Why Tensorflow and Kubernetes? — Secret Sauce
Cause when you listen, you improve.
Let’s begin with a short introduction for those of you who have been living under book racks or server racks.
Tensorflow: A library which helps you make sense of your data
Kubernetes: A platform to help manage your compute resources
So.. what’s so special about them? Let’s break it down.
Both were part of a system which has proven it’s caliber in the field. Borg powers infrastructure for Google, and it is the inspiration and in many ways father of kubernetes. You can expect better, if not similar capabilities from Kubernetes. On the other end, Tensorflow is what is actually being used at Google to improve on existing machine learning capabilities, as well as build some pretty rad new ones.
Yes! Both of them. And that might seem like a small thing, but a key aspect of open source projects it that they thrive on feedback. If you are not getting issues opened in your project, you are already dying. Not only both projects have been regraded well by their own respective communities, they have also embraced the world of open source. Continuous iterations, fully formed feedback loop, ability to question existing methods and features and driving the change from community are just a few reasons that both projects have become the brand for their fields.
Both projects embrace a philosophy. For Kubernetes, it’s the principle that everything that can fail, will fail. So start getting things back into shape. You start from a failed state and then system does it’s best to keep you up into desired state. Everything else is just how you achieve it. On the other hand with Tensorflow, it’s the prenomination that current machine is never enough. So why not use it to build just a flow of your expectations, and then let Tensorflow manage the execution. This enables it to become distributed, which makes everyone’s life easier
Not only by individuals and Google, both projects have gained significant support from the community and other industry leaders alike. Kubernetes is now part of Cloud Native Computing Foundation under Linux Foundation, which aims to simplify and provide better tools for cloud. Tensorflow has generated such an interest internally at Google that they are building Tensorflow Processing Units (TPUs) which are basically hardware optimized to run Tensorflow. Not just Google but Intel and others are also on the mission to help Tensorflow succeed.
Since a project is open source doesn’t mean that you can’t build great services around it. Google has been offering Kubernetes as a service for some time now as Google Container Engine (GKE). There are other offerings as well. Not just for kubernetes, but also for services that work along with it. As for Tensorflow, there’s a Machine Learning platform available by Google Cloud Platform. Alos the amount of machine learning models built on Tensorflow is growing at a rapid rate. It’s also expected to launch an even simpler, higher level API to learn and build models.
I hope this article was insightful. If you still have doubt, why not ask the folks directly?