Gonzalo Paniagua Javier gonzalop

Below is the analysis and the actual output of running the throughput test in https://github.com/gonzalop/mq, using examples/throughput/multiple_different_args.sh against a local mosquitto MQTT 2.1.0 server running on a container with host networking.

Note that Paho v3 lacks features and checks present in v5 clients, so it's not really comparing apples to apples, but it's a good reference. Also note that I increased the max inflight in the mosquitto server documentation to 1000 in order for paho v3 not to fail QoS 0 tests.

The analysis is what Gemini thought of the output from both the client side and the server side perspectives.

Below is the analysis and the actual output of running the throughput test in https://github.com/gonzalop/mq, using examples/throughput/multiple_different_args.sh against a local mosquitto MQTT 2.1rc2 server running on a container with host networking.

Note that Paho v3 lacks features and checks present in v5 clients, so it's not really comparing apples to apples, but it's a good reference.

The analysis is what Gemini thought of the output from both the client side and the server side perspectives.

Below is the analysis and the actual output of running the throughput test in https://github.com/gonzalop/mq, using examples/throughput/multiple_different_args.sh against a local mosquitto MQTT 2.0.22 server running on bare metal.

Note that Paho v3 lacks features and checks present in v5 clients, so it's not really comparing apples to apples, but it's a good reference.

The analysis is what Gemini thought of the output from both the client side and the server side perspectives.

Performance Optimization Report - Phase 2

1. Objective

The goal was to achieve an additional 5% performance improvement in the dns library's end-to-end throughput, building upon previous optimizations.

2. Optimizations Implemented

A. Zero-Allocation Unpacking for Empty Sections

Location: msg.go (unpackQuestions, unpackRRs)

Performance Optimization Report - Incremental Unpacking

This report details the performance gains achieved by implementing incremental unpacking in the dns library. The optimizations focus on the server's hot path where a message is often partially unpacked for filtering before being fully unpacked by a handler.

System Information

OS: Linux
Arch: amd64
CPU: AMD Ryzen AI 9 365 w/ Radeon 880M

1. Micro-Benchmarks (Incremental Unpacking)

Below is the analysis and the actual output of running the throughput test in https://github.com/gonzalop/mq, using examples/throughput/multiple_different_args.sh against a local VerneMQ MQTT server running in a podman conrtainer like:

podman run --replace -p 1883:1883 -e "DOCKER_VERNEMQ_ACCEPT_EULA=yes" -e "DOCKER_VERNEMQ_ALLOW_ANONYMOUS=on" --name vernemq1 -d vernemq/vernemq

and then another run using --network host.

Note that Paho v3 lacks features and checks present in v5 clients, so it's not really comparing apples to apples, but it's a good reference.

The analysis is what Gemini thought of the output from the client perspective.

Below is the analysis and the actual output of running the throughput test in https://github.com/gonzalop/mq, using examples/throughput/multiple_different_args.sh against a local Mochi MQTT server running on bare metal.

Note that Paho v3 lacks features and checks present in v5 clients, so it's not really comparing apples to apples, but it's a good reference.

The analysis is what Gemini thought of the output.

I really ❤️ the memory use + GC runs of mq, and the performance is not to shabby either.

	+ } else if isVerne {
	+ // VerneMQ configuration
	+ // Our custom image uses a wrapper script start_vernemq
	+ cmd := []string{"start_vernemq", "--port", port}
	+ // Parse mosquitto config to extract limits
	+ lines := strings.Split(configContent, "\n")
	+ for _, line := range lines {
	+ line = strings.TrimSpace(line)
	+ if strings.HasPrefix(line, "max_packet_size") {
	+ var size int