Discrete Resource Model for Multi-Processor Power Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing configuration and scheduling approaches for computing platforms fail to optimally reduce power consumption within a given time constraint due to their inability to account for discrete computing resource operating modes and transition costs, leading to unrealistic representations of available resources and suboptimal task scheduling.
Innovation Solution
A sophisticated resource model that accounts for discrete operating modes and transition costs between modes is used, providing a detailed representation of computing platform resources and applications, which can be processed by an integer linear programming (ILP) solver to select optimal task schedules and operating configurations for minimizing power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If existing configuration/scheduling approaches are used that treat processors and buses as having continuous operational settings, then the modeling is simpler, but the power consumption minimization is suboptimal because transition costs between modes are not accounted for
Solution Approach 1:
The patent segments the continuous operational settings into discrete operating modes for each processor and bus. Each resource is divided into specific modes (e.g., idle, active, high-performance) with defined transition costs between them. This segmentation allows the ILP solver to explicitly account for transition costs when scheduling tasks, leading to more accurate power consumption calculations and optimized scheduling decisions.
2Productivity
If existing approaches treat each task in isolation, then the scheduling process is simpler, but the overall task execution schedule is suboptimal because transition costs between modes are not considered
Solution Approach 1:
The patent merges the scheduling of multiple tasks with the selection of operating modes and transition timing into a unified ILP optimization problem. Instead of scheduling tasks in isolation, the formulation simultaneously determines task assignment, mode selection, and transition timing across all tasks, accounting for the cumulative impact of transition costs on overall power consumption.
Solution Approach 2:
The patent performs preliminary characterization of discrete operating modes and transition costs before task scheduling. The resource model pre-defines all possible modes and their associated transition costs, which are then used by the ILP solver to make informed scheduling decisions that minimize total power consumption including transition overheads.
3Use of energy by moving object
If a resource model accounts for discrete operating modes and transition costs, then power consumption optimization improves, but the model complexity increases
Solution Approach 1:
The patent replaces complex continuous optimization mechanisms with a discrete ILP formulation. By substituting continuous operational settings with discrete modes and using integer linear programming, the model achieves tractable complexity while capturing the essential discrete nature of processor and bus operating modes. The ILP solver efficiently handles the combinatorial optimization problem without requiring complex continuous optimization algorithms.
4Measurement precision
If discrete operating modes and transition costs are accounted for in the resource model, then task scheduling accuracy improves, but the computational processing requirements increase
Solution Approach 1:
The patent changes the parameter representation from continuous operational settings to discrete modes with associated costs. This parameter transformation simplifies the computational problem by converting it into a discrete optimization framework that the ILP solver can handle efficiently. The discrete parameterization maintains scheduling accuracy while reducing computational complexity compared to continuous optimization approaches.
Data Source
AI summary
Energy management modeling and scheduling techniques are described for reducing the power consumed to execute an application on a multi-processor computing platform within a certain time period. In one embodiment, a sophisticated resource model which accounts for discrete operating modes for computing components/resources on a computing platform and transition costs for transitioning between each of the discrete modes is described. This resource model provides information for a specific heterogeneous multi-processor computing platform and an application being implemented on the platform in a form that can be processed by a selection module, typically utilizing an integer linear programming (ILP) solver or algorithm, to select a task schedule and operating configuration(s) for executing the application within a given time.