Open mikemccand opened 1 year ago
I'll try to make a first cut at this.
OK I pushed a first cut. It produces many lines of output looking like this:
TOP_10_COUNT HighTerm
145.3% from 206.1 ms 84.0 ms count=7834704 (WARNING: tantivy_count=7834746)
143.5% were 88.1 ms 36.2 ms count=3319889 (WARNING: tantivy_count=3319906)
143.5% has 110.8 ms 45.5 ms count=4180955 (WARNING: tantivy_count=4180986)
143.4% which 118.0 ms 48.5 ms count=4469484 (WARNING: tantivy_count=4469528)
143.2% have 101.0 ms 41.5 ms count=3821616 (WARNING: tantivy_count=3821633)
142.9% also 97.7 ms 40.2 ms count=3702272 (WARNING: tantivy_count=3702292)
142.4% his 106.9 ms 44.1 ms count=4051071 (WARNING: tantivy_count=4051077)
142.4% first 80.8 ms 33.3 ms count=3044659 (WARNING: tantivy_count=3044671)
141.6% been 76.6 ms 31.7 ms count=2892317 (WARNING: tantivy_count=2892324)
141.6% had 72.8 ms 30.1 ms count=2733137 (WARNING: tantivy_count=2733149)
141.5% one 82.6 ms 34.2 ms count=3122527 (WARNING: tantivy_count=3122538)
141.3% he 86.8 ms 36.0 ms count=3278298 (WARNING: tantivy_count=3278314)
140.9% who 74.6 ms 31.0 ms count=2811995 (WARNING: tantivy_count=2812021)
140.4% its 67.9 ms 28.3 ms count=2548474 (WARNING: tantivy_count=2548493)
140.3% other 67.9 ms 28.2 ms count=2557746 (WARNING: tantivy_count=2557757)
139.9% name 62.7 ms 26.1 ms count=2348632 (WARNING: tantivy_count=2348644)
First column is what % slower Lucene is, then Lucene's time, then Tantivy's, then the hit count, and also Tantivy's different count if different.
Also, separately, I'm quite happy I got this issue number (42).
It's nice that we get a
results.json
with all the juicy details of each benchmark, but, for sharing purposes (e.g. the Graviton results I just attached to #36) it'd be nicer/easier to copy/paste a text summary so that we could see at a glance what the results look like.We could just make a new tool that parses the JSON and renders to flat text, in addition to the
make bench
that renders to an interactive Web UI.