SEED: Streaming Energy Consumption and Emissions Dataset

The Streaming Energy consumption and Emissions Dataset (SEED) is an open and FAIR-compliant benchmark designed for energy- and carbon-aware video streaming research. It provides segment-level measurements of energy usage and CO₂ emissions across cloud provisioning and end-user consumption stages: encoding, storage, retrieval, decoding, display.

📄 The complete dataset documentation and methodology are described in the paper:
"SEED: Energy and Emission Estimation Dataset for Adaptive Video Streaming" [PDF]

Dataset Structure

SEED/
├── README.md
├── LICENSE
├── Provisioning_Encoding_Storage.csv
├── Consumption_Retrieval_Decoding.csv
└── Consumption_Display.csv

Each CSV file contains detailed energy and emission measurements for over 500 video segments, under various configurations and conditions.

---

📚 CSV File Descriptions

1. Provisioning_Encoding_Storage.csv

Measures server-side energy and emissions during video encoding and storage.

Column	Description
video, codec, resolution, width, bitrate [kb/s]	Video and encoding parameters
instance_type, cpu_model	Cloud instance details
encoding_duration [s], average_cpu_utilization [%]	Encoding benchmarks
cpu_energy [kWh], storage_energy [kWh]	Energy consumed
provisioning_energy_<region>_<PUE> [kWh]	Region-specific PUE-adjusted energy
emission_<stage>_<region>_<CI> [g]	CO₂ emissions (see below)

---

2. Consumption_Retrieval_Decoding.csv

Captures client-side retrieval (LTE/Ethernet), decoding energy, and emissions.

Column	Description
video, codec, resolution, width, bitrate [kb/s]	Video and encoding parameters
net, userDevice, cpu_model	Network and hardware setup
decoding_duration [s], average_cpu_utilization [%]	Decodeing benchmarks
energy_decode [Wh]	Decoding energy
energy_NIC_<ETH/LTE>_<throughput> [kWh]	NIC energy at various throughputs
emission_<NIC>_<region>_<CI> [g]	CO₂ emissions from network usage

---

3. Consumption_Display.csv

Contains display energy consumption data.

Column	Description
video	Video segment identifier
display_energy [kWh]	Energy for rendering the video
emission_display_<region>_<CI> [g]	Associated CO₂ emissions

---

⚙️ Benchmark Summary

• VCD – Video Complexity Dataset: https://ftp.itec.aau.at/datasets/video-complexity/

• VEED – Video Encoding Energy and CO₂ Emissions Dataset: https://github.com/cd-athena/VEED-dataset

• Codecs: HEVC and AVC

• Resolutions: From 360p to 2160p

• Bitrates: 145 kb/s to 16800 kb/s

• Provisioning: AWS EC2 instances across Frankfurt, Ireland, Paris, and Northern Virginia

• Networking: LTE dongle and Ethernet at various throughputs (16.7 to 60 Mb/s)

• Decoding: Intel Xeon Gold 5218 on a PC

• Display: LG OLED 4K Monitor (model 27EP950-B)

---

🌍 Region-Specific PUE

PUE values obtained from AWS Cloud.

• Frankfurt: 1.35
• Ireland: 1.11
• Paris: 1.04
• Northern Virginia: 1.15

---

🏭 Regional Carbon Intensity (CI)

CI values obtained from Electricity Maps on July 11, 2025, at 2:00 AM (GMT-5):

• Frankfurt: 334 gCO₂e/kWh
• Ireland: 494 gCO₂e/kWh
• Paris: 13 gCO₂e/kWh
• Northern Virginia: 472 gCO₂e/kWh
• Klagenfurt: 157 gCO₂e/kWh

---

💻 Instance Types

Experiments used the following EC2 instance types:

c5.large, c5.xlarge, c5.2xlarge, c5.9xlarge, m5.2xlarge, r5.2xlarge