Abstract: Considering the diversity of node failures in the actual distributed storage systems, group repairable codes based on non-uniform cyclic coding (GRC-NCC) are proposed in this paper, to protect the nodes with high failure rate more effectively. Specifically, storage nodes are grouped non-uniformly according to the node failure rate, data blocks are sequentially stored into node groups with increasing length, and then group coding blocks and global check blocks are generated by using cross-band cyclic coding. Performance analyses and experimental simulations show that GRC-NCC has lower repair bandwidth overhead and repair locality during repairing the failed nodes, and better fault tolerance in the process of multi-node fault repair compared with reed-solomon (RS) codes and rotated group repairable codes (RGRC), although GRC-NCC has higher storage overhead than RS codes.