分布式存储中的克隆piggybacking框架

Cloned piggybacking framework for distributed storage

随着5G技术的快速发展,分布式存储系统凭借低成本、高可用性、高吞吐量和海量存储能力等特点,越来越受到人们的青睐。由于节点众多,节点故障频发,因此采用一定的数据容错技术来保证数据的可靠性尤为重要。近年来,piggybacking框架以其优异的修复性质受到了广泛关注。从piggybacking框架的修复度入手,为了同时降低修复带宽和修复度,提出了克隆piggybacking框架。与其他piggybacking框架不同,该框架只对一组节点进行嵌入设计和修复分析,其他组内对应节点通过克隆的方式简化设计过程。根据“一个节点的信息符号尽可能地只嵌入一个子条带”的设计思想,权衡修复带宽和修复度。与现有的piggybacking框架相比,克隆piggybacking框架设计简单,进一步降低了信息节点的修复带宽。

With the rapid development of 5G technology,distributed storage systems are popular with its features such as low cost,high availability,high throughput and massive storage capacity.Due to the frequent node failures,it is important to adopt data fault tolerance technology to ensure the reliability of data.In recent years,the piggybacking framework has received extensive attention for its excellent repair properties.In this paper,we propose a cloned piggybacking framework design to optimize average repair bandwidth and repair locality.Different from other piggybacking frameworks,the piggybacking and repairing methods are carried out on one set of nodes.Then the corresponding nodes in other groups are piggybacked in the same way.The trade-off between repair bandwidth and repair locality is established by the idea of“one parity node only piggybacks symbols from one substripe”.Compared with the existing piggybacking framework,the cloned piggybacking framework is simple in design and can further reduce the repair bandwidth of information nodes.