Scope
The paper introduces VEEP, an architecture designed to predict energy consumption and CO2 emissions in cloud-based video encoding. VEEP combines video analysis with machine learning to accurately estimate CPU energy usage and CO2 emissions during the encoding process. Trained on a comprehensive dataset and encoding results from AWS EC2 instances, VEEP achieves high accuracy, with potential emission reductions of up to 375 times.
The paper highlights the importance of considering environmental factors in cloud computing and proposes a scheme that can predict energy consumption and CO2 emissions with remarkable precision. Key contributions include a detailed methodology, experimental design, and results demonstrating the effectiveness of VEEP in predicting energy consumption and CO2 emissions.
Overall, VEEP represents a significant advancement in addressing environmental concerns in cloud-based video encoding.
Summary
The proposed Video Encoding Energy and CO2 Emission Prediction (VEEP) scheme leverages machine learning to forecast energy consumption and CO2 emissions in cloud-based video encoding. The methodology entails analysing video complexity, predicting energy consumption, fetching real-time carbon intensity data, and subsequently estimating CO2 emissions based on energy usage and carbon intensity.
Some of the key aspects of the paper are:
- A scheme capable of accurately predicting CPU energy consumption during the video encoding process with high-precision metrics.
- The ability to calculate CO2 emissions for encoding a video segment within a specified country based on real-time energy mix and carbon intensity.
- Demonstration of significant potential reductions in CO2 emissions by optimising cloud instance types and locations.
The architecture of VEEP comprises five key modules:
- Video analyser: Initiates the process by analysing video frames to extract complex features.
- Energy predictor: Employs an ML model to predict energy consumption based on video complexity features, instance types, and encoding parameters.
- CPU Energy Consumption: This metric quantifies the amount of energy consumed by the CPU during the video encoding process. It is a primary focus of the paper as VEEP aims to predict CPU energy consumption accurately. CPU energy consumption can be measured in various units, such as watt-hours (Wh) or joules (J), which quantify the amount of electrical energy consumed over a period of time.
- CO2 calculator: Estimates carbon emissions by fetching real-time carbon intensity data and multiplying it by the predicted CPU energy consumption.
- CO2 data source: Accesses an API to retrieve real-time carbon intensity data for various countries, providing detailed representations of carbon intensity.
Relevance for EXIGENCE
EXIGENCE Relevance for requirements and scenarios, energy metrics, energy measurements, and resource interconnectivity, orchestration and energy optimisation. The CPU energy consumption units that could be used as a global standard. Also, the potential use of real time data for the consumption of energy by the project’s proposed solutions could be a potentially relevant input. Moreover, the use of that data could be beneficial in future expansion of the Use Cases and scenarios that would be investigated in the project. The drawback of such endeavor, on the other hand, is the introduction of the difficulty to acquire those data (e.g. from the power network of a country) and how that data could be used. The correlation of CO2 and energy consumption could provide an insight for a proposed green solution.
