In distributed storage system for big data

the repair pipelining (RP) reduces repair time by 90%

which effectively solves the problem that erasure code is not suitable for storing hot data due to the heavy overhead of repair time. However

the existing RP has the problem of unbalanced load among nodes

which leads to the degradation of system performance. In this paper

a repair pipelining based on node load balancing (NLB-RP) is designed

and then the algorithms for calculating the load of nodes and repair time according to the evaluation index of performance are proposed. Theoretical analysis and experimental results both show that

from local to global

the NLB-RP can effectively balance and reduce the load of nodes without introducing extra repair cost. Compared with the RP

the load variance of the NLB-RP is zero

that is

the load of each node is the same. Thus

the proposed NLB-RP has the optimal load balance.