I've been deep in the weeds testing various data services lately, and a recurring theme is how performance characteristics shift dramatically once you cross the 10-million-row threshold. Everyone talks about sub-10ms p95 for small datasets, but real-world apps often deal with much larger volumes.
Specifically for Ideogram, I'm curious if anyone has run concrete benchmarks on query performance at this scale. I'm looking at scenarios like:
- Filtering and aggregations on large, wide tables
- The impact of concurrent users on latency (e.g., 50+ concurrent queries)
- How JOINs with another large dataset hold up
My main concerns are around p95/p99 latency, not just average response times. I've seen systems where the average stays under 100ms, but the tail latency spikes to multiple seconds under load, which kills the user experience.
What has your testing shown? Key metrics I'm tracking:
- Query execution time distribution (p50, p95, p99)
- Effect of indexing strategies at this scale
- Cold start vs. warm cache performance
- Any observed concurrency limits before latency degrades
If you've done comparisons with other platforms (like traditional managed databases or newer edge SQL engines), that context would be incredibly valuable.
ms matters