Data-Parallel Graph Aggregation for Multigrid Setup
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graph aggregation techniques are inefficient, especially as data sets grow in size, and are limited by serial methodologies, necessitating a more efficient approach for performing graph matching in a data-parallel manner.
Innovation Solution
The system performs graph aggregation by computing a graph matching in a data-parallel manner, utilizing methods such as one-phase and two-phase handshaking, and implementing it on a parallel architecture like a GPU, allowing for simultaneous processing of vertices and edges without requiring graph coloring or a greedy method, and integrating it with multigrid algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current serial methodologies are used for graph aggregation, then implementation simplicity is maintained, but processing efficiency deteriorates as data sets increase in size
Solution Approach 1:
The patent segments the graph aggregation process into independent parallel operations including vertex degree computation, edge selection, and aggregate formation. Each vertex can be processed independently to compute its degree, edges can be evaluated in parallel for inclusion in aggregates, and multiple aggregates can be formed simultaneously across different processors or GPU cores, thereby improving productivity while managing complexity through modular decomposition
Solution Approach 2:
The patent transitions from serial processing to parallel processing by utilizing a second dimension of computation through multiple processors or GPU cores. This dimensional shift allows simultaneous evaluation of multiple vertices and edges, transforming the computational approach from sequential (one operation at a time) to concurrent (multiple operations simultaneously), thereby resolving the efficiency-complexity contradiction
2Quantity of substance
If data sets utilized during aggregation increase in size, then problem resolution capability is improved, but processing inefficiency is amplified
Solution Approach 1:
The patent divides large data sets into manageable segments that can be processed in parallel. Vertices are partitioned across multiple processing units, and edge evaluations are distributed independently, allowing the system to handle larger data sets without proportionally increasing processing time. This segmentation enables linear or near-linear scaling of data set size while maintaining processing efficiency
Solution Approach 2:
The patent performs preliminary computations such as vertex degree calculations and edge eligibility determinations before the main aggregation process. By pre-computing these values and storing them in data structures optimized for parallel access, the system prepares the data set for efficient parallel processing, thereby handling larger data volumes without amplifying processing inefficiency
Data Source
AI summary
A system, method, and computer program product are provided for performing graph aggregation. In use, a graph with a plurality of vertices and a plurality of edges is identified. Additionally, aggregation is performed on the vertices and edges of the graph by computing a graph matching, where such graph matching is performed in a data-parallel manner.