Abstract: Considering access frequency differences of data in distributed storage systems, a heterogeneous variable fractional repetition (HVFR) code based on Huffman tree is proposed. First, taking the data blocks with different access frequencies as the weighted leaf nodes of the Huffman tree, the Huffman tree is constructed and the duplication of the data blocks is determined. Then, the pairwise balanced design is used to construct heterogeneous fractional repetition codes. Performance analysis and experimental results show that, compared with reed-solomon codes and simple regenerating codes, HVFR codes can significantly reduce the repair time and repair locality of the failed nodes, improves the parallel access speed of the hot data, and achieves load balance with low computation complexity.

Key words: distributed storage, hot and cold data, fractional repetition code, Huffman tree