The invention discloses a prefetching method and
system based on a distributed block storage
system, a medium and a product, and relates to the technical field of electric
digital data processing. The method comprises the following steps: sampling and integrating historical access records to generate an access
record sequence, constructing a space-time fragment, combining the sequence to obtain a first matrix, performing quantization
processing to obtain second matrixes, and aggregating a plurality of second matrixes of a volume to generate a third matrix; predicting a fourth matrix in a future preset
time based on a third matrix by using a prefetching determination model which is trained in advance through
deep learning of a third matrix sample set with a real access
label; and calculating the priority
score of each data block according to the fourth matrix, forming a global
score list, screening the to-be-prefetched data block, and sending a lifting suggestion to the
data service unit of the corresponding node to complete the migration of the prefetched data block. According to the scheme, non-periodic and randomized access scenes can be accurately adapted, the prefetching real-time performance and accuracy are improved, the
resource consumption of invalid prefetching is reduced, and the IO response efficiency of the distributed hierarchical block storage
system is optimized.