Flutter Protobuf vs. JSON Performance Benchmark

Date: August 29, 2025 Target Hardware: Texas Instruments AM62x SoC

---

1. Project Overview

This project provides a benchmark comparing Protocol Buffers (Protobuf) against JSON for high-frequency data exchange. The simulation mimics EV subsystem communication with packets sent every 50ms.

Metrics measured:

• Data payload size
• Round-trip time (RTT)
• Serialization/deserialization speed
• Network throughput
• Latency jitter

---

2. Project Structure

/
├── protos/
│   └── ev_data.proto      # The single source of truth for data structures
│
├── lib/
│   ├── main.dart          # Main Flutter app: UI + TCP server + benchmark client
│   └── src/
│       └── generated/     # Auto-generated by protoc
│           ├── ev_data.pb.dart
│           ├── ev_data.pbenum.dart
│           └── ev_data.pbjson.dart
│
└── pubspec.yaml           # Project dependencies

• protos/ev_data.proto → Protobuf definitions
• lib/main.dart → Flutter app (UI, TCP server, benchmark client, isolates)
• lib/src/generated/ → Protoc-generated Dart classes

---

3. Setup and Execution Guide

Prerequisites

1. Flutter SDK installed.
2. Protobuf Compiler (protoc) installed and available in PATH.
3. Dart protoc Plugin:

dart pub global activate protoc_plugin

Ensure the Dart pub cache binary path is added to your system PATH:

• Windows:

  C:\Users\<YourUser>\AppData\Local\Pub\Cache\bin

• Linux/macOS:

  ~/.pub-cache/bin

---

Steps

1. Clone repo

   git clone <repo-url>
   cd <project-root>

2. Install dependencies

   flutter pub get

3. Generate Protobuf Dart files

   protoc --dart_out=lib/src/generated -Iprotos protos/ev_data.proto

   This produces:

   • ev_data.pb.dart
   • ev_data.pbenum.dart
   • ev_data.pbjson.dart

4. Run the app

   flutter run

---

4. Benchmark Results and Analysis

Test Platforms

• Windows 11 PC: Intel Core i7 9th Gen (x86-64), 32GB RAM
• TI AM62x EVM: ARM Cortex-A53, 2GB RAM, Flutter-Pi

---

Initial Benchmark (Core Metrics)

Platform	Metric	JSON	Protobuf	% Diff	Winner
Windows	Avg. Size	995.11 bytes	276.72	-72.2%	Protobuf
	Avg. RTT	603.82 µs	531.50	-12.0%	Protobuf
	Serialization	59.19 µs	18.29	-69.1%	Protobuf
	Deserialization	41.55 µs	15.71	-62.2%	Protobuf
AM62x	Avg. Size	994.98 bytes	277.00	-72.2%	Protobuf
	Avg. RTT	1904.29 µs	1952.26 µs	+2.5%	JSON
	Serialization	389.92 µs	148.26	-62.0%	Protobuf
	Deserialization	288.98 µs	115.10	-60.2%	Protobuf

---

Final Benchmark (Extended Metrics)

Platform	Metric	JSON	Protobuf	% Diff	Winner
Windows	Avg. Size	995.16 bytes	276.84	-72.2%	Protobuf
	Avg. RTT	632.27 µs	596.41	-5.7%	Protobuf
	RTT Jitter (σ)	909.58 µs	533.48	-41.4%	Protobuf
	Serialization	62.70 µs	19.89	-68.3%	Protobuf
	Deserialization	45.86 µs	17.20	-62.5%	Protobuf
	Throughput	19.34 KB/s	5.38 KB/s	-72.2%	Protobuf
AM62x	Avg. Size	995.29 bytes	276.74	-72.2%	Protobuf
	Avg. RTT	2482.61 µs	1799.80	-27.5%	Protobuf
	RTT Jitter (σ)	925.66 µs	593.24	-35.9%	Protobuf
	Serialization	354.17 µs	141.15	-60.1%	Protobuf
	Deserialization	290.30 µs	105.85	-63.5%	Protobuf
	Throughput	19.34 KB/s	5.38 KB/s	-72.2%	Protobuf

---

Key Insights

1. Data Size & Throughput

   • Protobuf payloads are ~72% smaller → massive bandwidth + buffer savings.
   • Embedded devices benefit the most from this reduction.

2. CPU Performance

   • Windows (x86): Protobuf faster due to binary efficiency.
   • AM62x (ARM): Gap widens further → Protobuf uses half the CPU time vs. JSON.

3. Latency & Jitter

   • Protobuf improved RTT by 27.5% on AM62x.
   • Jitter reduced by 36%, making it more predictable → crucial for EV real-time control.

---

5. Conclusion

On the TI AM62x SoC, Protobuf is unequivocally superior to JSON:

• ~72% reduction in payload size
• >60% faster (serialization & deserialization)
• Lower RTT & jitter
• Much better CPU efficiency

• JSON’s human readability does not justify its cost in embedded systems.
• Recommendation: Use Protobuf for EV subsystem communication on AM62x.

---