RadRouter Performance

Performance

The router stores routes in a path-compressed trie:

• Static lookups: O(L) where L = path length (single-pass walk with one integer cursor).
• Parameter / greedy lookups: O(L) average; worst case adds a small constant for backtracking when a :name:<suffix> doesn't match.
• No allocation per lookup on the common path (path.split('/') isn't called; the cursor walks path directly).
• No Map.get() per hop — staticChildren uses an indexed Object.create(null).

Table of Contents

• Reproducing
• Per-shape lookup
• Case-sensitive vs case-insensitive
• Router shootout

Reproducing

Two benchmark suites live in the package:

• RadRouter.bench.ts — six lookup shapes against a fully-loaded router (1150 routes), plus a case-sensitive vs case-insensitive comparison. Static lookup is the baseline within each group.
• RadRouter.compare.bench.ts — head-to-head against find-my-way (Fastify) and radix3 (Nitro/Nuxt) under an identical workload.

deno bench --allow-all packages/radrouter/RadRouter.bench.ts
deno bench --allow-all packages/radrouter/RadRouter.compare.bench.ts

Per-shape lookup

One run on an Apple M2 Max / Deno 2.7.14. Treat these as a snapshot, not a contract — re-run on your own hardware.

1150 routes registered; static is the baseline within the group.

Lookup shape	time/iter	iter/s	vs static
static (versioned)	102.1 ns	9.79M	1.00×
static (deep, unversioned)	100.3 ns	9.97M	1.02×
greedy (multi segment)	119.0 ns	8.40M	0.86×
greedy (single segment)	122.2 ns	8.18M	0.84×
param (2 params)	236.4 ns	4.23M	0.43×
miss	32.1 ns	31.2M	3.18×

A miss is faster than a hit because the walk terminates as soon as the trie diverges from the path — no further node traversal or parameter capture happens.

Case-sensitive vs case-insensitive

Same lookup workload, just the option flipped:

Mode	time/iter	iter/s	vs CS
case-sensitive	102.2 ns	9.78M	1.00×
case-insensitive	128.8 ns	7.77M	0.79×

CI mode lowercases the URL once at lookup entry and threads both the case-folded view (for label matching) and the original view (for param-value extraction) through the trie walk. That costs ~26% on this workload. Set caseSensitive: true (the default) if you don't actually need forgiving matching.

Router shootout

1150 routes registered, blended 5-path workload, one consumer running each router back-to-back. Same machine, same workload, single Deno process per row.

Router	time/iter	iter/s	Setup
RadRouter (CS)	133.1 ns	7.51M	2.18ms
RadRouter (CI)	145.0 ns	6.90M	1.34ms
radix3 (Nitro)	147.1 ns	6.80M	1.80ms
find-my-way (Fastify)	159.1 ns	6.29M	18.95ms

Practical reading: RadRouter (CS) leads both peers on per-iteration lookup; CI mode and radix3 sit in a similar tier ~9–11% behind; find- my-way trails by ~20% on lookup and ~9× on setup. The benchmark file at RadRouter.compare.bench.ts is the source — re-run on your own hardware to get current numbers.

---

← Back to RadRouter