Design of AI-enhanced hybrid storage engine for multimodal data management

To enhance the management effect of multimodal data and increase the data access speed, this paper first uses dynamic random access memory (DRAM) to complete the caching of non-volatile memory (NVM) in the hybrid storage module. When there is a cache missing in DRAM, a high-speed acquisition card is used to collect historical access records of NVM. After encoding historical access records into access vectors, they are used as the input of the deep learning model. The spatio-temporal attention mechanism is introduced to enhance access coding features and improve the prediction accuracy of the access frequency. The multimodal data with prediction results higher than the set threshold are read into the hybrid storage module for storage. Experimental outcome implies that the average performance of the offered approach in sequential reading is at least 1.1 times that of the benchmark approach, significantly improving the access speed.