One of the most fundamental changes in computing occurring today is the shift toward processing and analyzing data at the point where it is being created and consumed. The goal is to deploy applications at the network edge that drive near real-time experiences that increasingly are at the core of a wide range of digital business transformation initiatives.
The definition of what the network edge is, of course, differs by use case but the one thing that is increasingly clear is the applications being deployed all have one thing in common: Kubernetes. The primary reason for this is Kubernetes provides a standard set of application programming interfaces (APIs) that make deploying applications across a wide range of class of processors much simpler. Given all the types of processors employed across an edge computing platform, the layer of abstraction provided by Kubernetes is essential.
Depending on the use case there are three distinct flavors of Kubernetes being employed at the network edge. The first is the full distribution of Kubernetes that, for example, might be deployed on a hyperconverged computing platform that pushes that type of horsepower as close to the point where data needs to be processed as possible.
The second is a lighter weight distribution of Kubernetes that might be alternatively employed, for example, within a point-of sale (PoS) environment or anywhere else that has a limited amount of computing capacity.
The third is endpoints that rather than running any type of Kubernetes cluster are simply taking advantage of the Kubernetes API to make external calls to a cluster at the network edge or Kubernetes resources deployed in the cloud.
Regardless of the use case, organizations that embrace Kubernetes are able for the first time automate the deployment and subsequent updates of applications in a consistent fashion. The challenge is that Kubernetes itself is challenging to manage. A survey of 500 platform teams working in organizations with more than 1,000 employees conducted by Rafay Systems, a provider of a Kubernetes management platform, find 61% reporting that environment provisioning is a major roadblock to accelerating the timeframe for application deployments, with 25% taking three months or longer to deploy a modern application or service.
The survey of both platform engineers and developers in those also finds nearly half (45%) of respondents are not satisfied or just somewhat satisfied with their current process. Issues include having to wait on someone else or on a ticketing-based system to provision environments (57%), the fact that there are too many software/service dependencies between the application and environment that need to be tested/approved/validated (49%), that it takes too long to gain/configure/approve access to new environments (30%) and a lack of automation to procure environments or environments that instead must be manually deployed (27%).
Among the 39% of platform engineers that are unsatisfied, issues include the lack of a standard way to deploy and manage environments (43%), it takes too much time and effort to train development teams on how to provision environments (41%), lack of visibility into environment resources including usage, costs and performance metrics (38%) and lack of governance (35%) and too few guardrails around operations (27%).
Each IT organization will need to decide for themselves to what degree it makes sense to manage instances of Kubernetes at the edge themselves versus relying on the expertise of an external IT services provider. Every dollar applied to managing infrastructure is, after all, one less than can be applied to building and deploying applications.
At the very least, however, each IT team should understand what’s required to many Kubernetes at the edge just in case an issue that leads to a services disruption does arise. The important thing to remember is that just because an IT team knows how to do something it doesn’t always follow that they should be the ones to do it given all the other tasks that there are only so many hours in a day wind up being still left undone.
