Case study — 2025 — Syberry
Perseus: 109M+ Records. 720x Faster.
Perseus is industrial IoT analytics for water treatment telemetry. A hardware-aware PostgreSQL partitioning strategy cut critical queries from 1 hour to under 5 seconds across 109M+ records by sending enterprise-tier clients with expensive monitoring devices to dedicated partitions.
The problem
The system ingests continuous telemetry from water treatment facilities. More than a hundred million rows accumulate across the device fleet, and the nightly critical reports aggregate over most of it. A single important query took about an hour to complete.
Adding indexes and more replicas was no longer a strategy. The table needed a physical layout that matched how the business actually reads the data.
The approach
Profile reading patterns before touching schema
Users rarely query the fleet as one blob. Enterprise-tier clients with many high-value monitoring devices read their own history; operational staff scan recent windows. The access profile defined the partition key.
Hardware-aware partitioning
Partitioning was designed around the monitoring hardware tiers: enterprise-grade devices with expensive sensors went to dedicated partitions. Partition pruning then excludes the rest of the fleet for those queries automatically.
Land the change without an outage window
The migration ran on PHP 8 / Symfony services so it could execute incrementally while the old table kept serving reads, then queries were flipped to the partitioned tables once validated.
Dashboards verify the win
React and TypeScript dashboards on top meant the improvement was visible to non-engineers immediately, not just in EXPLAIN plans.
Results
109M+
Records queried in under 5 seconds
720x
Faster critical query path
1h → <5s
Nightly critical reporting runtime
Other case studies
My role on Syberry: PostgreSQL performance & data architecture.