GKE

GKE is an offering from the company Google that can be rented. You upload your own program, and Google has it run on its computers. The name stands for Google Kubernetes Engine. Kubernetes is software that organizes many programs on many computers at the same time. It decides which program runs on which computer, restarts it after a crash, and adds extra copies when there is high demand. GKE takes over the work of setting up and maintaining this software yourself.

Why companies no longer run their own servers

In the past, a company bought its own servers and put them in a basement. If the number of users grew, someone had to order and install new devices. That took weeks and cost a lot of money upfront. If a device failed at night, an employee had to drive there.

With GKE, you rent computing power by the hour and pay only for what you use. An online shop can multiply its capacity tenfold before the Christmas season and reduce it again in January. That happens via a configuration file, not a delivery truck. For start-ups, this is crucial, because it lets them run an application for millions of users without large amounts of starting capital.

Competition also matters here: GKE is Google’s answer to similar offerings from Amazon and Microsoft. This cloud business, meaning the renting out of computing power over the internet, is among the most profitable divisions of these corporations. It therefore regularly appears as its own line item in quarterly earnings reports.

Containers, pods, and the Autopilot

The smallest unit in GKE is a container. This is a package that contains the program along with everything it needs to run. You can picture it like a shipping container: the contents don’t matter, the outer dimensions are always the same. That’s why the same package runs identically on every computer, without anyone needing to make adjustments.

Kubernetes is given only a target by the operator, for example: five copies of this container should always be running. From then on, it constantly compares the actual state with this target. If a copy crashes, it starts a new one within seconds. If an entire computer fails, it distributes that computer’s tasks among the others. This principle is called self-healing and is the core of the whole thing.

GKE comes in two variants. In Standard mode, you choose yourself how many and how powerful the computers running in the background are. In Autopilot mode, Google also takes over this decision and bills only for actual usage. A common misconception is that GKE is itself the software Kubernetes. GKE is the management service built around it; Kubernetes is a freely available open-source project that anyone could also install on their own.

From streaming services to AI data centers

As a user, you never see GKE directly. You only notice it in that an app still works even when very many people are accessing it at the same time. Banks run their apps on it, as do streaming providers and online retailers. At most, the service becomes visible in the IT department’s bill.

In the news, GKE mainly comes up in connection with artificial intelligence. AI models need special chips, so-called GPUs, which were originally developed for graphics. Through GKE, hundreds of such chips can be linked together into a cluster and released again. Companies that want to train their own model rent this capacity instead of building a data center.

Anyone interested in computer science will sooner or later come across this topic in job postings as well. Job titles like DevOps Engineer or Site Reliability Engineer describe people who look after exactly such systems. Knowledge of Kubernetes is now among the most frequently requested skills there.