1 of 15

Unleashing Cost Optimization : Freshworks' Journey from Startup to Scale

Sravya Mallapu

Lead Software Engineer - SRE

Freshworks

2 of 15

Growth phase : Startup to Scale

3 of 15

Growth phase : Startup to Scale

Single stack architecture

4 of 15

Growth phase : Startup to Scale

Blue Green architecture

5 of 15

Growth phase : Startup to Scale

  • Partial bootup / Shutdown
  • Reserved Instances
  • Traffic Switch phases
  • Time based instances
  • Test switch
  • Canary / Partial switch
  • Full Switch

A Reserved Instance is a pricing model that allows customers to reserve specific compute capacity in the AWS cloud for a one- or three-year term. By making this commitment, customers receive a significant discount compared to the pay-as-you-go, On-Demand pricing for the same instances.

6 of 15

Navigating the Transition Phase : Agile Solutions

  • Resource tagging
  • Time based instances
  • Spot adoption
  • Cloud Custodian policies
  • own:primary
  • own:team
  • svc:name

auto-ops:auto-start-time, auto-ops:auto-stop-time auto-ops:weekend-shut

Spot is a pricing model where customers can bid on unused or spare computing resources. The pricing in the Spot model is based on the concept of supply and demand, and it typically offers significant cost savings compared to other pricing models like On-Demand Instances.

Cloud Custodian is an open-source used for managing and ensuring compliance of cloud resources. It provides a simple and flexible way to define policies that govern the behavior of cloud resources.

7 of 15

Infrastructure Improvements : Best Practices and Strategies

Better Resource Utilisation

Fine Tuned Resource Allocations

Native Auto-scaling Capabilities

40% Compute savings

Bin Packing

8 of 15

Infrastructure Improvements : Best Practices and Strategies

  • AWS private link and VPC endpoints
  • Inter product / platform communication
  • Zone aware routing
  • Log optimization
  • GP2 to GP3 migration
  • RDS Optimisations

9 of 15

Cost Control and Proactive Monitoring

  • Budget utilisation reports
  • Cost anomaly detection

10 of 15

Cost Control and Proactive Monitoring

11 of 15

Cost Control and Proactive Monitoring

  • Cloud Custodian Policies
    • Underutilised EC2 Instances/EBS/RDS/NAT/LB
    • Unattached EBS volumes and long living EBS/RDS snapshots
    • S3 lifecycle policies
    • Unused DynamoDB and Backup policy
    • Un-encrypted EBS/S3/RDS
    • Cloudwatch retention policies
    • Mandatory tags

12 of 15

Cost Control and Proactive Monitoring

13 of 15

Cost Control and Proactive Monitoring

14 of 15

Looking Ahead : Future Initiatives

  • Graviton and Intel Instances
  • SQS Long Polling & Kafka
  • Cell based architecture
  • Spot adoption for production
  • Karpenter

15 of 15

Thank you