CASE 154 · CONDUIT · 2025
Indexes that match the queries, not the wishes.
An IoT telemetry platform’s DynamoDB tables had been designed for the queries the team imagined they would need. A year in, the actual access patterns had drifted. Read latency was up, costs were up, and one GSI was scanning 80% of the table on every query. We redesigned the indexes around the real access patterns.
IoT telemetry
RELIABILITY
2025
RESULTS
What changed, by the numbers.
READ COST
−63%
p99 QUERY LATENCY
−71%
GSIs OPTIMISED
5 / 7
HOT-KEY VIOLATIONS
0
HOW IT WENT
CloudWatch Contributor Insights gave us the truth — which GSIs were actually used, which were near-empty, which queries were scanning rather than seeking. Two of the seven GSIs had no queries against them at all (they had been intended for a planned feature that pivoted).
We retired the unused GSIs, redesigned the partition keys of the three most expensive GSIs to eliminate hot keys, and added a composite sort key to enable the range queries the application was emulating with scans. DAX absorbed the read-heavy patterns where caching helped.
Read cost dropped 63%. p99 query latency dropped 71% — the previous setup had been doing far more work than queries required. Two GSIs retired entirely. Hot-key violations dropped to zero because the partition design now matched the access pattern.
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.