
Unofficial EKS Cluster Creation Performance
Creating a Kubernetes cluster on Amazon EKS (Elastic Kubernetes Service) is a crucial first step for many cloud-native applications. While the process is straightforward, I was curious about how long it takes to provision these clusters across different AWS regions. If you are interested to the EKS cluster creation performance improvement, in this analysis, let’s explore cluster creation performance across all major AWS regions.
Overview
I conducted extensive testing across 21 AWS regions using identical configurations to measure and compare cluster creation times. The results reveal interesting variations in regional performance. This unofficial analysis tracks EKS control plane performance improvements but does not suggest deployment preferences.
Regional Performance Analysis
Region Name | Region | Creation Time (seconds) | Approx. time (mm:ss) |
---|---|---|---|
Europe (London) | eu-west-2 | 792.85 | 13 minutes 13 seconds |
Asia Pacific (Seoul) | ap-northeast-2 | 834.38 | 13 minutes 54 seconds |
Asia Pacific (Sydney) | ap-southeast-2 | 840.96 | 14 minutes 01 seconds |
Europe (Paris) | eu-west-3 | 854.17 | 14 minutes 14 seconds |
South America (São Paulo) | sa-east-1 | 860.00 | 14 minutes 20 seconds |
Asia Pacific (Tokyo) | ap-northeast-1 | 893.76 | 14 minutes 54 seconds |
Asia Pacific (Singapore) | ap-southeast-1 | 897.10 | 14 minutes 57 seconds |
US West (Oregon) | us-west-2 | 899.90 | 15 minutes 00 seconds |
US East (Ohio) | us-east-2 | 904.62 | 15 minutes 05 seconds |
Europe (Ireland) | eu-west-1 | 912.86 | 15 minutes 13 seconds |
Canada (Central) | ca-central-1 | 913.77 | 15 minutes 14 seconds |
Europe (Frankfurt) | eu-central-1 | 916.08 | 15 minutes 16 seconds |
Europe (Stockholm) | eu-north-1 | 919.79 | 15 minutes 20 seconds |
Africa (Cape Town) | af-south-1 | 936.71 | 15 minutes 37 seconds |
Asia Pacific (Mumbai) | ap-south-1 | 965.44 | 16 minutes 05 seconds |
Asia Pacific (Hong Kong) | ap-east-1 | 968.06 | 16 minutes 08 seconds |
US West (N. California) | us-west-1 | 970.99 | 16 minutes 11 seconds |
Europe (Milan) | eu-south-1 | 986.76 | 16 minutes 27 seconds |
US East (N. Virginia) | us-east-1 | 1232.10 | 20 minutes 32 seconds |
Asia Pacific (Osaka) | ap-northeast-3 | 1277.37 | 21 minutes 17 seconds |
Middle East (Bahrain) | me-south-1 | 1297.92 | 21 minutes 38 seconds |
Key Findings
- The average cluster creation time across all regions is approximately 15 minutes
- Fastest cluster creation is in Europe (London) at ~13 minutes, while Middle East (Bahrain) is at ~21.6 minutes
- European regions generally show shorter creation time, with 4 out of 6 European regions completing in under 15.5 minutes
- Notable performance gap between fastest (792.85s) and slowest (1297.92s) regions - approximately 8.5 minutes difference
- US East (N. Virginia) are taking a bit longer than expected - around 20.5 minutes to get everything up and running
- Asia Pacific regions show mixed results, with Seoul performing well (834.38s) but Osaka being second-slowest (1277.37s)
How was this measured?
The measurements were conducted using eksctl, a popular command-line tool for creating and managing EKS clusters. To ensure consistency, we ran multiple concurrent tests during similar time periods, using identical configurations across all regions. Each test created a Fargate-based cluster using the command shown below.
eksctl create cluster --name {cluster_name} --region {region} --fargate
We chose AWS Fargate for these tests because it provides a more controlled and consistent environment for measuring cluster creation times. Unlike EC2-based nodes, which can be affected by instance availability and capacity constraints across regions, Fargate’s serverless nature eliminates these variables from our measurements. This allows us to focus specifically on EKS control plane provisioning performance across regions and ensure proper setup of critical components such as CoreDNS.
Conclusion
Based on the performance data collected across AWS regions, there are significant variations in EKS cluster creation times that appear to correlate with regional infrastructure. This analysis reveals a performance spectrum ranging from approximately 13 minutes to 21 minutes.
This is a general test that simulates a basic running cluster. Production deployments should consider additional factors—such as node group configurations, add-ons, and specific networking requirements—all of which can affect cluster creation times1. Keep in mind that these measurements represent baseline performance, which may fluctuate based on AWS service load and regional infrastructure updates.
References