Things I heard (and said) at #DevopsSummit
#devops #cloud #sdn
It was a long and full day for me at Devops Summit. I could fill a week or more with blogs on ideas I both agreed - and of course, disagreed - with from sessions and panels.
But for now, here's a digestible selection.
From John Willis (@botchagalupe) session on networking and devops (which is something you'll be hearing more and more about, even if we are calling it something else, like maybe SDN or a part of the more overarching SDDC or even <gasp> cloud): "servers became ephemeral infrastructure"
This is the descriptor I've been looking for, "ephemeral infrastructure." Whether it's servers or switches, the use of the term "virtual" implies virtualization, a la Hyper-V or VMware. And maybe for servers that's okay, though as we move toward container-based architectures (a la Docker) and into the network with its virtualized (but not virtual machine) form of multi-tenancy, there needed to be something to call it that distinguished it from physical infrastructure without conjuring up images of virtual machines (sorry, pun not intended). John nails it, I think, with the term "ephemeral infrastructure."
The Silos that Lie Beneath the Silos
Problem of #devops is you can't break down IT silos until you first break out of systems silos.
This was my simple observation and I touched on it in my session: infrastructure services aren't islands, they're part of a larger process and there exist dependencies and relationships between them that need to be bridged before we start worrying about bridging IT silos. That latter statement was one I heard multiple times in many different devops sessions said many different ways, though the best was definitely during a presentation by Jeff Sussna (@jeffsussna) discussing the need for empathy in moving forward with successful devops initiatives.
The fact is that we talk about Devops as cultural change and about bridging gaps between dev and ops (two of IT's four silos) but until we can break out of the systems silos that operations has built we're not going to get very far in practical terms. We need to recognize that just as applications have interdependencies and can impact each other in terms of scale, stability and performance, so too can the systems and infrastructure upon which those applications are deployed. And we need to recognize and manage those dependencies as a first step toward bridging other, more significant, gaps between IT groups.
Continuous is the new Cloud
If I had a dollar for every time I heard the phrase "Continuous <delivery, integration, deployment, improvement, <insert IT-related thing here>>" today I'd be flying home first class, let me tell you.
Within devops itself seems to be a variety of sub-categories, if you will, that bring the focus of devops onto a specific IT concern, like deployment, integration and delivery. There are a whole lot of them right now, and the formula seems to be to tack "continuous" onto the front of a traditional IT concern and then it becomes devops.
It's the new cloud, because many of these continuous "things" are not so much emulating cloud, but attempting to achieve the same goals - seamless application deployment through abstraction of infrastructure that enables operationalization. And because it's likely to experience the same hype cycle as cloud, watch out for continuous washing.
SDN and Devops
John Willis did a session on essentially SDN and how its operational aspects are related to devops. While I'm not in agreement with his view on limiting SDN to OpenFlow, the truth is that SDN is still in its nascent phase and it is in fact evolving toward a more devops-oriented approach to operationalizing the network. In that respect, John hit on what's key: network provisioning and management must evolve along with operations if it's going to meet the challenges ahead.
While most enterprises may think they don't need webscale infrastructure and networks, if they're going to take advantage of the Internet of Things, they will. The explosion of applications and services that go along with "things" is going to require a lot more scale and stability in the network, and a devops-like approach is one of the ways in which IT can realize both.