DynamoDB Cost Optimization Masterclass: Deep Dive on DynamoDB Pricing
0 likes172 views
The document provides a comprehensive overview of Amazon DynamoDB's pricing model, highlighting its predictability, flexibility, and integration with resource usage. Key elements discussed include the core pricing units (read and write capacity units), billing modes, and strategies for managing costs effectively, such as auditing usage and limiting resource-heavy features. Ultimately, the insights aim to guide developers in optimizing their application architecture based on DynamoDB's pricing structure.
DynamoDB Cost Optimization Masterclass: Deep Dive on DynamoDB Pricing
- 1. Deep Dive on DynamoDB Pricing Alex DeBrie Principal, DeBrie Advisory
- 2. Why I love DynamoDB's pricing.
- 4. DynamoDB's pricing … ■ is predictable.
- 5. Predictable pricing ■ "able to reason about in advance" ■ Can reason about your costs before building ■ Can see the impact of changes
- 6. DynamoDB's pricing … ■ is predictable.
- 7. DynamoDB's pricing … ■ is predictable. ■ is flexible.
- 8. Flexible pricing ■ Can scale capacity up + down to match traffic ■ Can switch between modes You get all this with world-class reliability + availability
- 9. DynamoDB's pricing … ■ is predictable. ■ is flexible.
- 10. DynamoDB's pricing … ■ is predictable. ■ is flexible. ■ has a tight feedback loop.
- 11. Tight feedback loop ■ Understand what's driving your costs ■ Adjust as needed ■ Tight integration between resource usage + your costs
- 12. DynamoDB's pricing should change how you build applications.
- 13. Outline ■ Why I love DynamoDB's pricing model ■ Basics of DynamoDB pricing ■ Tips for managing DynamoDB costs
- 14. Who am I? Alex DeBrie AWS Data Hero Author, The DynamoDB Book [email protected] @alexbdebrie
- 15. Basics of DynamoDB Pricing
- 16. Basics of DynamoDB pricing ■ The core units ■ Macro factors ■ Micro factors
- 17. The core units: RCUs + WCUs ■ Pay for reads and writes directly rather than resources ○ Traditional databases: CPU, RAM, IOPS ○ DynamoDB: RCUs and WCUs ■ Read Capacity Units (RCUs) → up to 4KB of data read ■ Write Capacity Units (WCUs) → up to 1KB of data written ■ Storage: priced per GB-month Billing is (mostly) independent from performance!
- 21. The core units: RCUs + WCUs ■ Pay for reads and writes directly rather than resources ○ Traditional databases: CPU, RAM, IOPS ○ DynamoDB: RCUs and WCUs ■ Read Capacity Units (RCUs) → up to 4KB of data read ■ Write Capacity Units (WCUs) → up to 1KB of data written ■ Storage: priced per GB-month Billing is (mostly) independent from performance!
- 22. Basics of DynamoDB pricing ■ The core units ■ Macro factors ■ Micro factors
- 23. Basics of DynamoDB pricing ■ The core units ■ Macro factors ■ Micro factors
- 24. The macro: DynamoDB modes ■ DynamoDB bill == operations + storage. ■ Billing modes ○ Provisioned throughput → RCUs & WCUs available on a per-second basis ○ On-demand mode → Charged per request ■ Table storage class ○ Standard → Cheaper operations but more expensive storage ○ Standard-IA → Cheaper storage but more expensive operations
- 25. Basics of DynamoDB pricing ■ The core units ■ Macro factors ■ Micro factors
- 26. Basics of DynamoDB pricing ■ The core units ■ Macro factors ■ Micro factors
- 27. The micro: WCU + RCU multipliers ■ Item size ■ Secondary indexes ■ Transactions ■ Global tables ■ Consistent Read ■ Query vs. BatchGet
- 28. Tight feedback loop ■ Understand what's driving your costs ■ Adjust as needed ■ Tight integration between resource usage + your costs
- 29. The micro: WCU + RCU multipliers ■ Item size ■ Secondary indexes ■ Transactions ■ Global tables ■ Consistent Read ■ Query vs. BatchGet ← Larger items require more resources ← More writes ← Coordination ← Requests to more storage nodes ← Requires routing to leader ← Replication infra + conflict resolution
- 30. Basics of DynamoDB pricing ■ The core units ■ Macro factors ■ Micro factors
- 32. Tips for managing DynamoDB costs ■ Mind your multipliers!
- 33. The micro: WCU + RCU multipliers ■ Item size ■ Secondary indexes ■ Transactions ■ Global tables ■ Consistent Read ■ Query vs. BatchGet
- 34. Mind your multipliers ■ Review item sizes carefully ○ Remove unused attributes ○ Compress large values (or send to S3) ○ Reduce attribute names ○ Split items if needed
- 35. Mind your multipliers ■ Review item sizes carefully
- 36. Mind your multipliers ■ Review item sizes carefully ■ Limit secondary indexes ○ Do you need an index? Writes are ~20x more than reads. ○ Use projections to limit size of items in index
- 37. Mind your multipliers ■ Review item sizes carefully ■ Limit secondary indexes
- 38. Mind your multipliers ■ Review item sizes carefully ■ Limit secondary indexes ■ Limit transactions
- 39. Mind your multipliers ■ Review item sizes carefully ■ Limit secondary indexes ■ Limit transactions ■ Ensure you need Global Tables
- 40. Mind your multipliers ■ Review item sizes carefully ■ Limit secondary indexes ■ Limit transactions ■ Ensure you need Global Tables ■ Don't use consistent reads
- 41. Tips for managing DynamoDB costs ■ Mind your multipliers!
- 42. Tips for managing DynamoDB costs ■ Mind your multipliers! ■ Audit your usage
- 43. The macro: DynamoDB modes ■ DynamoDB bill == operations + storage. ■ Billing modes ○ Provisioned throughput → RCUs & WCUs available on a per-second basis ○ On-demand mode → Charged per request ■ Table storage class ○ Standard → Cheaper operations but more expensive storage ○ Standard-IA → Cheaper storage but more expensive operations
- 44. Audit your usage ■ Billing mode
- 45. Choosing a billing mode ■ On-Demand costs 6.94x fully-utilized Provisioned Capacity ○ But: full utilization is hard to achieve! ○ Requires >14.4% utilization on average ■ My advice: ○ Use On-Demand until it hurts. ○ If using Provisioned, ensure you're hitting >14.4% utilization. ○ Add Auto-Scaling with caution ○ Explore Reserved Capacity
- 46. Audit your usage ■ Billing mode
- 47. Audit your usage ■ Billing mode ■ Table storage class
- 48. Choosing a table storage class ■ Indifferent whenever Operations costs are 2.4x Storage costs ○ If >2.4x, then use Standard storage ○ If <2.4x, then use Standard-IA ■ Standard storage will (almost) always be cheaper at first ○ Writing 1KB == 5 months of storage ○ But: this shifts over time ■ Audit this monthly! ■ Use TTL to expire data
- 49. Audit your usage ■ Billing mode ■ Table storage class
- 50. Audit your usage ■ Billing mode ■ Table storage class ■ Review secondary indexes
- 51. Audit your usage ■ Billing mode ■ Table storage class ■ Review secondary indexes ○ Does my secondary index have any reads?
- 52. Audit your usage ■ Billing mode ■ Table storage class ■ Review secondary indexes ○ Does my secondary index have any reads? ○ What is the write:read cost ratio on my secondary index?
- 53. Audit your usage ■ Billing mode ■ Table storage class ■ Review secondary indexes
- 54. Tips for managing DynamoDB costs ■ Mind your multipliers! ■ Audit your usage
- 55. Tips for managing DynamoDB costs ■ Mind your multipliers! ■ Audit your usage ■ Run the numbers
- 56. DynamoDB's pricing should change how you build applications.
- 57. The micro: WCU + RCU multipliers ■ Item size ■ Secondary indexes ■ Transactions ■ Global tables ■ Consistent Read ■ Query vs. BatchGet ← Larger items require more resources ← More writes ← Coordination ← Requests to more storage nodes ← Requires routing to leader ← Replication infra + conflict resolution
- 58. DynamoDB's pricing should change how you build applications.
- 59. Keep in touch! Alex DeBrie AWS Data Hero Author, The DynamoDB Book [email protected] @alexbdebrie