Longevity: methodology
PlannedHow queries slow down over weeks of commits, and what maintenance wins back.
Not run yet. Borrows LST-Bench's method (Microsoft, SIGMOD 2024), applied to our own time-series data. This page is the plan.
What we'll run
Only LST-Bench's method: our own data, queries, and runs, not its code or workload.
The idea
LST-Bench measures how table formats degrade over many rounds of writes and maintenance. We apply the method to time series: rounds of streamed commits, with each system's maintenance between some rounds, and the same queries after every round.
How we'll run it
- TSBS IoT and, later, the industrial IoT workload, extended to 30 days of commits.
- Query time, files, metadata size, and requests after each round, charted against the number of commits.