WhizMQ - A message queue library implementing ZMTP 3.0 in Nim.
Built on top of PowPow event library

nimble install whiz

API reference

😍 Key Features

• ZMTP 3.0 wire protocol, full greeting, framing, and command handling
• Socket patterns: PAIR, PUB/SUB, REQ/REP, PUSH/PULL, File Transfer
• Security mechanisms:
  • NULL: no authentication (plaintext)
  • PLAIN: username/password authentication with ZAP callback
  • CURVE: X25519 + XChaCha20-Poly1305 AEAD encryption and mutual authentication
• Transport TCP and IPC (Unix domain sockets)
• Built on PowPow event notification library in Nim

Note

CURVE requires Monocypher (via the e2ee package). Install it with your system package manager (brew install monocypher, apt install libmonocypher-dev, etc.) or build from source.

🗺 Roadmap

• ZMTP 3.0 wire protocol (greeting, framing, commands)
• PAIR, PUB/SUB, REQ/REP, PUSH/PULL socket patterns
• File transfer over ZMTP
• PLAIN security mechanism + ZAP auth callbacks
• CURVE security mechanism (X25519 + AEAD)
• RADIO/DISH (dgram) socket pattern
• CLIENT/SERVER (stream) socket pattern
• CURVE vouch (Ed25519 signature in HELLO for key continuity)
• TLS/DTLS transport
• GSSAPI / Kerberos mechanism
• WebSocket transport (RFC 7692)
• Formal ZMTP conformance tests
• Performance benchmarks vs ØMQ

Examples

...

Benchmarks

Note

Benchmark results are not consistent and may show different results across runs and environments.

Column legend:

• benchmark — test name: {pattern}_{scenario}_{transport} (_tcp or _unix)
• subscriber — specific subscriber/worker for multi-endpoint tests; total is the aggregate row, - for single-endpoint
• n — number of messages sent (throughput) or round-trips (latency)
• size — payload size in bytes
• total(μs) — wall-clock duration of the test in microseconds
• throughput/s — messages per second = n / total(μs) * 1_000_000
• min(μs) — fastest observed per-message time (minimum)
• p50(μs) — median per-message time (50th percentile)
• p75(μs) — 75th percentile per-message time
• p90(μs) — 90th percentile per-message time
• p99(μs) — 99th percentile per-message time
• avg(μs) — arithmetic mean per-message time
• σ(μs) — population standard deviation of per-message times

Per-message time collection:

• Latency benchmarks (pair_latency, reqrep): each value is one request–reply round-trip measured from the client
• Throughput benchmarks (all others): inter-arrival deltas between consecutive receives — approximates per-message service time distribution

benchmark	subscriber	n	size	total(μs)	throughput/s	min(μs)	p50(μs)	p75(μs)	p90(μs)	p99(μs)	avg(μs)	σ(μs)
bench_pair_latency_tcp	-	5000	1	100325	49837	17.9	18.4	18.9	23.1	41.7	20.0	4.7
bench_pair_latency_unix	-	5000	1	64249	77821	12.2	12.5	12.6	12.9	18.5	12.8	1.2
bench_pair_throughput_tcp	-	10000	512	16445	608080	0.0	0.0	0.1	1.3	4.0	0.4	3.5
bench_pair_throughput_unix	-	10000	512	15291	653975	0.0	0.0	0.1	1.3	12.5	0.9	3.0
bench_pubsub_1sub_tcp	-	10000	512	11360	880222	0.0	0.0	0.1	1.3	3.8	0.4	3.4
bench_pubsub_1sub_unix	-	10000	512	13933	717704	0.0	0.0	0.1	1.3	12.1	0.9	2.9
bench_pubsub_2sub_tcp	sub0	5000	256	15679	318881	0.0	0.0	0.0	0.1	6.9	0.5	17.3
bench_pubsub_2sub_tcp	sub1	4143	256	15679	264225	0.0	0.0	0.0	0.1	7.3	0.5	12.2
bench_pubsub_2sub_tcp	total	9143	256	15679	583106	0.0	0.0	0.0	0.1	7.1	0.5	15.2
bench_pubsub_2sub_unix	sub0	5000	256	5279	947043	0.0	0.0	0.0	0.1	19.2	0.7	3.4
bench_pubsub_2sub_unix	sub1	5000	256	5279	947043	0.0	0.0	0.0	0.1	19.3	0.7	3.4
bench_pubsub_2sub_unix	total	10000	256	5279	1894086	0.0	0.0	0.0	0.1	19.2	0.7	3.4
bench_reqrep_tcp	-	5000	128	100401	49800	16.6	18.9	19.2	23.5	38.2	20.0	3.7
bench_reqrep_unix	-	5000	128	64793	77168	12.5	12.8	12.8	12.9	16.1	12.9	0.9
bench_sizes_pair_tcp	-	2000	64	5586	357990	0.0	0.0	0.0	0.0	3.1	0.1	0.5
bench_sizes_pair_tcp	-	2000	1024	3461	577759	0.0	0.0	1.2	1.6	2.7	0.6	3.6
bench_sizes_pair_tcp	-	2000	65536	258856	7726	19.4	20.4	21.8	25.6	253.5	28.4	39.9
bench_sizes_pair_unix	-	2000	64	582	3435942	0.0	0.0	0.0	0.0	1.5	0.1	1.1
bench_sizes_pair_unix	-	2000	1024	4677	427548	0.0	0.0	1.3	11.4	12.3	1.7	3.9
bench_sizes_pair_unix	-	2000	65536	411682	4858	99.7	104.1	107.3	110.7	126.9	105.3	5.3
bench_pushpull_throughput_tcp	-	10000	512	10436	958154	0.0	0.0	0.1	1.3	3.9	0.3	3.4
bench_pushpull_throughput_unix	-	10000	512	11985	834370	0.0	0.0	0.1	1.3	12.2	0.9	3.0
bench_pushpull_1worker_tcp	-	10000	128	3979	2512663	0.0	0.0	0.0	0.0	1.6	0.1	2.4
bench_pushpull_1worker_unix	-	10000	128	3643	2744939	0.0	0.0	0.0	0.0	11.5	0.2	1.5
bench_pushpull_3workers_tcp	worker0	1667	128	6350	262516	0.0	0.0	0.0	0.0	3.2	0.4	9.4
bench_pushpull_3workers_tcp	worker1	1667	128	6350	262516	0.0	0.0	0.0	0.0	2.8	0.3	8.2
bench_pushpull_3workers_tcp	worker2	1666	128	6350	262358	0.0	0.0	0.0	0.0	2.8	0.3	7.0
bench_pushpull_3workers_tcp	total	5000	128	6350	787391	0.0	0.0	0.0	0.0	4.3	0.3	8.3
bench_pushpull_3workers_unix	worker0	1667	128	1599	1041951	0.0	0.0	0.0	0.1	27.7	0.5	3.7
bench_pushpull_3workers_unix	worker1	1667	128	1599	1041951	0.0	0.0	0.0	0.0	27.7	0.5	3.6
bench_pushpull_3workers_unix	worker2	1666	128	1599	1041326	0.0	0.0	0.0	0.1	27.7	0.5	3.6
bench_pushpull_3workers_unix	total	5000	128	1599	3125228	0.0	0.0	0.0	0.0	27.7	0.5	3.6

Key observations:

• Unix domain sockets are faster for small messages — latency benchmarks show ~55% higher throughput on Unix (pair_latency: 78k/s vs 50k/s; reqrep: 77k/s vs 50k/s) by skipping TCP stack overhead.
• Unix latency is more consistent — σ drops from 4.7→1.2 (pair_latency) and 3.7→0.9 (reqrep), indicating less jitter.
• Large payloads (≥64 KiB) favor TCP — Unix has smaller default socket buffers, causing fragmentation: 65536B pair throughput is 7.7k/s (TCP) vs 4.9k/s (Unix), with avg latency 28μs vs 105μs.
• Multi-worker pushpull scales poorly on TCP — 3 workers saturate at 787k/s (less than a single worker's 2.5M/s) due to TCP fairness/contention. Unix handles it near-linearly at 3.1M/s aggregate.
• Pub/sub delivery is imbalanced on TCP — pubsub_2sub TCP delivers 5000/4143 unevenly; Unix delivers evenly (5000/5000), suggesting single-subscriber fairness differences with Nagle/ACK delaying.
• Throughput is consistent across runs — no multi-second stalls or order-of-magnitude variance, confirming the earlier TCP cork and edge-triggered event fixes are effective.

❤ Contributions & Support

• 🐛 Found a bug? Create a new Issue
• 👋 Wanna help? Fork it!

	Switch to Open-Source LLMs via OpenCode GO, choosing from a variety of powerful models such as DeepSeek, Qwen, Kimi, GLM-5, MiniMax, MiMo. 🍕 Use our referral link to get started!

🎩 License

MIT license. Made by Humans from OpenPeeps.
Copyright OpenPeeps & Contributors — All rights reserved.