idx

idx vs grep Benchmark Report

Date: 2026-05-06
Repository: idx (self-benchmark in this repo)

Scope

This report compares search performance between:

• idx service (in-process search service)
• idx CLI (idx search needle)
• grep (grep -Rnw --exclude-dir=.git --exclude-dir=.idx needle .)

Benchmark command used:

go test ./internal/core/services/search -run '^$' -bench '^BenchmarkSearchVsGrep$' -benchmem -count=3

Corpus profiles

• files-500: 20 directories x 25 files (500 files)
• files-2000: 40 directories x 50 files (2000 files)

Results (average of 3 runs)

Corpus	Method	ns/op (avg)	ms/op	B/op	allocs/op
files-500	service	28,625	0.029	15,166	486
files-500	cli	8,379,099	8.38	14,555	43
files-500	grep	14,750,834	14.75	18,229	80
files-2000	service	115,375	0.115	60,562	1,910
files-2000	cli	18,603,417	18.60	14,784	43
files-2000	grep	55,202,165	55.20	18,796	80

Key comparisons

idx CLI vs grep

• files-500: idx is approximately 1.76x faster than grep
• files-2000: idx is approximately 2.97x faster than grep

Allocations and memory

• idx CLI uses fewer allocations than grep (43 vs 88 allocs/op)
• idx CLI also uses fewer bytes per operation in these runs

Interpretation

• grep scans raw files on each run.
• idx searches prebuilt indexes, so query-time work is lower.
• The in-process service is much faster than CLI because it avoids process-spawn overhead.
• The relative advantage of idx over grep increases with larger corpus size in this benchmark.

Notes and limitations

• This benchmark focuses on query latency, not index creation cost (init/sync).
• CLI results include process startup cost.
• Absolute numbers may vary by machine, filesystem, and system load.
• Use this benchmark for directional comparison under the same environment.

Reproduce

make bench-search-vs-grep
# or
go test ./internal/core/services/search -run '^$' -bench '^BenchmarkSearchVsGrep$' -benchmem -count=3

This site is open source. Improve this page.