PERFORMANCE.md
Markdown-style heading precedes a code fenceAt least four Markdown section headings
SHA-25694593865e07f0c30a002cd99de8bad08e02bb8ccec7446e6ce03b71fd7ddc37c
MaleculeH(F)Md
Evidence
1# Performance baseline
2
3Measure the cost of sending a no-op job before choosing optimizations. The
4end-to-end harness uses generated data and explicitly selected configured
5peers. It never snapshots the developer's working tree.
6
7## Reproduce
8
9Build the client being measured, then choose idle peers:
10
11```sh
12mkdir -p dist
13go build -trimpath -o dist/errand-benchmark ./cmd/errand
14python3 scripts/benchmark.py --binary ./dist/errand-benchmark \
15 --on …
26:37… are selected only through repeated `--on`.
27
28Run this separately from builds, tests, and other benchmarks. The harness
29refuses a runner with active or queued jobs at the initial probe, but does not
30reserve it or exclude unrelated proces …
37:56… h timed invocation;
38 require the CLI to report that all file content was shipped.
39- **Cached snapshot:** seed a separate fixed tree with one warmup …
74:10… ith all file content already cached. Two shapes have the same 8 MiB
75payload: 128 files of 64 KiB and 4,096 files of 2 KiB. Fixtures are prepared
76outside timing on a warm local filesystem, without Git metadata or ignore
77rules. Archive output goe …