09 am

4 min read

Cloud has turned most IT shops upside-down, and agencies have spent the last few years feverishly planning their response. As with most trends, it started with maximalists: everybody was going to public clouds, and if you didn’t join them, you faced irrelevance and death.

Opponents declared that the public cloud was far too risky, far too dangerous and would take food from the mouths of hard-working IT staff. In the meantime, the vast silent majority have taken the virtualized data centers they created during the Federal Data Center Consolidation Initiative (FDCCI), slapped a “CLOUD” sticker on them, and declared their mission accomplished.

Gunnar Hellekson is the chief strategist for Red Hat’s U.S. Public Sector group

We now know who was right: everyone. Sure, agencies such as the Federal Communications Commission and the Consumer Financial Protection Bureau have jumped head first into public clouds, such as Amazon EC2. The CIA even took the unusual step of buying its own Amazon region outright. But we know these stories only because they are outliers.

How an agency might take advantage of cloud is a complicated decision based on budget, internal skills, strategy, authority and a tolerance for change–an approach which makes black-or-white ideologies insufficient.

Stealing a turn of phrase from Stephen Fry, government applications are “human-shaped, not idea-shaped, and humans are great heaps of inconsistency, ambiguity and complexity.” This inconsistency demands that many kinds of infrastructure – public, private, on-premise, off-premise – be in play simultaneously. Some systems will work great in public clouds, while others belong in a government-owned data center. Most agencies will use both for the foreseeable future. This “hybrid cloud” approach is not doctrine or strategy, it’s a concession to reality and a compromise between the attractive economics of a cloud platform and the grim realities of the incumbent workloads.

With the hybrid cloud compromise in mind, agencies will reliably lay the challenge at the feet of the operations and infrastructure teams because it is so obviously an operations problem: “Here’s the portfolio of applications. Please build or rent the best home for each, put a self-service portal on all of it, and make sure they all stay reliable and compliant. See you on Monday.”

The results are uniformly terrible. Most “hybrid” clouds are little more than the existing virtualized infrastructures with some management tools on top.

Low success rate for private clouds

Even without the complexity of hybrid cloud, Gartner estimates private cloud efforts fail 95 percent of the time. As Matt Asay puts it, this infrastructure-centric approach “lets enterprises pretend to be innovative, embracing pseudo-cloud computing even as they dress up antiquated IT in fancy nomenclature.”

The problem is that operations is ill equipped to solve the hybrid problem on its own. Their traditional measures of efficiency, reliability and predictability don’t apply in the same way when cloud is involved. Manufacturing teaches us that by chasing efficiency and perfection, we rob ourselves of the “slack” we need to stay agile and responsive. “Slack” is exactly what the cloud provides. The pursuit of “slack” is the profession of the other half of an IT shop: developers and application owners.

Where traditional infrastructure will put reliability in a platform, a cloud prefers resilience in an application. Any transition between the two requires a renegotiation of traditional IT responsibilities. Rather than asking operations for a high-availability server cluster with failovers and redundancy, for example, developers may create applications that recognize and accommodate system failures themselves. A hybrid cloud approach that does not allow for this renegotiation will fail.

Don’t misunderstand this as a power grab by developers. Infrastructure and development should have a symbiotic relationship, with each influencing the other. This was true before cloud arrived, and it’s just as true now. Without that cooperation, we get an “impedance mismatch.” If you only release new software every six months, it doesn’t matter if you’re using OpenStack or your infrastructure can be built with a click on a website. If you’re deploying software 10 times a day, you’ll be frustrated if you have to open tickets for the servers you need.

Instead of starting bottom-up with a fancy new infrastructure management initiative and inappropriate measures of success, start with a goal that will make a meaningful difference to performance and unite these two camps. It’s not uptime, application density, or applications per operator – those are incremental, marginal improvements. Instead, measure the time it takes for an improvement to make its way from a developer’s fingertips to production. “Time to market” or “deployments per day” are both leading indicators for the overall performance of an operation, and an excellent way to focus the efforts of entire teams.

This process of continuous improvement has no end. Every incremental improvement is the responsibility of the entire IT staff, not one department or another. Each constraint will drive the correct infrastructure choices, whether on-premise, cloud, or anything else. In this way, application development strategies drive hybrid cloud strategies, not the other way around. This makes hybrid cloud the symptom of success, not its cause.

Gunnar Hellekson is the chief strategist for Red Hat’s U.S. Public Sector group, where he works with systems integrators and government agencies to encourage the use of open source software in government.