Neural Network Query Processing on DNN Runtimes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Query processing systems configured for CPUs are unable to leverage the benefits of specialized hardware and hardware acceleration techniques tailored for deep learning operations, preventing efficient execution of database operations on deep learning systems.
Innovation Solution
A query processing system that generates a deep neural network (DNN) program from database queries and executes it over a multi-platform DNN runtime, utilizing DNN operations to perform queries and improve performance while reducing development effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If query processing systems use CPU-based configurations, then they maintain compatibility with traditional database operations, but they cannot leverage specialized hardware acceleration for deep learning operations
Solution Approach 1:
The query processing system is designed to execute both traditional SQL queries and deep learning operations through a unified framework. The system translates SQL query operators into corresponding deep learning framework operators, enabling a single system to handle multiple types of operations across different hardware platforms (CPU, GPU, TPU) without requiring separate processing systems.
2Productivity
If specialized hardware is used for deep learning operations, then execution performance improves, but query processing systems cannot access these hardware benefits
Solution Approach 1:
The system acts as an intermediary between traditional query processing systems and specialized deep learning hardware. It translates SQL query operators into deep learning framework operators that can be executed on GPU, TPU, or other specialized hardware, thereby enabling access to hardware acceleration benefits while maintaining compatibility with existing query interfaces.
3Productivity
If query operators are translated to neural network operators, then deep learning hardware acceleration is enabled, but system complexity increases
Solution Approach 1:
The system performs query translation in advance by converting SQL query operators into deep learning framework operators before execution. This preliminary translation step enables the use of optimized deep learning runtime systems and hardware acceleration, while the translation logic is encapsulated to minimize ongoing complexity during query execution.
Data Source
AI summary
Example aspects include techniques for query processing over deep neural network runtimes. These techniques may include receiving a query including one or more query operators and determining a query representation based on the one or more query operators. In addition, the techniques may include determining a neural network program based on the query representation, the neural network program including one or more neural network operators for performing the query in a neural network runtime, generating a neural network data structure based on a dataset associated with the query, and executing the neural network program in the neural network runtime over the neural network data structure to generate a query result.