cloud-test: Open Source + Hosted Containers: A recipe for workload mobility

Google Cloud Platform Blog

Open Source + Hosted Containers: A recipe for workload mobility

Monday, November 10, 2014

we announcedKubernetes

Develop and perform scale out testing in the cloud, but deploy to an on-premises production data center. A huge benefit to many of our customers is being able to do high throughput scale out testing of an application on resources that are paid for by the minute because it reduces iteration time and improves team productivity. Because Google Compute Engine is billed in minute quanta, the incremental cost of accelerated scale out testing is low. You pay what you would for sequential testing-- it just happens on more cores and finishes much more quickly. This only works if the framework that runs your application is available in both your production and test environment. It also helps to have a management framework that makes it easy to deploy, orchestrate and wire together individual tests. Google Container Engine with Docker containers provide a framework that supports the easy deployment and management of an app, and also lets you easily integrate test management and orchestration frameworks.

Migrate a new piece of an application to the cloud, but have parts of it stay on-premises. With Google Cloud Platform’s newly announced direct peering and carrier interconnect network features, it’s now easier to connect a part of an application deployed in the cloud to Google’s data centers with the on-premises parts. With 70+ peering locations in 33 countries, it’s possible to get unprecedented levels of throughput and low latency access to your cloud resources. Many of our customers also highly value a common toolchain and management paradigm, as it makes sense to build modern applications using the same tools, packaging format and management services, but let the pieces that need to stay on-premises remain there.

Burst to the cloud during peak load. The cloud offers the ability to quickly and easily spin up a large number of VM instances that are charged on a per minute basis. Compute Engine instances tend to boot in around 30 seconds, giving our customers the ability to react quickly to unexpected demand spikes.

Container based. Docker has created a highly portable application container framework and is committed to the vision of making it run everywhere. The natural decoupling of application pieces from the OS and infrastructure environment is a really important ingredient in achieving high levels of portability.

Modular to the core. To become broadly adopted, it was important to allow providers to adapt and extend pieces of the stack without invalidating the core API. We focused on rigorous and principled modularity in the design, and pretty much everything in Kubernetes can be unplugged and replaced by other technologies.

Decoupled services. A key insight into what allows portability and mobility is the idea that different pieces of an application may be moved to a different cloud at different times. Kubernetes is built with a focus on micro-services based architecture and ensuring that the pieces of an application are not tied together. The beauty of Kubernetes is that its naturally decoupled model creates the feeling like the pieces are co-deployed. You don’t have to jump through hoops to get a decoupled deployment.