Amazon DynamoDB Read / Write Capacity Modes | Overview.

Amazon DynamoDB Read / Write Capacity Modes - Overview. 

Scope:

• Intro,
• Provisioned Capacity Mode,
• On-Demand Capacity Mode,
• Best for predictable workloads.
• When to use what,
• Key Differences Table,
• Key takeaway.

Intro:

• In Amazon DynamoDB, Read/Write Capacity Modes determine how twtech pays for throughput and how DynamoDB allocates resources to serve its reads and writes attributes.
• There are two main modes , plus an optional on-demand backup/restore & autoscaling layer that can be added on top.

1. Provisioned Capacity Mode

• twtech explicitly configures how many read and write units it wants for the table (or GSI).

Best for predictable workloads.

• Read Capacity Units (RCUs) → 1 strongly consistent read per second for an item up to 4 KB, or 2 eventually consistent reads per second for 4 KB.
• Write Capacity Units (WCUs) → 1 write per second for an item up to 1 KB.
• Auto Scaling (optional) → DynamoDB can automatically adjust RCUs/WCUs based on load.
• Over-provision risk → twtech might pay for unused capacity if traffic is lower than expected.
• Under-provision risk → twtech may get throttled if requests exceed provisioned RCUs/WCUs.

 When to use: 

• Predictable traffic patterns, steady baseline with occasional spikes handled by auto scaling.

2. On-Demand Capacity Mode

• twtech doesn’t predefine RCUs/WCUs. 
• DynamoDB instantly accommodates the workload.

Best for unpredictable or new workloads.

• twtech pay per request:
• Read request units (RRUs)
• Write request units (WRUs)
• Instantly scales up and down with traffic.
• No throttling (except in rare, extreme, sustained spikes).
• No need to manage capacity settings.
• Typically more expensive at high sustained throughput than provisioned.

 When to use: 

• Unpredictable traffic, new apps without a baseline, or dev/test workloads.

Key Differences Table

Feature	Provisioned Capacity	On-Demand Capacity
Billing model.	Pay per RCU/WCU provisioned per hour	Pay per request (RRU/WRU)
Scaling.	Manual or auto-scaling	Fully automatic
Best for.	Predictable workloads	Unpredictable workloads
Throttling.	Possible if exceed provision	Rare; only on extreme spikes
Price predictability.	High (fixed hourly rate)	Variable (depends on requests)

Insights:

• twtech can switch between modes once every 24 hours per table (but GSIs follow the table’s mode). 
• Also, a table’s mode applies to all indexes on it.

A diagram of difference  between Provisioned and On-Demand DynamoDB capacity modes … how reads/writes are handled.

Key takeaway:

• Provisioned mode - has a fixed throughput meter — once it’s full, extra requests get throttled (blocked) unless twtech has set auto-scaling.
• On-Demand mode - is like a stretchy pipeline that adjusts instantly to the flow, billing per operation.