Memory Preprocessing for Reduced I/O Read Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing multi-level memory architecture in computer devices, with cache, primary memory, and external memory, faces inefficiencies due to limited I/O bandwidth, leading to frequent and time-consuming I/O read and write operations during machine learning processes, which hinder processor performance.
Innovation Solution
A pre-processing method that stores target output data in the primary memory to reduce data interaction between primary and secondary memory, thereby minimizing I/O read operations and enhancing processor speed and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If data is frequently read from external memory during machine learning operations, then the processor can access input data, but the I/O bandwidth limitation causes long occupation time and low processing efficiency
Solution Approach 1:
The patent applies preliminary action by pre-loading input data from external memory into primary memory before the processor needs it for operations. This advance data preparation ensures that when the processor requires input data, it can be quickly retrieved from primary memory rather than waiting for slow external memory access, thus reducing I/O occupation time and improving processor speed.
Solution Approach 2:
The patent uses primary memory as an intermediary between external memory and the processor. Instead of direct access between the processor and external memory, data flows through primary memory which acts as a buffer. This intermediary approach allows the processor to access data quickly from primary memory while external memory operations occur in the background, effectively decoupling the processor's operational speed from external memory bandwidth limitations.
2Quantity of substance
If data is stored in external memory to increase storage capacity, then more data can be stored, but frequent I/O read and write operations between primary memory and external memory reduce processing efficiency
Solution Approach 1:
The patent applies copying by creating copies of input data from external memory and placing them in primary memory. Instead of repeatedly accessing the original data in external memory, the system works with copies in primary memory. This copying strategy allows the processor to operate efficiently on data copies while the original data remains stored in external memory, thus maintaining both high storage capacity and high processing efficiency.
Solution Approach 2:
The patent pre-transfers data from external memory to primary memory before processing operations begin. This preliminary data movement prepares the working data in fast primary memory in advance, so that subsequent processing operations can proceed at high speed without being bottlenecked by slow external memory access, thereby improving overall processing efficiency while maintaining adequate storage capacity in external memory.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The present disclosure provides a pre-processing method and device of data, a computer device, and a storage medium. By storing the target output data corresponding to the target operation into the first memory close to the processor and reducing the times of reading the target output data, the occupation time of the I/O read operation during the operation process can be reduced, so that the speed and efficiency of the processor can be improved.