Directed-Graph Scheduling With Parallel Algorithm Node Groups
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing tasks in computer vision and graphics are inefficient due to algorithms being executed sequentially, which affects throughput rate and calculation efficiency.
Innovation Solution
A scheduling method that groups algorithm nodes into parallel node groups and schedules them in series, allowing parallel execution of processing algorithms within each group to improve efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If algorithms are executed sequentially in a certain order, then the execution order is simple to control, but the throughput rate and calculation efficiency are low
Solution Approach 1:
The patent segments the algorithm nodes into multiple independent groups based on data dependency relationships. Each group can be executed independently in parallel, transforming a single sequential execution stream into multiple parallel execution streams. This segmentation resolves the contradiction by enabling parallel execution (improving throughput) while maintaining clear group boundaries (controlling scheduling complexity).
Solution Approach 2:
The patent transitions from one-dimensional sequential execution to two-dimensional parallel execution by introducing the group dimension. Algorithm nodes are arranged in groups where intra-group nodes can execute in parallel while inter-group nodes maintain sequential dependency. This dimensional change enables simultaneous execution of multiple algorithms without violating dependency constraints, thereby improving throughput rate while keeping scheduling manageable through the group structure.
2Productivity
If algorithms are executed sequentially, then resource allocation is simple, but calculation efficiency is affected
Solution Approach 1:
By segmenting algorithm nodes into independent groups, the system can allocate computing resources to multiple groups simultaneously. Each group receives dedicated resource allocation while groups execute in parallel, thereby improving calculation efficiency. The segmentation maintains allocation simplicity at the group level while enabling parallel resource utilization.
Solution Approach 2:
The patent merges multiple algorithm nodes into execution groups that can be allocated resources collectively. By combining nodes with similar dependency characteristics into the same group, the system achieves efficient bulk resource allocation while maintaining the ability to parallelize across groups, thus improving calculation efficiency without proportionally increasing allocation complexity.
3Productivity
If algorithm nodes are executed in parallel, then the throughput rate improves, but the scheduling complexity increases
Solution Approach 1:
The segmentation of algorithm nodes into dependency-based groups provides a natural scheduling hierarchy. Within each group, nodes can execute in parallel without complex coordination, while between groups, sequential execution follows clear dependency rules. This segmentation enables parallel execution (improving throughput) while maintaining simple scheduling logic at each level.
Solution Approach 2:
The patent implements dynamic scheduling where the parallel execution of groups is determined by runtime dependency evaluation. The system dynamically identifies which groups are ready for execution based on completed predecessor groups, enabling flexible parallelization without requiring static complex scheduling tables. This dynamic approach improves throughput while keeping scheduling adaptable and manageable.
Data Source
AI summary
The embodiment discloses a scheduling method, apparatus, device and storage medium. The method includes: obtaining an algorithm directed graph corresponding to a target task, the algorithm directed graph comprising a plurality of algorithm nodes; grouping the plurality of algorithm nodes in the algorithm directed graph to obtain a plurality of node groups; scheduling the plurality of node groups in series, and scheduling a processing algorithm corresponding to at least one algorithm node in the node group in parallel.