CPU Core Power Estimation for Real-Time Workload Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power monitoring solutions for computing devices, such as edge servers in telecommunication networks, require significant computing resources and are not suitable for real-time performance, especially those with low computational capabilities.
Innovation Solution
A system that measures total power consumption, processor core residency in different states, and utilization to estimate power consumption of processes, using a single energy exporter agent per worker node, without requiring a full containerized application.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If prior power monitoring solutions (e.g., KEPLER) are used, then power usage can be monitored, but significant computing resources are required on each computing device, further adding to power consumption
Solution Approach 1:
The system segments power monitoring functionality by separating the measurement agent (running on the monitored computing device) from the analysis system (running on a separate computing device). This allows the monitored device to perform only lightweight measurements and reporting, while the computationally intensive analysis is performed remotely, reducing the power consumption burden on individual computing devices.
Solution Approach 2:
The system introduces an intermediary analysis system that receives power measurement data from multiple computing devices and performs the computationally intensive analysis and estimation. This intermediary handles the heavy processing workload, allowing the original computing devices to maintain low power consumption while still enabling comprehensive power monitoring and estimation capabilities.
2Measurement precision
If prior power monitoring solutions are used, then power analysis can be performed, but real-time performance is compromised due to offline analysis requirements
Solution Approach 1:
The measurement agent continuously collects and buffers power-related metrics (CPU usage, memory usage, storage I/O, network I/O) in real-time before analysis is performed. This preliminary data collection ensures that when analysis is needed, the data is already prepared and available, enabling faster response times and real-time power estimation without requiring computationally intensive real-time analysis on the monitored devices.
3Measurement precision
If prior power monitoring solutions are used, then power estimation can be performed, but the approach is not suitable for installations with low computing resources, such as edge servers
Solution Approach 1:
The system extracts the computationally intensive analysis and estimation functions from the monitored computing devices and places them on a separate analysis system. The computing devices only need to run a lightweight measurement agent that collects and reports metrics, significantly reducing their computing resource requirements and making the solution suitable for resource-constrained environments like edge servers.
Solution Approach 2:
The measurement agent implements only the essential measurement and reporting functions needed for power monitoring, leaving the complex analysis and estimation tasks to the external analysis system. This partial implementation on the monitored device minimizes resource consumption while still enabling comprehensive power estimation capabilities through the combined system.
Data Source
AI summary
An energy estimator runs workloads to determine power consumed by processor cores in various power states and utilizations. This power consumption data is then used by an energy export agent for further calculations on different scenarios and cases. The energy export agent reads a total power consumption value from a CPU, data describing execution of processes from an operating system, and information regarding software components executing on the CPU from an orchestrator control plane. The agent uses data describing states of operating of the processor cores of the CPU and the total power consumption value to estimate power consumed by each processor core. Allocation and utilization of each processor core by a process and/or a container that spawned a process are then used to determine power consumption attributable to the process and/or container. A software hierarchy may be used to aggregate power consumption for applications, namespaces, or other software components.