Compute Resource Scheduling by Processor Power Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing resource scheduling methods result in prolonged waste of computing power due to processors with faster execution rates waiting for slower processors to synchronize, leading to inefficient utilization.
Innovation Solution
A resource scheduling method that selects processors based on computing power topology to ensure minimal deviation in execution times, allowing processors to synchronize closely and reducing idle time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the scheduler selects multiple processors of the same model to process the job in parallel, then the job can be executed using standardized resources, but the processors reach the synchronization point at different moments causing faster processors to remain idle
Solution Approach 1:
The patent changes the selection criterion from processor model (categorical parameter) to computing power (quantitative parameter). By measuring actual computing power through test jobs and selecting processors based on computing power similarity rather than model uniformity, the system achieves better synchronization while maintaining parallel execution efficiency.
Solution Approach 2:
The system introduces feedback by measuring actual processor performance through test jobs and using this information to inform scheduling decisions. The computing power topology is built based on measured performance data, creating a closed-loop system that adapts to actual processor capabilities rather than relying on theoretical specifications.
2Adaptability or versatility
If processors with different execution rates are used for parallel processing, then resource utilization flexibility increases, but computing power of faster processors goes unused during extended periods
Solution Approach 1:
The patent transforms the scheduling criterion from processor model (providing high flexibility) to computing power similarity (optimizing synchronization). This parameter change maintains adaptability by allowing different processor models to be grouped based on their actual performance characteristics, while eliminating energy waste through better-matched processor selection.
Solution Approach 2:
The system creates equipotential conditions by selecting processors with similar computing power levels, ensuring that all processors in a job experience comparable execution speeds. This balances the system state so that no processor is significantly faster or slower than others, preventing idle waiting and energy waste.
3Device complexity
If the scheduler selects processors based on model uniformity, then resource allocation is simplified, but processors reach synchronization points at different moments reducing overall efficiency
Solution Approach 1:
The system performs preliminary action by pre-measuring and recording the computing power of each processor before actual job scheduling. This advance characterization of processor capabilities allows the scheduler to make informed decisions without increasing runtime complexity, as the computing power topology is built in advance and reused for scheduling decisions.
Solution Approach 2:
The patent creates a virtual representation (computing power topology) that copies and stores the performance characteristics of physical processors. This virtual model allows complex performance-based matching without requiring complex real-time measurements, simplifying the actual scheduling operation while maintaining high parallel processing efficiency.
Data Source
AI summary
A resource scheduling method, a job processing method, a scheduler, a system, and a related device are related, and relates to the field of computer technologies. The scheduler obtains a to-be-processed first job; determines, based on a computing power topology, a plurality of first processors from a plurality of processors included in a resource pool, where the computing power topology indicates computing powers of the plurality of processors, and a difference between computing powers of the plurality of first processors does not exceed a threshold; and schedules the plurality of first processors for the first job, where the first job is executed in parallel based on the plurality of first processors. The difference between the computing powers of the plurality of first processors scheduled for the first job does not exceed the threshold.