CASE 44 · FORGE · 2024
DynamoDB on-demand was cheaper. Until it wasn’t.
A gaming backend ran 140 DynamoDB tables, all on on-demand capacity, because the team had read "start with on-demand" three years ago and never revisited. Half the tables had stable, predictable traffic. The DynamoDB bill was $48k a month. We rebalanced and brought it to $19k.
Gaming backend
COST
2024
RESULTS
What changed, by the numbers.
DYNAMO BILL
−60%
TABLES REBALANCED
74
p99 LATENCY
0% Δ
THROTTLES
< 0.01%
HOW IT WENT
The team’s instinct against switching to provisioned capacity was traffic spikes — every game release caused one. But analysis showed only 14 of 140 tables actually had spike-driven workloads. The rest had steady traffic with daily cycles.
We modelled each table’s traffic curve from CloudWatch metrics over 90 days and recommended a per-table capacity mode. Tables with under 30% peak-to-trough ratio moved to provisioned with auto-scaling. Tables with bursty traffic stayed on-demand. Reserved capacity covered the predictable floor on the provisioned set.
Bill dropped 60% across the migrated set. Throttling stayed under 0.01% — well below the noise floor. The 14 spike-driven tables stayed on-demand and accounted for most of the remaining DynamoDB spend. The team now reviews capacity quarterly.
RELATED · SAME DOMAIN
Other engagements in this space.
READY WHEN YOU ARE
Let's get your AWS bill (and architecture) in order.
The discovery call is free. You walk away with at least one concrete idea — even if we never work together.