SIMD Workload Distribution for Ontology Tagging Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing workload management solutions for concept tagging with large ontology files suffer from coarse granularity, leading to inefficient use of computing resources and high latency due to their non-SIMD based architecture.
Innovation Solution
A system and method for distributing workload over a SIMD based parallel processing architecture, where an ontology file is loaded to determine synonyms, concepts, and characters, and a mapping relationship is established between synonyms and concepts, with arrays formed to store length, characters, and start positions, allowing concurrent access by software threads on a SIMD processing device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If non-SIMD based architecture (CPU) is used for ontology tagging, then multiple instructions can be executed simultaneously on multiple processors, but coarse granularity results in inefficient use of computing resources and high latency
Solution Approach 1:
The patent segments the ontology file into multiple synthetic records, each containing a concept identifier and its associated synonyms. This segmentation transforms the large ontology file into manageable units that can be distributed across thousands of SIMD processing units, enabling fine-grained parallel processing and eliminating the coarse granularity limitation of traditional CPU-based approaches.
Solution Approach 2:
The patent transitions from traditional CPU-based sequential/instruction-level parallelism to SIMD-based data-level parallelism by organizing ontology data into arrays that can be processed simultaneously across multiple processing units. This dimensional change in processing architecture enables thousands of operations to execute in lockstep, dramatically reducing latency while improving resource utilization.
2Adaptability or versatility
If large ontology files are processed with traditional workload management, then concept tagging can be performed, but computing resources are not optimized due to coarse granularity
Solution Approach 1:
The patent divides the large ontology file into multiple synthetic records with fixed-size buffers, allowing the system to handle large ontology files while maintaining fine-grained control over processing units. Each synthetic record represents a manageable unit that can be independently processed, enabling efficient resource allocation across the SIMD architecture.
Solution Approach 2:
The patent changes the parameter of data organization from traditional processor-centric structures to array-based structures optimized for SIMD processing. By organizing ontology data into arrays of synthetic records with standardized fields (concept identifier, synonyms, bitmasks), the system achieves both adaptability to large files and optimal utilization of computing resources through fine-grained parallel processing.
Data Source
AI summary
A method of workload management for distributing workload over a single instruction and multiple data (SIMD) based parallel processing architecture includes pre-processing of vert large ontology file on the host processor. The outcome of the preprocessing step is a set of arrays which are loaded on the SIMD based parallel processing architecture to process the ontology file over input text documents of any kind and generate the tagged outcome. The method provides the maximum granularity to facilitate allocation of the maximum number of software threads on the SIMD based parallel processing architecture to achieve minimum document processing latency.