eperbit

Energy per bit measurements for ID codes. This repository contains the code, scripts, and measurement data used to evaluate the latency and energy consumption of several identification (ID) codes and hash-based identification schemes on a testbed consisting of a Raspberry Pi 5 and Keysight DC Power Analyzer.

What’s Being Evaluated

We benchmark the following ID code families:

• RSID (Reed–Solomon ID)

• RS2ID (Concatenated Reed–Solomon ID)

• RMID (Reed–Muller ID)

• SHA256ID (SHA-2 256 based identification code)

• PMHID (Polymur Hash–based ID)

Each code is implemented over Galois Field (GF) symbol sizes:

• 8

• 16

• 32

• 64

The goal is to compare their:

• Encoding latency

• Power and Energy consumption

• Energy cost of a bit of information encoded

All tests are run repeatedly (typically ~20 seconds or ≥10,000 iterations) to clearly separate active processing from background noise.

Testbed Overview

All measurements were performed on:

• Device Under Test (DUT): 64-bit Raspberry Pi 5

  • BCM2712, Cortex-A76

  • 8GB RAM

• OS: Debian 12

• Kernel: 6.12.47

• Compiler: GCC 12.2.0

• IDCODES Library:

  pip install ecidcodes

Compile Flags:

-O3 -mtune=cortex-a76 -march=armv8-a+crypto

• OpenSSL Version: 3.0.18

Power Measurement Setup

• Power Analyzer: Keysight N6705B DC Power Analyzer

  • N6785A module

• Voltage supply: 5.2V via Source to GPIO on Rpi 5

• Data Logger Sampling Period: 0.1 ms

We record .dlog files and post-process them to extract:

• Time

• Voltage

• Current

Background noise was minimized:

• WiFi and Bluetooth disabled

• Ethernet physically removed during tests

• LEDs disabled after boot

• Passive cooling (heatsink only)

• Controlled room temperature

Warmup and sleep phases are added to stabilized temperature and CPU frequency and allow us to programmatically separate active compute phases from background fluctuations.

---

Repository Structure

.
├── DATASET20260212/
├── RPI/
├── eperbit.py
├── parselogs.py
├── finalsummary.csv
└── ...

Top-Level Files

• eperbit.py — energy per bit computation and metrics evaluation

• parselogs.py — parsing log files including Time and Temperature

• finalsummary.csv — final aggregated benchmark results from .log files

• version.txt — experiment version reference

• LICENSE / COPYING — licensing information

DATASET20260212/

Contains:

• Measurement logs and labeled segments for:

  • PMHID-{8,16,32,64}

  • RMID-{8,16,32,64}

  • RSID-{8,16,32,64}

  • RS2ID-{16,32,64}

  • SHA256ID-{8,16,32,64}

• Log processing and segmentation scripts in DATASET20260212:

  • dataset.py

  • segmentify.py

  • segmentify_all.sh

  • parselogs.py

  • dlogviewer.py

  • exportdlog_to_csv.sh

  • msgsizeveclen.sh

  • sortfiles.sh

• Aggregated data:

  • combined_summarylogs.csv

These scripts convert raw Keysight .dlog files into structured CSV data and labeled segments for energy analysis.

---

RPI/

Scripts used directly on the Raspberry Pi 5 during benchmarking:

• warmup.sh — CPU warmup phase

• test.sh — run benchmark

• time.sh — Set System Clock Time

• killservices.sh — disable background services

• screen_test.sh — run test.sh inside screen session

• stress_test_results.txt — stress test output

# Configure RPI to allow boot and power over GPIO and prevent hard crashes
sudo rpi-eeprom-config --edit
PSU_MAX_CURRENT=5000 

How the Benchmarks Work?

Each benchmark run:

1. Generates a random input message (seeded by date/time).

2. Uses a fixed tag position for consistent energy comparison.

3. Executes the encoding algorithm repeatedly.

4. Inserts warmup and sleep phases.

5. Records high-resolution voltage and current traces.

6. Segments and labels active compute regions.

7. Computes:

   • Average latency

   • Energy consumption

   • Energy per bit

The repetition ensures that the encoding phase stands out clearly against Linux background process activity.

---

Why Energy per bit Matters?

This repository provides:

• Real measured power traces with 0.1ms accuracy

• Repeatable benchmarking scripts

• Segmented dataset ready for analysis including timestamped periods of activity during benchmark and idle state

• Direct comparison across GF sizes and ID constructions

---

Reproducing the Experiments

1. Set up a Raspberry Pi 5 with Debian 12 (64-bit).

2. Disable background services, radios, and LEDs using killservices.sh.

3. Use a high-resolution DC power analyzer and connect via USB-C or GPIO 5V.

4. Record .dlog files and convert to CSV using dlogviewer.py.

5. Process logs using scripts in DATASET20260212/.

6. Analyze aggregated results via eperbit.py.

---

License

See LICENSE and COPYING for full licensing details.

Copyright (c) 2026 Ecologic Computing GmbH info@ecologic-computing.com

Visit ecologic-computing.com/research for more details.