One million completed messages

v12.0.0 vs v14.0.0

Each value compares node-ipc 12.0.0 with 14.0.0 on the same runner, Node version, transport, payload, and paired sample. Results stay grouped by operating system and socket semantics.

Paired timing snapshot

Paired node-ipc v12.0.0 and recorded-version median time for one million messages, grouped by Linux, macOS, and Windows transport — The chart names the measured version from its retained cohort. Fixed platform and transport order prevents timing-based sorting. A shorter time inside one exact pair is meaningful; rows and platforms are not a universal transport ranking.

Accepted 14.0.0 cohort

Six environments passed.

Workflow run 32623449214 measured commit 2692d32 on Windows, macOS, and Linux with Node.js 22.13.0 and 24.18.1. v14.0.0 was faster in all 30 exact environment/transport pairs; paired median speedups ranged from 1.14× to 4.96×. Every accepted record passed provenance, content, sequence, byte, count, process-exit, endpoint-release, and temporary-state gates.

Accepted paired results

Platforms appear as Linux, macOS, then Windows. Every table keeps native local IPC, TCP, TLS, UDP4, and UDP6 in that fixed order.

Loading compact transport data…

Platform	Native local IPC	TCP	TLS	UDP4	UDP6
Linux	Unix domain socket · Pending	Pending	Pending	Pending	Pending
macOS	Unix domain socket · Pending	Pending	Pending	Pending	Pending
Windows	Windows named pipe · Pending	Pending	Pending	Pending	Pending

Read within a pair

Native local IPC

Linux and macOS use Unix domain sockets. Windows uses a named pipe. Those are host-native implementations, not one portable socket primitive. Both versions use the Node byte reflector.

TCP + TLS

Both versions use the Node byte reflector. TCP is a loopback ordered stream. TLS is an encryption-only bulk-transfer lane: peer verification is disabled consistently for both recorded versions, and the handshake is outside timing. It is not authenticated-TLS evidence.

UDP4 + UDP6

Official datagram evidence uses the standard-C exact-count reflector for both versions. The lanes retain their message boundaries and address families, and accepted samples verify the expected content, sequence, counts, and cleanup.

Directional CI evidence.

These GitHub-hosted hybrid-oracle runs are classified snapshot-noisy. They show paired observations on ephemeral runners, not hardware-normalized rankings, certifications, or a performance regression gate. No result is hosted until the complete matrix and every oracle-provenance, exact-count, correctness, and cleanup gate are valid.

Measurement contract

Field	Paired value
Application payload	Deterministic 64 bytes
Correctness probe	64 sequenced messages before timing
Warm-up	100,000 completed messages
Measured work	1,000,000 completed messages per version and transport
Samples	Seven alternating v12/recorded-version pairs
Oracle	Same implementation within each pair: Node byte reflector for local/TCP/TLS; standard-C exact-count reflector for official UDP4/UDP6 evidence
C provenance	Source and binary SHA-256, compiler and version, fixed flags, and platform/architecture target
Primary value	Median milliseconds per million completed messages

Open the per-environment raw samples from the live tables, or start from the transport manifest. The methodology documents pairing, exact-count gates, oracle provenance, and interpretation.

← Benchmark overview Methodology →