Skip to content

Slash EKS Cluster Costs by 20-30% Instantly with AWS Graviton

If you’re running Kubernetes workloads on Amazon EKS backed by Intel-based instances, you’re leaving significant savings on the table. In this blog, we will look at how many Rafay customers have been able to immediately cut compute costs by ~20-30% with minimal effort and quickly comply with internal cost saving mandates.

Graviton Ready

AWS Graviton Processors

AWS offers 3 processor types for EC2 as well as EC2-backed EKS managed node groups. Customers have the choice of Intel, AMD, and ARM (AWS Graviton) processors.

AWS Graviton processors are designed by AWS to deliver the best price performance for your cloud workloads running in Amazon EC2. Graviton-based instances can be identified by the letter g in the Processor family section of the Instance type naming convention.

Graviton Instance

AWS Graviton processors are built on AWS Nitro. Graviton instances can use almost all of the compute and memory resources of the host hardware. This is achieved by breaking apart the hypervisor functions and management capabilities from the hosts and offloading them to dedicated hardware and software.

Info

Traditional virtualization platforms that run the hypervisor software on the same physical host as the virtual machines cannot utilize 100% of the host’s resources

Migrating to ARM based Graviton

Migrating your EKS workloads to Graviton is easier than you might think. Most modern containerized applications are architecture-neutral. As a result, container images can be rebuilt for ARM with a simple addition to your CI/CD pipeline. Many customers start with non-critical workloads or stateless microservices to test compatibility. Once validated, expanding to production services is straightforward.

Since containers share the host kernel, the code that's running inside the container must be compatible with the host's architecture. This is why you can't run a linux/amd64 container on an arm64 host.Multi-platform builds solve this problem by packaging multiple variants of the same application into a single image.

EKS Cluster Lifecycle using Rafay

Amazon EKS Clusters can be provisioned and managed using Rafay Kubernetes Management with Graviton based node groups. In the image below, the managed node group is based on t4g.xlarge instance type.

Graviton Node Group

EKS Cluster Lifecycle Management can be performed using all the supported automation options: Declarative Cluster Spec using CLI or GitOps or Terraform or REST API.

Rafay In-Cluster Components

Kubernetes Management products such as Rafay require the deployment of in-cluster software components. These components need to be available for ARM64 processors such as AWS's Graviton.

All on-cluster software components from Rafay have also been optimized for AWS Graviton. As you can see from the kubectl command below, both the Rafay Management components and Rafay Services are available for ARM based AWS Graviton nodes. 

kubectl get po -n rafay-system

NAME                                             READY   STATUS    RESTARTS      AGE
cilium-operator-69d6c786fb-d2jvf                 1/1     Running   0             40h
cilium-wr5gp                                     1/1     Running   0             4h47m
controller-manager-v3-7d594549bc-pvd9z           1/1     Running   0             40h
csi-secrets-store-provider-aws-m6pjx             1/1     Running   0             4h47m
edge-client-f5b8d9d79-jl956                      1/1     Running   3 (62m ago)   40h
gatekeeper-audit-c7d76f766-j45qb                 2/2     Running   0             40h
gatekeeper-controller-manager-5b46bcd4c8-pf5nh   2/2     Running   0             40h
gatekeeper-controller-manager-5b46bcd4c8-q6zcx   2/2     Running   0             40h
gatekeeper-controller-manager-5b46bcd4c8-spcnv   2/2     Running   0             40h
rafay-connector-v3-595569c488-2w4gv              1/1     Running   0             40h
secrets-store-csi-driver-6pw7j                   3/3     Running   0             4h47m
v2-relay-agent-78b649f59b-kvft6                  1/1     Running   0             40h

Real Life Customer Example

One of our customers, a Fortune 500 enterprise had a mandate from their leadership to cut cloud costs by 30% in less than 60 days. Approximately, 90% of their modern customer facing apps were deployed across ~100 Amazon EKS clusters managed by the Rafay Platform.

For the last few years, they have been using EKS clusters with nodes based on amd64 architecture based m5.xlarge instance type.

They successfully migrated to AWS Graviton based m6g.xlarge instances, immediately saving ~20% in cloud costs.

In their case, m6g.xlarge was the direct cost-optimized alternative (see comparison table below) to m5.xlarge which is ideal for most general workloads. They are now evaluating m7g.xlarge instance type as well since it offers higher performance, more bandwidth, and support for more compute-intensive workloads while remaining cheaper than m5.xlarge.

Feature m5.xlarge m6g.xlarge m7g.xlarge
Processor Architecture Intel Xeon Platinum 8175 (x86_64) AWS Graviton2 (Arm Neoverse N1, arm64) AWS Graviton3 (Arm Neoverse V1, arm64)
vCPUs 4 4 4
Memory 16 GiB 16 GiB 16 GiB
Networking Bandwidth Up to 10 Gbps Up to 10 Gbps Up to 12.5 Gbps
EBS Bandwidth Up to 2,120 Mbps Up to 4,750 Mbps Up to 10,000 Mbps
Clock Speed 2.5 GHz (base) ~2.5 GHz equivalent (estimated) ~2.6–3.0 GHz (estimated)
Price (us-east-1) $0.192/hour $0.154/hour $0.1728/hour
Price Difference — ~20% cheaper ~10% cheaper
Performance per Dollar Baseline Up to 40% better than m5 Up to 25% better than m6g
Instruction Set x86_64 Armv8.2 Armv9
Ideal For General-purpose workloads Cost-optimized general workloads, containers Higher-performance ARM workloads, ML inference, microservices

The customer found the remaining 10% in cost savings by (a) right sizing their containerized applications and (b) automatically scaling down their pre-prod clusters to zero based on a schedule controlled using Rafay's Environment Manager.

Conclusion

The switch to AWS Graviton for your EKS clusters can be a quick win. With rising cloud spend scrutiny, this architectural upgrade offers instant ROI, improved performance, and better sustainability metrics.

Ready to save ~30% on your next AWS bill?

Rebuild your container images for ARM, update your Amazon EKS node groups, and let Rafay and Graviton do the rest.