Reducing Multi-Node Repair Bandwidth with ERN-MBR Codes

Minimum Bandwidth Regenerating (MBR) codes, grounded in the theory of network coding, represent a novel class of erasure codes. These codes can be employed in distributed storage systems to ensure data reliability and achieve low singlenode repair bandwidth comparable to that of replication-based methods. However, the inherent structure of MBR codes can result in significant storage overhead. To address this, we have identified that the recently proposed wide-stripe strategy, which increases the proportion of data nodes within the system, can be applied to MBR codes to reduce storage overhead. Nonetheless, the wide-stripe strategy increases the frequency of multi-chunk failures compared to traditional erasure coding methods. Since conventional MBR codes focus solely on optimizing single-node repair performance, they are insufficient for addressing the increased occurrence of multi-node failures. To overcome this challenge, we propose extra-redundant-node based MBR codes (ERN-MBR), which introduce an additional redundancy node to participate in the multi-node repair process, thereby significantly reducing the system's multi-node repair bandwidth. We conducted an analysis of the repair bandwidth and storage overhead associated with ERN-MBR codes. Experimental results show that with a stripe size of 128, ERN-MBR achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$10.78 \times$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$9.79 \times$</tex> higher repair throughput for 2-node and 3-node failures, respectively, compared to traditional MBR codes.