認証・query benchmark

実測結果 — 2026年9月8日

2026年9月8日 · NRT edge · 各100回 · warmup 5回を除外 · 並列度1 · 1 entity

Biscuit発行

32.23 ms

p50 · ミリ秒

100/100件成功

Biscuit検証

18.26 ms

p50 · ミリ秒

100/100件成功

認証済みwarm query

741.67 ms

p50 · ミリ秒

100/100件成功

実測レイテンシ — ミリ秒
Operation	p50	p95	p99	Max
Biscuit発行	32.23	110.02	123.46	144.59
Biscuit検証	18.26	80.78	105.49	105.64
認証済みwarm query	741.67	961.66	1073.58	3667.31

測定方法と限界

Single-client production HTTPS latency, including network and authorization. A small two-pattern Datalog point selection; not a throughput, large-graph, controlled cold-cache, or competitor ranking claim.

測定したすべてのqueryで書き込んだmarkerを確認しました。計300件の測定requestはすべて成功しました。

First observed write: 1806.39 ms. First query read-back: 1009.52 ms. Neither is a controlled cold-cache measurement.

The client ran Node 26.7.0. Edge location was NRT; client geography and negotiated HTTP version were not independently measured. Percentiles use nearest-rank. Connection reuse was enabled; caches were not forcibly cleared.

測定結果JSONをダウンロード

KOTOBASE_LIVE_BISCUIT_BENCH=1 KOTOBASE_BENCH_SAMPLES=100 KOTOBASE_BENCH_WARMUPS=5 node authn/scripts/live_biscuit_query_bench.mjs --provision

Reproduction creates an isolated test tenant and writes one test entity. No payment or blockchain transaction is required. The raw receipt includes the harness hash and source revision.