Attributed Priority Queue for Computing Task Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scheduling methods for computing tasks in heterogeneous computing environments are inefficient, requiring numerous priority queues and resulting in significant CPU overhead, as they fail to effectively account for varying processing unit attributes and priorities.
Innovation Solution
The implementation of attributed priority queues using numerical representations that combine priority and attribute bits, allowing for efficient sorting and distribution of computing tasks across processing units, with mechanisms to detect and address notification events indicating inefficient task processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If separate priority queues are used for computing tasks requiring various attributes of processing units, then task distribution accuracy is improved, but device complexity and CPU overhead increase significantly
Solution Approach 1:
The patent merges multiple separate priority queues into a single unified priority queue. Each computing task is represented by a numerical value that encodes both its priority level and required processing unit attributes. This consolidation reduces the number of queues from potentially thousands (n!+2 for n attributes) to just one, while maintaining accurate task distribution through the encoded numerical representations.
Solution Approach 2:
The patent transforms the representation of computing tasks by encoding both priority and attribute information into a single numerical value. This parameter transformation allows the system to compare and sort tasks using simple numerical operations rather than complex multi-dimensional attribute matching, thereby reducing computational overhead while preserving distribution accuracy.
2Manufacturing precision
If multiple separate priority queues are maintained for different processing unit attributes, then task scheduling precision is improved, but productivity decreases due to significant CPU overhead
Solution Approach 1:
The patent combines multiple attribute-specific priority queues into a single queue, eliminating the need to maintain and manage numerous separate queue structures. This merging reduces memory access patterns and simplifies queue management operations, thereby improving CPU efficiency while preserving scheduling precision through encoded task representations.
Solution Approach 2:
The patent replaces complex mechanical-like operations of managing multiple separate queues with a simplified numerical comparison system. By encoding task attributes and priorities into numerical values, the system uses efficient bitwise and arithmetic operations instead of complex data structure manipulations, significantly reducing CPU overhead.
3Device complexity
If traditional scheduling methods are used without attributed priority queues, then device complexity is reduced, but task distribution efficiency and adaptability deteriorate
Solution Approach 1:
The patent creates a universal priority queue system that can handle multiple processing unit attributes simultaneously through a single queue structure. The numerical encoding scheme allows the same queue to efficiently manage tasks with different attribute requirements (such as GPU acceleration, memory capacity, processor speed) without requiring separate specialized queues for each attribute type.
Solution Approach 2:
The patent transforms traditional scheduling parameters into a unified numerical representation that captures both priority and attribute information. This parameter transformation enables efficient sorting and distribution using simple numerical comparisons, achieving high task distribution efficiency while maintaining system simplicity through a single queue structure.
4Adaptability or versatility
If numerous priority queues are used to account for processing unit attributes, then adaptability to heterogeneous environments is improved, but loss of time increases due to queue management overhead
Solution Approach 1:
The patent merges multiple attribute-specific queues into one unified queue, dramatically reducing the time required for queue management operations. By consolidating queue maintenance, sorting, and task assignment into a single structure, the system eliminates redundant operations across numerous separate queues while maintaining full adaptability to heterogeneous computing environments through encoded task representations.
Solution Approach 2:
The patent performs preliminary encoding of task attributes and priorities into numerical values before task submission to the queue. This preliminary action ensures that when tasks are sorted and distributed, the system can use efficient numerical comparisons rather than performing complex attribute matching operations, significantly reducing time loss while maintaining adaptability.
Data Source
AI summary
Aspects of the disclosure provide for mechanisms for scheduling computing tasks in a computer system. A method of the disclosure includes determining one or more attributes associated with a computing task, determining an ordered list of the attributes in view of priorities associated with the attributes, generating a first numerical representation of the attributes in view of the ordered list of the attributes, determining a second numerical representation of a priority of the computing task, and determining a third numerical representation of a total priority of the computing task in view of the first numerical representation and the second numerical representation.