Success Story
Alephee migrates from Azure to AWS with no data loss and sustains continuous FinOps with ~50% fewer EKS instances
Project overview
Alephee is a B2B/B2C software company for e-commerce and marketplace integrations, with a high volume of sales and integrations processed on its platform. Craftech supported it across three stages: the migration from Azure to AWS, continuous FinOps work and the modernization to AWS Graviton and .NET 8. The migration was completed with no data loss or serious disruptions for users. After moving workloads to Karpenter, the platform needs ~50% fewer instances.
~50%
fewer instances needed after moving workloads to Karpenter
~USD 850/month
saved in DEV during the first cleanup phases
~USD 565/month
saved in PROD during the first cleanup phases
+USD 720/month
in direct savings from tuning IOPS in MongoDB Atlas
Alephee is a B2B/B2C software company for e-commerce and marketplace integrations, with a high volume of sales and integrations processed on its platform. The work with Craftech was organized in three stages, between August 2023 and June 2026.
Solution implemented
- Staged migration plan: pre-production, management and production
- AKS clusters and Docker Swarm workloads migrated to Amazon EKS
- Azure App Services replaced by AWS Elastic Beanstalk with automated Blue/Green deployments
- SQL Server on VMs migrated to Amazon RDS for SQL Server with AWS DMS and native backups to S3
- Azure Blob Storage replaced by Amazon S3 with Rclone and AWS DataSync
- Provisioning with Terraform and Azure DevOps pipelines adapted to deploy to AWS
- Karpenter v1 on Amazon EKS and workers and consumers moved to Spot Instances
- Automatic nightly shutdown of DEV/Test environments with AWS Lambda and AWS SAM
- CPU and memory rightsizing with Goldilocks and Kubecost, and database optimization
- Multi-architecture CI/CD that publishes amd64 and arm64 images to Amazon ECR
- Modernization of C# applications to .NET 8 with AWS Transform
- Karpenter NodePools for Graviton instances and new observability with InfluxDB v3, AWS Lambda and Amazon S3
Business Challenge
Alephee's Azure infrastructure was fragmented, expensive and hard to scale, with scattered technologies (AKS, Docker Swarm, App Services and VMs running SQL Server) that created high maintenance overhead. It had to be modernized toward a cloud-native model on AWS while keeping the platform fully available. Once on AWS, new challenges emerged: high monthly costs from idle resources, oversized instances and environments running 24/7, and later x86/AMD64 and .NET 6 architectures nearing end of life, with telemetry that did not scale.
Solution
Craftech tackled the project in three stages. First, it migrated the platform from Azure to AWS in stages, with all provisioning in Terraform. Next, it sustained continuous FinOps work with Karpenter, Spot Instances, nightly environment shutdowns, rightsizing and database optimization. Finally, it modernized the workloads toward AWS Graviton instances and .NET 8, with multi-architecture CI/CD and a new observability layer.
Outcomes
- Successful migration with no data loss or serious disruptions for users
- Architecture consolidated on AWS with better security, resilience and scalability
- CI/CD 100% integrated with the existing repositories
- ~USD 850/month saved in DEV and ~USD 565/month in PROD during the first cleanup phases
- ~50% fewer instances needed after moving workloads to Karpenter
- More than USD 720/month in direct savings from tuning IOPS in MongoDB Atlas
- Multi-architecture pipelines
- The heaviest workloads run on Graviton EKS instances, with better price/performance than x86
- Historical e-commerce metrics retained for more than 30 days without compromising cost or memory
Stage 1 — Migration from Azure to AWS (August 2023 – May 2024)
The Azure infrastructure was fragmented, expensive and hard to scale, and the platform had to stay fully available during the modernization. Craftech carried out a staged migration plan (pre-production, management and production): it moved the AKS clusters and Docker Swarm workloads to Amazon EKS, replaced Azure App Services with AWS Elastic Beanstalk with automated Blue/Green deployments, migrated SQL Server on VMs to Amazon RDS for SQL Server with AWS DMS and native backups to S3, and replaced Azure Blob Storage with Amazon S3 using Rclone and AWS DataSync. All provisioning was done with Terraform, and the Azure DevOps pipelines were adapted to deploy to AWS.
- Successful migration with no data loss or serious disruptions for users
- Architecture consolidated on AWS with better security, resilience and scalability
- CI/CD 100% integrated with the existing repositories
Stage 2 — Continuous FinOps (October 2023 – February 2026)
After stabilizing on AWS, monthly costs were high due to idle resources, oversized instances and development and test environments running 24/7. The bill had to come down without affecting performance or availability. Craftech implemented Karpenter v1 on Amazon EKS and moved workers and consumers to Spot Instances; automated the nightly shutdown of DEV/Test environments (Elastic Beanstalk, RDS and EKS) with AWS Lambda and AWS SAM; tuned CPU and memory requests and limits with Goldilocks and Kubecost; and optimized the databases: it removed unused clusters, reduced instance types and set MongoDB Atlas IOPS back to their default values.
- ~USD 850/month saved in DEV and ~USD 565/month in PROD during the first cleanup phases
- ~50% fewer instances needed after moving workloads to Karpenter
- More than USD 720/month in direct savings from tuning IOPS in MongoDB Atlas
Stage 3 — Modernization to AWS Graviton and .NET 8 (December 2025 – June 2026)
The x86/AMD64 and .NET 6 architectures were nearing end of life, and telemetry did not scale: noisy alerts and little visibility into the real performance of the e-commerce consumers. Craftech built multi-architecture CI/CD with multi-stage Dockerfiles and docker buildx on Azure DevOps to publish amd64 and arm64 images to Amazon ECR; assessed and modernized the C# applications to .NET 8 with AWS Transform; created Karpenter NodePools for Graviton instances and moved a first batch of workloads (Node.js frontends and .NET APIs) to ARM64; and set up new observability with InfluxDB v3, AWS Lambda and Amazon S3 in Parquet format for massive event ingestion, with centralized dashboards.
- Multi-architecture pipelines
- The heaviest workloads run on Graviton EKS instances, with better price/performance than x86
- Historical e-commerce metrics retained for more than 30 days without compromising cost or memory
Stack & technologies
