Considering SDN? Adopt Chaos Monkey
The way Ivan Pepelnjak sees it, every company could benefit from its own Chaos Monkey.
What's Chaos Monkey, you ask? It's a service Netflix built -- and now relies upon --to ensure the resiliency of its applications running on Amazon Web Services. It does this by randomly shutting down servers to see what the effect is on applications, and thus the customer experience. Netflix kills thousands of virtual instances each year this way, solving any problems that arise, and each time its applications get a bit stronger and more reliable.
As a result, the company doesn't care about hardware availability. It has dozens of virtual servers supporting the same application, and it doesn't even worry about replacing the servers it kills as it's constantly adding new ones to handle increased user loads.
"Their application architecture is resilient," Pepelnjak, chief technology adviser at consultancy NIL Data Communications, said in an interview. "If any component fails, the whole operational stack is still working."
It's an approach that's long proved successful at Netflix and other big-name online services, such as Amazon and Google. And much to Pepelnjak's pleasure, it's starting to catch on elsewhere.
"That type of mentality is slowly moving into the enterprise side of things," Pepelnjak said. "It will make things simpler, cheaper and easier to manage. You stop worrying about all the possible failure scenarios. You only worry about how resilient is your application."
According to Pepelnjak, companies can design their data centers to be hardware independent, and thus much more agile, by tapping into the software-defined phenomenon.
Doing so, he said, will enable companies to do application development in environments that duplicate the live network rather than in the stand-alone testing environments they've long depended on, which often don't adequately prepare apps for life in the real world.
Already, he said, many company are realizing the benefits of that transition. They're discovering that it doesn't make sense to buy vendor-branded boxes, and are instead buying generic servers and deploying firewalls, load balancers and other layers as virtual instances on those commodity machines.
Whether the resulting environments are called software-defined data centers, software defined networks, or network functions virtualization doesn't matter -- those are all just buzzwords in Pepelnjak's view.
What's most important, he said, is that companies not expect that they can throw technologies labeled as software-defined at a problem and expect results. Taking full advantage of software-defined data center and networking innovations requires a bit of IT soul searching.
"If you want to really get the business benefits, you should go back and totally change the way you think about applications, and totally change the way you think about security," Pepelnjak said.
[Read how Ethan Banks thinks the job of a network pro will change with software-defined networking "Network Engineers: Don't Be The Dinosaur."]
For example, an organization can place virtual layers in front of every application, thus making app teams responsible for their own firewalls, load balancers and the like, and turning network teams into consultants. Doing so can remove the roadblocks to rapid app development by enabling app teams to test their work in sandboxes that imitate production environments.
In order to make this approach work, though, Pepelnjak said companies would have to tear down walled gardens and shift responsibilities so that network and application teams work more in tandem. The traditional paradigm in which apps teams work in a disconnected silo and then point fingers when their apps don't run as expected will no longer work.
"You can deploy these new technologies and keep doing things in the old way, but then you're just wasting money," he said. "You have to reorganize the way you do business in IT."
Combining a new approach to app development with adoption of emerging software-defined technologies could result in a compelling value proposition for organizations willing to embrace change.
"Building private clouds will be way simpler and cheaper," Pepelnjak said. "And they'll be way more robust than they were a couple of years ago."
[Don't miss Ivan Pepelnjak's workshop, "Designing the Virtual Network for the Software-Defined Data Center" March 31 at Interop Las Vegas. Register today!]
Recommended For You
Innovative companies are leveraging open frameworks such as composable infrastructure to forge a path toward making fabric-based infrastructures a reality.
Anyone deploying NVMe-oF should examine their existing infrastructure investments, IT roadmaps, and expected workloads to choose the right flavor of NVMe-oF.
By enabling up to 30 times faster deployment than firewalls, and simplified management, segmentation reduces CAPEX and OPEX compared to traditional techniques.
DevOps principals have influenced how infrastructure is managed at scale using IaC strategies.
To meet increasing consumer and enterprise demand for data (especially from video streaming and IoT devices), data centers must upgrade their infrastructure.