The Cranky Admin

Should You Build Your Own Cloud?

Private and hybrid clouds are becoming the new normal.

Private and hybrid clouds are becoming the new normal. As the number of vendors offering relevant software increases, it increasingly makes sense to front-end datacenter resources with the sort of self-service, management and orchestration tools that have come to be called "cloud." For most organizations, however, trying to roll your own cloud doesn't make much sense.

Clouds have a lot to offer. They solve technological issues with resource provisioning, business issues concerning the speed of provisioning IT services and political issues related to giving users solutions they want to use so that they don't work around IT, and thus corporate governance.

Clouds can also be transformative in that they often make it easier—or at least more obviously desirable—to use desired state configuration tools such as Puppet, Chef and so forth. While clouds can be deployed and run using cloneable templates as the basis for workload deployment, IT teams will rapidly discover that this is unsustainable in the long run and that composable workloads are the future.

All of this is a "Good Thing." In general, clouds are a Good Thing. Like the last great advancement in IT, virtualization, clouds are a rational and even necessary advancement of the state of the IT art. They're also a right royal pain to design, manage and maintain if you try to build them one component at a time.

Remember the days of SANs and the "standard" three-tier data­centers? Remember when we needed dedicated storage administrators, network administrators, application administrators, OS administrators, virtualization administrators and Jibbers knows how many other roles? It wasn't so long ago. Many organizations are still there.

Hyper-converged infrastructure (HCI) largely removed the need for dedicated storage administrators. Software-defined networking (SDN) is bringing networking to the same level. Desired-state configuration and workload composability are eliminating the need for OS administrators, and so on down the line.

Instead of a group of dedicated administrators needed to individually (and manually) configure each virtual machine (VM), switch or storage array, we're entering a world where all parts of IT infrastructure can be defined by configuration files, and these files can be dynamically generated by scripts or management applications. Even without layering the self-service bits on top that make things cloudy, datacenters are evolving toward something where there's simply an "infrastructure administrator" that spends their day defining everything below the application in YAML.

Of course, for a configuration-driven infra­structure to work, you need management applications, scripts and other bits of software to drive it all. The traditional datacenter roles may be dissolving, but new challenges are on the horizon.

One reason there were dedicated administrators was because they needed to learn the individual products in use. While this is going away with the rise of software-defined infrastructure, the other reason dedicated administrators are needed—fundamental, domain-specific knowledge—remains.

While it's certainly possible to have infrastructure managed and maintained by generalized systems administrators armed with config files, they still need to know enough about the theory behind what they're managing to boss the infrastructure around efficiently. This leads us to the design conundrum.

Specialized domain knowledge is required to design any IT infrastructure, especially one that runs at scale. As software-defined infrastructure places pressure on IT teams to evolve into generalists, they will either have to get the domain knowledge required for design from expensive consultants, or from vendors.

Fortunately, the market is responding. Clouds are no longer restricted to a series of individual applications that must be layered one on top of the other like OpenStack. Cloud-in-a-can solutions, usually based on HCI clusters at their core, are increasingly common.

The future isn't just composable workloads running on software-defined infrastructure. That infrastructure now comes pre-integrated out-of-the-box.

About the Author

Trevor Pott is a full-time nerd from Edmonton, Alberta, Canada. He splits his time between systems administration, technology writing, and consulting. As a consultant he helps Silicon Valley startups better understand systems administrators and how to sell to them.


Subscribe on YouTube