Search Benchmark, the Game

Setup

Text analysis

- both engines use whitespace tokenizer

Index Settings

- both engines have index sorted by a numeric ID field

- both engines have 2% documents deleted. The deleted subset are exactly the same so we can still expect both engine to return same results.

Benchmark run

- Ran on m5.12xlarge machine

- Each engine run 10 iterations, after the first warmup iteration (excluded from stats)

- The latencies below are the min across all iterations