Introducing the next evolution of container platforms
Chen Goldberg
GM & VP, Cloud Runtimes
Google Cloud has been on a mission to be the best place to run containerized workloads. This started back in 2014, when, inspired by Google’s internal cluster management system, Borg, we invented Kubernetes and introduced Google Kubernetes Engine (GKE), the first managed Kubernetes service in the world. GKE is the most scalable leading Kubernetes service available in the industry today1. In 2019, we launched Cloud Run, the first serverless platform to combine the benefits of containers and serverless. Today, Cloud Run provides one of the leading developer experiences amongst all cloud providers. We also extended GKE to hybrid and multi-cloud environments with Anthos in 2019, and introduced Autopilot mode in GKE in 2021. Finally, this year, we expanded Anthos’ reach with Google Distributed Cloud. Continuing on our mission, this year at Next, we are announcing three enhancements to our container management products:
First, we are launching GKE Enterprise, a new premium edition of GKE. With GKE Enterprise, companies can increase velocity across multiple teams, easily and securely run their most important business-critical workloads, and reduce total cost of ownership with a fully integrated and managed solution from Google Cloud.
Second, for organizations developing the next generation of AI applications, GKE now supports the AI-optimized Cloud TPU v5e. In addition, support for both the A3 VM with NVIDIA H100 GPU as well as Cloud Storage FUSE are now generally available for GKE.
Third, for platform teams that want to leverage the power of generative AI to drive productivity, Duet AI in GKE and Cloud Run provide gen AI assistance specifically trained on our documentation to cut down on time it takes to run containerized applications.
GKE Enterprise: the next evolution of Kubernetes
GKE Enterprise builds on Google Cloud’s leadership in containers and Kubernetes, bringing together the best of GKE and Anthos into an integrated and intuitive container platform, with a unified console experience.
GKE Enterprise edition includes a new multi-cluster feature (“fleets”) that lets platform engineers easily group similar workloads into dedicated clusters, apply custom configurations and policy guardrails per fleet, isolate sensitive workloads, and even delegate cluster management to other teams. GKE Enterprise comes with managed security features, including advanced workload vulnerability insights, governance and policy controls, and managed service mesh — all based on the best of the Kubernetes open-source ecosystem. And because GKE Enterprise is a fully integrated and fully managed platform, with a simple, intuitive, in-context observability dashboard, customers spend less time and effort managing the platform and more time creating amazing apps and experiences for their customers. Plus, GKE Enterprise includes hybrid and multi-cloud support so you can run container workloads anywhere — on GKE, in other public clouds, or on-premises with Google Distributed Cloud.
In short, GKE Enterprise makes it faster and safer for distributed teams to run even their more business-critical workloads at scale, without growing costs or headcount. In fact, GKE Enterprise is producing amazing results with customers, improving their productivity by 45%, while reducing software deployment times by over 70%.2
Credit reporting provider Equifax has 14,000 employees around the world, and uses GKE to run data and analytics applications that are critical to its operations. As an early adopter of the new multi-cluster and multi-team capabilities in GKE Enterprise, Equifax is excited about the security posture and improved efficiency that it brings to their table.
“Google Kubernetes Engine is the foundation of Equifax's global data fabric and helps Equifax customers around the world live their financial best. With GKE Enterprise edition, we can efficiently manage hundreds of clusters using fleets to ensure operational consistency everywhere. GKE Enterprise edition has allowed us to scale rapidly with strong security and governance controls and meet our customer's service level requirements while keeping costs down.” - Vipul Mapara, Equifax Fellow and SRE Leader, Equifax
GKE Enterprise edition will be available in preview in early September. To enable its capabilities, please reach out to your account manager. Also, you can work with our launch partners Accenture, CDW, Deloitte, DoiT International, SADA, Searce, and 66 degrees to get started.
TPU support in GKE: a catalyst for AI success
Almost every organization is either already using, or plans to use, AI to accelerate their business. The tremendous growth in machine learning is reflected in how customers are using our products: Today, the 15 largest GKE customers are already using it to power their AI workloads. In fact, over the last year, the use of GPUs with GKE has doubled.
As organizations develop and deploy larger and more capable AI models, they need more compute power and more cost-efficient AI accelerators. The new Cloud TPU v5e can scale to tens of thousands of chips, making it ideal for developing more complex AI models. Cloud TPU v5e achieves up to 2x higher training performance and up to 2.5x higher inference performance per dollar for Large Language Models (LLMs) and gen AI models compared to Cloud TPU v4. Running Cloud TPU workloads on GKE enables you to leverage the robust features that many of our most successful customers rely on such as autoscaling, workload orchestration and support for up to 15,000 node clusters.
Grammarly provides AI writing assistance at no charge powered by Google Cloud, and has begun testing TPUs alongside GKE.
“In our research on large language model alignment, Grammarly utilized the power of Google Cloud, TPUs, and JAX. We were impressed by the remarkable performance, robustness, and reliability of the platform, which outperformed many similar offerings we evaluated.” - Max Gubin Engineering Director, Intelligence, Grammarly
Running your workloads in GKE helps save on valuable compute cycles by scaling up when demand rises and scaling down when demand falls. You only pay for the TPU resources you've provisioned, so GKE makes it easy to delay provisioning TPUs until they're needed and shut them down just as easily.
In addition to support for TPUs, GKE is adding GA support for A3 VM with NVIDIA H100 GPU, which can be perfect if you’re training large models.
Finally, Google Cloud Storage FUSE is generally available on GKE. So if your workloads fetch unstructured data – perhaps TensorFlow, PyTorch, Ray, or Spark workloads – now you can move those workloads to GKE without changing how you access your data.
Duet AI in GKE and Cloud Run: productivity in a bottle
The demand for cloud skills vastly exceeds available talent. At Google Cloud, we aim to help your operations and platform engineering team unlock productivity growth and work on the most impactful ideas. Earlier this year, we introduced Duet AI in Google Cloud – your always-on AI collaborator powered by Google’s state-of-art gen AI foundation models – to help Google Cloud users accomplish tasks more effectively and efficiently. And today, we are excited to be introducing Duet AI in our runtimes, including GKE and Cloud Run. Duet AI helps platform teams running their containers on Google Cloud reduce much of the manual, repetitive work they encounter on a day-to-day basis. Duet AI in GKE and Cloud Run is available in preview.
Enabling a container-first world
We're committed to providing the best place to run containers. These new additions built on recent innovations make it easier for customers to scale new workloads with containers, whether with Cloud Run or GKE, or both. For example, with recent enhancements such as the Cloud Run-Eventarc integration, news media outlets like the BBC use Cloud Run to handle dramatic traffic spikes, scale from 150 - 200 container instances to over 1,000 in under a minute, and entertain over 498 million adults per week. We're also focused on giving customers options. Some organizations move workloads between GKE and Cloud Run. Other organizations like Carrefour leverage both GKE and Cloud Run together to run new ecommerce apps. Finally, for telcos like Orange with regulatory and sovereignty requirements, we’ve extended GKE to hybrid environments with Google Distributed Cloud.
We’ve come a long way since 2014, from powering web applications to enabling customers to leverage the cloud and containers to run AI-driven, business-critical applications that transform their businesses. Today, more than ever before, there are infinite possibilities for your workloads, and we can't wait to partner with you on your digital journey.
1. As of August, 2023.
2. Source: “The Total Economic Impact™ Of GKE Enterprise”, a Forrester Consulting study commissioned by Google.
