NYSE TAQ Benchmark — Thread Scaling

How one solution’s query and data load performance scales with the number of threads on one machine. Use the filters to choose the solution, machine and thread counts; the charts and tables update as you go.

Methodology notes

  1. A failed query counts in the geometric mean with a penalty ratio: twice the worst time posted for that query by any compared solution, divided by the baseline’s time. Queries the baseline fails are left out. A solution whose run failed before producing any result is listed with its exit code instead of a score.

  2. Times are wall-clock times of each query; the cold run follows a cache flush, the hot runs repeat the query immediately.

  3. Benchmark methodology: see Methodology and motivation(opens in new tab).