Computer Power Modeling Through Variable Resource Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies fail to predict power consumption at computer nodes in a hybrid cloud environment, hindering efficient utilization of renewable energy.
Innovation Solution
A computer system with a management node that measures and models power consumption by executing a power measurement program, adjusting resource allocation, and generating a power consumption model to predict power usage based on resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If data and application programs are disposed in appropriate bases to satisfy target performance, then execution time and cost are optimized, but power consumption cannot be predicted and renewable energy cannot be efficiently utilized
Solution Approach 1:
The system performs preliminary measurements of power consumption at computer nodes before actual data analysis workloads are executed. By measuring power consumption in advance under various conditions and storing this data in a database, the system enables subsequent prediction of power consumption for different workload scenarios without requiring real-time measurement during actual operations.
Solution Approach 2:
The system creates a power consumption model that copies and generalizes the measured power consumption characteristics from specific computer nodes to predict power consumption across different nodes and workloads. The model captures the relationship between resource allocation and power consumption, allowing virtual copying of power consumption patterns for prediction purposes.
2Measurement precision
If power measurement program uses hardware resources allocated to the program, then accurate power consumption measurement is achieved, but resource allocation for measurement affects program execution
Solution Approach 1:
The system segments the power measurement process from the actual program execution by using a dedicated power measurement program that operates independently. The measurement program is executed separately to gather power consumption data, which is then used to create models for predicting program execution power consumption, avoiding interference with actual workload performance.
Solution Approach 2:
The system introduces a power consumption model as an intermediary between the power measurement program and the actual program execution. The model serves as a mediator that translates measured power consumption data into predictions for different resource allocation scenarios, allowing accurate measurement without directly affecting program execution performance.
Data Source
AI summary
The CPU of the management node measures the power consumption of a computer node while causing the computer node to execute the power measurement benchmark that uses hardware whose resource is allocated to a program to be executed by the computer node, where the CPU is changing the use amount of the resource while causing the computer node to execute the power measurement benchmark. The CPU generates a power consumption model representing a relationship between an allocation amount of the resource to be allocated to the program and the power consumption on the basis of a measurement result obtained by measuring the power consumption.