Design_file_cost_optimisation__1_.pptx.pdf

Transcript

1. Strategies for Thriving in the Cloud: Sustainability and Scalability in a Changing Macroeconomic Environment A $400k/M heist By Sandeep Raghuwanshi Lead DevOps | Razorpay Sandeepraghuwanshi
2. “Last year we have created stacks of infra and scaled it till the moon and Sandeep been a pivotal part of the process, ” Proud Manager Cranking It Up • Team expands and infra creation starts rapidly. • As the consumption peaks scale happens. • At this stage the goal is to serve the requirements. The story begins …
3. Rapid changes?
4. Contemplating about cloud cost?
5. Chaos?
6. • AWS ◦ Compute ◦ Storage ◦ Network ◦ Database • K8S ◦ Over provisioning ◦ Request/Limit on resources ◦ Imbalanced cluster • Monitoring • Logging Bills, Thrills, and Big Bills: Service Expense Categorization
7. Efficiency and Cost Optimization in DevOps AWS K8S Monitoring Logging
8. • High expenditure on data transfer • Cluster Autoscaler & AutoScaling groups • Priority: One AZ (availability zone) • Secondary AZ fallback for availability assurance • 100% service cost visibility • 30% AWS cost reduction • 40% overall cost/txn reduction • Cost governance process implemented • AWS Graviton processor • Arm64 CPU architecture • 25% worker node transition • 25% cost decrease vs. AMD AWS Compute Storage Network Graviton • Scalability with Spot - 50% lower cost than On-demand • Over 60% of production EC2 instances are on Spot • New applications utilize Spot instead of On-demand • GitHub actions running in K8S on spot
9. Efficiency and Cost Optimization in DevOps AWS K8S Monitoring Logging
10. K8S • Goal: Reduce cost and increase instance size • Action items aligned with optimizing resource allocation. • Descheduler implementation • Variety of node type • Weekly review and implementation of CRDs for policy enforcement • Increased number of nodes in the system • Elevated operational costs. • Ephemeral cluster creation. • Night stage shutdown • Effectively manage request vs usage • Get 100% visibility on the changes • Keep auto scaling on requirement basis Bin packing Over Provision Imbalance Tuning R/R • Evaluate existing bin packing algorithm tools (eg: Cast AI, Karpenter) • Assess their performance and capabilities • Make sure every node is well utilized • Keep vertical scaling in mind
12. Bin packing Primary objective: Reduce compute cost • Investigate merging AWS AutoScaling groups •
13. Efficiency and Cost Optimization in DevOps AWS K8S Monitoring Logging
14. Monitoring Choose monitoring based on below • Rapid data retrieval and analysis • Embedded compression algorithm • Cardinality has to be managed while exposing any metrics. • Monitor the Monitoring system • Keep sample size in check • Reliable time series database • Read load segregation to enhance resilience during zone failures
16. Efficiency and Cost Optimization in DevOps AWS Monitoring K8S Logging
17. Logging • Use light weight log exporter (fluentbit > Fluentd) • Keep tab on verbosity of logs(whenever possible switch off the debug or whole trail) • Divide the log tier in Hot , Cold, Archive • Make sure to ship the logs in same zone for curtailing any network cost • Make use of compression algorithms of log shippers
18. Database • DAM stands for Database activity monitoring • Involves observing, identifying and reporting a database’s activities • Database firewall acts as a protective layer for the database • Helps detect and prevent database-specific attacks
19. Thank You