vlt/SwD/GSF K8s/K8s Proposal for DC.md

• Benefits

  • simplifies and automates HA
    • EKS control planes are multi AZ by default
    • self healing
    • requires no developer interaction
  • can still use managed services
    • rds
    • redis
      • can deploy a second redis for event streaming if redpanda not suitable
  • reduces number of AWS accounts
    • cost savings, see below
  • cloud agnostic
  • security improvements
    • no need for devs to SSH into machines to debug
  • flux for gitops
  • allows for easy deployments of new OEMs
  • maintains single tenancy with namespaces

• Cost

  • Last Month's Bill:

    PROD

    Total USD 1,079.60 $0.0146/hr t2.micro USD 10.51 $0.20/hr c5a.xlarge USD 144.00 c5a.xlarge Compute Savings Plans USD 144.00 $0.80/hr c5a.4xlarge USD 576.00 $1.60/hr c5a.8xlarge Instance Hour 1,440 Hrs USD 2,304.00 c5a.8xlarge Compute Savings Plans 1,131.817 Hrs

    DANDIS

    Total USD 1,020.67 $0.0238 per t3a.small USD 51.38 $0.0475 per t3a.medium USD 34.15 $0.0695 per r6gd.medium USD 50.04 $0.20 per c5a.xlarge USD 431.60 c5a.xlarge Compute Savings Plans USD 431.60 $0.432 per m6a.2xlarge USD 311.04 $0.80 per c5a.4xlarge USD 575.20 c5a.4xlarge Compute Savings Plans USD 1.14

  • EC2 Cost Estimate

  • utilises servers more

    • DANDIS m6a.2xlarge running at ~10% cpu usually
    • autoscaling
    • fargate possible but annoying

  • EKS

    • dev + prod all we need
    • flat $0.10/hr per cluster
      • 2 Clusters x 0.10 USD per hour x 730 hours per month = 146.00 USD
      • less than a single c5a.xlarge 3 of which in DANDIS account doing nothing
    • In prod instances max 27% cpu across the last week !
      • DANDIS Instance Utilisation !