Beyond the MDS Bound in Distributed Cloud Storage
Beyond the MDS Bound in Distributed Cloud Storage
复制标题
DOI：
10.1109/tit.2020.2993038
复制
发表时间：
2020-05
影响因子：
2.5
通讯作者：
Jian Li;Tongtong Li;Jian Ren
中科院分区：
文献类型：
--
作者：
Jian Li;Tongtong Li;Jian Ren
再生代码是一类分布式存储代码，它可以最佳地利用每个节点存储的数据量来修复故障节点。在最优再生平衡曲线上存在两个极值点，分别对应于最小存储再生（MSR）和最小带宽再生（MBR）。最近，基于RS码的再生码（RS-RC）在乘积矩阵框架下被构造。它还可以实现最大距离可分离（MDS）性质的代码再生和重建。然而，如果网络是敌对的，存储节点可能会受到损害或数据包被修改，重新生成或重建原始文件所需的存储容量和带宽可能会受到显着影响。在本文中，我们提出了基于厄米特码再生码（H-RC）的产品矩阵框架下的最小存储再生（H-MSR）和最小带宽再生（H-MBR）的发展建设。我们还提出了数据再生和重建算法的H-MSR和H-MBR码下的无错和敌对的网络。我们证明，所提出的算法也可以成功地确定在敌对网络中的错误解码。理论评估表明，我们提出的H-RC可以检测和纠正更多的错误，在敌对网络远远超过RS-RC相同的码率。我们的分析表明，建议的H-RC具有较低的计算复杂度比RS-RC的代码再生和代码重构。
Regenerating code is a class of distributed storage codes that can optimally trade the bandwidth with the amount of data stored per node to repair a failed node. There are two extreme points in the optimal regenerating trade-off curve, which correspond to minimum-storage regenerating (MSR) and minimum-bandwidth regenerating (MBR). Recently, Reed-Solomon (RS) code based regenerating codes (RS-RC) were constructed under the product-matrix framework. It can also achieve the maximum distance separable (MDS) property in code regeneration and reconstruction. However, in case that the network is hostile and the storage nodes could be compromised or packets be modified, the storage capacity and the bandwidth required to regenerate or reconstruct the original file can be significantly affected. In this paper, we propose Hermitian code based regenerating codes (H-RC) by developing constructions under the product-matrix framework for minimum storage regenerating (H-MSR) and the minimum bandwidth regenerating (H-MBR). We also propose data regeneration and reconstruction algorithms for both H-MSR and H-MBR codes under both error-free and hostile networks. We demonstrate that the proposed algorithms can also successfully determine the erroneous decodings in hostile networks. Theoretical evaluation shows that our proposed H-RC can detect and correct more errors in hostile networks well beyond the RS-RC with the same code rate. Our analysis shows that the proposed H-RC have lower computational complexity than the RS-RC for both code regeneration and code reconstruction.