Computing Task Operator Reconstruction for Lower External Data I/O
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing systems in neural network learning models experience inefficiencies due to frequent data read and write operations between computing devices and external data storage, which reduce task computing efficiency.
Innovation Solution
Reconstruct multiple operators in the operator execution logic graph into a single reconstructible operator structure, allowing data exchange without writing intermediate results to external storage, thereby reducing data operations and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple operators are executed separately with external data storage, then data exchange is performed, but the quantity of data read and write operations increases and computing efficiency decreases
Solution Approach 1:
The patent merges multiple operators that can be reconstructed into a single operator structure. This consolidation allows the system to treat multiple computational operations as one unified operator, thereby reducing the number of times intermediate results need to be written to and read from external data storage, which directly improves computing efficiency and reduces operational time.
Solution Approach 2:
The patent implements a nested operator structure where multiple operators are nested within a reconstructed operator framework. This nesting allows intermediate computational results to be retained within the nested structure rather than being externally stored, enabling seamless continuation of computations without repeated I/O operations to external storage.
2Productivity
If operator reconstruction is performed to reduce data operations, then computing efficiency improves, but operator execution logic graph complexity increases
Solution Approach 1:
The patent applies segmentation by identifying and separating operators that are suitable for reconstruction from those that are not. This selective approach allows the system to reconstruct only the appropriate operator subsets, managing complexity while achieving efficiency gains. The segmentation criterion is based on whether operators can be effectively merged without compromising computational integrity.
Solution Approach 2:
The reconstructed operator structure is designed to be universal, capable of encompassing multiple different operator types within a single unified framework. This multi-functionality allows the system to handle diverse computational operations through a standardized reconstructed operator interface, simplifying the overall execution logic while maintaining versatility.
Data Source
AI summary
Implementations of the present specification provide a computing task processing method and apparatus. In the computing task processing method, a first computing task can be obtained, and the first computing task includes a first operator execution logic graph used for task computing; operator reconstruction is performed on the first operator execution logic graph based on an operator dependency in the first operator execution logic graph and a data cache capacity for operator computing, to obtain a second operator execution logic graph; and a second computing task including the second operator execution logic graph is provided to a task execution apparatus for execution by the task execution apparatus.