On this page

Benchmark (0.0.2)

Synthetic publish throughput benchmarks using cometmq-bench against a local single-node broker.

Environment

Broker: cometmq-broker 0.0.2 (release build)
Tool: cometmq-bench (publisher mode)
Concurrency: 10 workers
Payload pattern: zeros
Compression: lz4 (threshold 1024 bytes)
Publisher confirms: disabled unless stated
Persistence: as noted per test

Throughput (Non-Persistent, No Confirms)

Payload (bytes)	Msg/s	Approx MB/s*
64	99,902	6.37
128	102,593	12.57
512	69,481	35.96
1,024	26,222	25.79
16,384	24,601	384.05
131,072	3,120	392.93
1,048,576	3,015	3,014.97
8,388,608	1,332	11,162.88

Throughput (Persistent, No Confirms)

Payload (bytes)	Msg/s	Approx MB/s*
64	68,900	4.41
128	66,615	8.17
512	47,349	24.48
1,024	26,445	26.03
16,384	24,789	386.98
131,072	2,973	374.60
1,048,576	2,714	2,713.60
8,388,608	1,228	10,301.44

Throughput (Persistent + Confirms)

Payload (bytes)	Msg/s	Approx MB/s*
64	34,470	2.21
128	33,337	4.09
512	27,678	14.33
1,024	17,910	17.63
16,384	17,669	275.67
131,072	2,969	374.10
1,048,576	2,930	2,929.69
8,388,608	1,299	10,896.38

*Approx MB/s = (Msg/s * PayloadBytes) / 1_000_000.

Summary (Msg/s)

Mode	64B	128B	1KB	16KB	1MB
Non-persistent	99,902	102,593	26,222	24,601	3,015
Persistent	68,900	66,615	26,445	24,789	2,930
Persistent + confirms	34,470	33,337	17,910	17,669	2,414

Notes

lz4 compression activates above 1024 bytes.
Numbers reflect loopback network (no external latency).
Persistence + confirms show expected throughput reduction due to fsync and acknowledgment round-trips.

Tool Invocation (Example)

bash

cometmq-bench pub \
  --count 50000 \
  --payload-bytes 128 \
  --concurrency 10 \
  --persistent false \
  --confirm false

Adjust parameters for different payload sizes and durability/confirm modes.