nmson/pi5-agricultural-iot-32day

Pi5 Agricultural IoT — 32-Day Continuous Performance Dataset

Dataset Summary

This dataset contains 9,409 system performance records collected from a Raspberry Pi 5 (8 GB) deployed as an edge computing node at an agricultural farm in Biên Hòa, Đồng Nai, Vietnam. The device managed 10 simultaneous IP camera streams, performing real-time RTSP→RTMP transcoding via FFmpeg, orchestrated by 10 independent systemd services.

Data was collected continuously for 32 days (23/03/2026 – 25/04/2026) at 5-minute intervals with no gaps. Each record contains 18 system-level metrics covering CPU, memory, thermal, network, and disk performance.

Key findings:

• Pi 5 sustained 10-channel live streaming at mean CPU usage of only 4.23% using FFmpeg -c:v copy passthrough strategy
• Zero reboots, zero swap usage, zero thermal throttling events across the entire 32-day period
• Core temperature strongly correlated with ambient thermal conditions (r ≈ 0.65 with network throughput), not CPU load
• ISP reliability of 97.5% (zero packet loss) with mean latency of 53 ms — sufficient for continuous live streaming

All sensitive information (camera IPs, credentials, RTMP stream keys, serial numbers) has been anonymized before public release.

---

Dataset Structure

File Overview

pi5-agricultural-iot/
├── README.md                  ← This file (Dataset Card)
├── pi5_metrics_research.csv   ← Main dataset (9,409 records × 18 metrics)
├── logger.sh                  ← Data collection script (Bash/cron)
├── setup_10_cams.sh           ← systemd service setup (anonymized)
└── cameras_anonymized.json    ← Camera config (SN/SC/IP anonymized)

Main Dataset: pi5_metrics_research.csv

Field	Metric	Unit	Description
Time	Timestamp	YYYY-MM-DD HH:MM:SS	Collection time (local time UTC+7)
Uptime_s	System uptime	seconds	Time since last reboot
CPU_Usage_%	CPU utilization	%	Busy percentage (100 – idle), 1-second vmstat sample
CPU_Freq_MHz	CPU frequency	MHz	Current scaling frequency of CPU core 0
RAM_Usage_%	RAM utilization	%	Used / total physical memory ratio
RAM_Avail_MB	Available RAM	MB	Available memory per kernel (free + reclaimable)
Swap_Usage_%	Swap utilization	%	Swap space usage (0.0% throughout entire dataset)
Temp_C	Core temperature	°C	Thermal zone 0 via /sys/class/thermal
Load_1m	1-min load average	—	1-minute exponential moving average of run queue
Procs_Total	Process count	count	Total active processes (ps -e)
Net_RX_KBps	Network receive	KB/s	Inbound throughput on primary interface, 1-second delta
Net_TX_KBps	Network transmit	KB/s	Outbound throughput on primary interface, 1-second delta
Net_Latency_ms	Network latency	ms	ICMP RTT average to 8.8.8.8 (3 packets)
Net_Loss_%	Packet loss	%	ICMP packet loss percentage to 8.8.8.8
Net_RX_Drops	Hardware RX drops	count	Cumulative NIC RX drop counter
Disk_R_KBps	Disk read	KB/s	Block device read throughput, 1-second delta
Disk_W_KBps	Disk write	KB/s	Block device write throughput, 1-second delta
Disk_Usage_%	Disk usage	%	Root filesystem (/) utilization percentage

Data Statistics (N = 9,409)

Metric	Mean	Std	Min	Max	Median
CPU_Usage_%	4.23	1.25	0	19	4.00
CPU_Freq_MHz	1,588.6	128.8	1,500	2,400	1,600
RAM_Usage_%	12.99	0.33	11.3	14.5	13.00
RAM_Avail_MB	7,014.3	28.4	6,873	7,142	7,018
Temp_C (°C)	50.15	5.64	35.3	63.4	48.50
Load_1m	0.241	0.131	0.00	1.13	0.220
Net_RX_KBps	321.9	223.5	0	1,289	237
Net_TX_KBps	352.7	231.4	0	2,824	262
Net_Latency_ms	53.07	33.79	43.29	2,373.2	51.23
Net_Loss_%	8.34	165.3	0	6,667	0

---

System Architecture

Farm Site (LAN)                Edge Node               Internet
┌──────────────┐               ┌─────────────────┐
│ 10× IP Cams  │               │  Raspberry Pi 5 │
│ H.264 · RTSP │──── RTSP ────▶│  8GB · NVMe     │──── RTMP ────▶ OK.ru Live
│ Dahua 5469   │               │  10× FFmpeg     │
│ 192.168.x.x  │               │  -c:v copy      │
└──────────────┘               │  10× systemd    │
                               │  logger.sh cron │──── ping ────▶ 8.8.8.8
                               └─────────────────┘
                                        │
                               pi5_metrics_research.csv
                               9,409 records · 18 metrics

Hardware: Raspberry Pi 5 (8 GB RAM), NVMe SSD via PCIe 2.0, Gigabit Ethernet, passive cooling only
OS: Raspberry Pi OS 64-bit (Debian Bookworm, Linux kernel 6.x)
Cameras: 10× Dahua IP cameras (PID: 5469ZS4B, H.264, RTSP port 554/TCP)
Streaming: FFmpeg -rtsp_transport tcp -c:v copy -c:a aac -b:a 64k → RTMP → OK.ru
Process management: 10 independent systemd units (farm-cam1.service – farm-cam10.service), Restart=always
Data collection: Custom logger.sh via cron every 5 minutes

---

Usage

Python (pandas)

import pandas as pd

df = pd.read_csv("pi5_metrics_research.csv", parse_dates=["Time"])
df = df.set_index("Time")

# Basic stats
print(df.describe())

# Filter stable network periods (no packet loss)
df_stable = df[df["Net_Loss_%"] == 0]  # N = 9,177 (97.5%)

# Thermal analysis — hourly average
hourly_temp = df.groupby(df.index.hour)["Temp_C"].mean()
print(hourly_temp)

Python — Load & Plot Core Temperature

import pandas as pd
import matplotlib.pyplot as plt

df = pd.read_csv("pi5_metrics_research.csv", parse_dates=["Time"])

plt.figure(figsize=(14, 4))
plt.plot(df["Time"], df["Temp_C"], linewidth=0.5, color="tomato")
plt.axhline(y=60, color="orange", linestyle="--", label="60°C boundary")
plt.axhline(y=80, color="red", linestyle="--", label="80°C throttle threshold")
plt.xlabel("Date")
plt.ylabel("Core Temperature (°C)")
plt.title("Raspberry Pi 5 — 32-Day Core Temperature Timeline")
plt.legend()
plt.tight_layout()
plt.savefig("temp_timeline.png", dpi=150)

R

library(tidyverse)

df <- read_csv("pi5_metrics_research.csv") %>%
  mutate(Time = as.POSIXct(Time))

# Summary statistics
summary(df %>% select(CPU_Usage_., Temp_C, Net_RX_KBps, Net_TX_KBps))

# Pearson correlation matrix
df %>%
  select(CPU_Usage_., CPU_Freq_MHz, Temp_C, RAM_Usage_.,
         Net_RX_KBps, Net_TX_KBps, Net_Latency_ms, Load_1m) %>%
  cor(method = "pearson")

---

Data Quality Notes

• Temporal continuity: 9,409 records across 32 days at exactly 5.0-minute intervals — no gaps detected
• Uptime monotonicity: Uptime_s increases continuously from 354,366 s to 3,176,634 s — confirms zero reboots
• Physical range validation: All values within Pi 5 hardware limits (Temp: 35.3–63.4°C; CPU freq in {1500, 1600, 1700, 1900, 2400} MHz)
• Network hardware stability: Net_RX_Drops constant at 6 throughout — no NIC queue overflow
• Packet loss events: 218 records with 100% packet loss represent genuine internet outage events (not measurement errors). Net_Latency_ms is empty in these rows — correct POSIX ping behavior. Filter with Net_Loss_% == 0 to obtain stable-network subset (N = 9,177, 97.5%)

---

Potential Use Cases

1. Edge computing benchmarking — Real-world Pi 5 workload profile under 10-channel continuous streaming; enables fair comparison with other SBC platforms (Jetson Nano, Rock 5B, Orange Pi 5)
2. Thermal modeling & prediction — Strong correlation between ambient temperature (proxied by network throughput) and device temperature provides training data for tropical-climate thermal prediction models
3. Anomaly detection & predictive maintenance — 32-day multivariate time series serves as normal-behavior baseline for training unsupervised anomaly detection models (Isolation Forest, LSTM Autoencoder)
4. Rural IoT network reliability analysis — Latency distribution and packet loss events characterize Vietnamese rural ISP quality under live-streaming load

---

Anonymization

The following sensitive information has been removed or replaced before public release:

Original	Replaced with
Camera IP addresses	192.168.x.xxx
Camera passwords (SC field)	XXXXXXXX
Camera serial numbers (SN)	XXXXXXXXXXXXXXX
RTMP stream keys	<STREAM_KEY_ANONYMIZED>
System username	<user>

The metric data in pi5_metrics_research.csv contains no personally identifiable information or network-identifying information.

---

Citation

If you use this dataset in your research, please cite:

@article{lhu2026pi5agri,
  title     = {A 32-Day Continuous Performance Dataset of Raspberry Pi 5 Deployed
               for 10-Channel Live Video Streaming in an Agricultural IoT Environment},
  author    = {TBD},  %% To be updated after paper acceptance
  journal   = {IEEE DataPort / Zenodo},
  year      = {2026},
  doi       = {TBD},  %% To be updated after paper acceptance
  note      = {Dataset. CC BY 4.0},
  url       = {https://huggingface.co/datasets/<your-username>/pi5-agricultural-iot}
}

Note: Author names and DOI will be updated upon paper acceptance. Dataset is fully functional for use and citation in the interim.

---

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
You are free to share and adapt the data for any purpose, provided appropriate credit is given.

See: https://creativecommons.org/licenses/by/4.0/

---

Contact

Faculty of Information Technology
Lac Hong University, Biên Hòa, Đồng Nai, Vietnam