Amazon DynamoDB offers two throughput modes: on-demand and provisioned, each with its unique characteristics and ideal use cases.
On-demand mode automatically scales to handle variable workloads without capacity planning, making it ideal for serverless applications and unpredictable traffic patterns.
Provisioned mode requires specifying read and write capacities, making it suitable for steady, predictable workloads with consistent utilization.
On-demand mode provides cost predictability with pay-per-request pricing, while provisioned mode is based on provisioned capacity regardless of utilization.
Throttling risks vary between modes, with on-demand being less likely to throttle automatically compared to provisioned mode.
Auto-scaling in provisioned mode adjusts capacity based on configured utilization thresholds, while on-demand mode scales instantly based on traffic.
Hot-partition throttling can occur in both modes, necessitating proper data modeling to distribute traffic across partitions effectively.
Choosing between on-demand and provisioned modes depends on workload predictability, traffic changes, cost considerations, and operational value.
On-demand mode simplifies capacity management and aligns well with modern applications, emphasizing development agility and consistency.
Provisioned mode can be more cost-effective for predictable workloads with high utilization and requires ongoing monitoring and optimization for optimal performance.
The decision to choose on-demand or provisioned mode should consider traffic variability, operational complexity, and the need for immediate response to workload changes.