Abstract:

中文摘要: 针对RAID存储容灾系统中数据存储的可靠性和扩展性等问题，提出一种具有较高容灾能力且易扩展的存储容灾方法，称之为随机阵列码。通过研究 GF(2)上随机矩阵列满秩的性质，并将其应用在RAID存储容灾方案中。首先，依据RAID存储系统的环境配置和容灾需求设置条带参数；其次，构建相应规模且满足特定性质的随机矩阵作为编码矩阵；最后，将原始数据等分成块，利用编码矩阵将其编码并折叠存储到不同磁盘上。当发生磁盘损毁、扇区失效等原因造成数据丢失时，可依据相应的校验矩阵及剩余的编码分块进行失效数据的高概率译码恢复，从而，实现了数据高效、可靠地容灾存储。实验验证及理论分析表明：1) GF(2)上的随机高矩阵，在随机概率 p=0.5，矩阵行列差 δ ≥15时，即具有高概率列满秩的性质；2)随机阵列码的编码参数，不再受到素数或有限域规模的限制，可灵活设置，其容灾能力也可根据容灾需求进行扩展，并可实现较多的容错模式；3)随机阵列码由于基于XOR运算，在均匀随机时与RS码、CRS码相比，具有较高的编译码速率，特别是在较大规模的编码构造中表现良好；4)随机阵列码随着规模的增长，可趋于近似MDS码，具有较高的存储空间利用率。基于随机阵列码高效，可靠，易扩展等特点，可实现一般化RAID存储容灾方案的构造，此外，也可与其他存储容灾技术结合使用，共同构建特定需求下的RAID存储容灾系统。Abstract:To improve the reliability and scalability of data storage in redundant arrays of inexpensive disks (RAID) storage system,a new kind of storage fault-tolerance method with high fault-tolerance and flexible scalability was proposed,named random RAID.A research on the properties of random matrices in GF(2) was conducted and applied in the RAID storage fault-tolerance system.At first,the stripe parameters were set by the storage environment configuration and fault-tolerance requirement,then a random matrix was created as the generator matrix,with corresponding scale and some specific properties.Finally,the origin data was split into blocks with equal size, and encoded by the generator matrix, then folded into different disks.When there were data loss caused by disk damage or sector failures,the lost data could be recovered by the corresponding parity-check matrix and the remaining encoded blocks with high probability,enabling efficient and reliable data storage.Theoretical and experimental results show that:1) When random probability p=0.5 and the subtraction of row and column δ ≥15,the random high matrix in GF(2) could be full column rank with high probability;2)The encoding parameters of random RAID weren't constrained by the prime or scale of finite field any more.Instead,they can be set flexibly.The fault-tolerance can also be scaled with different fault-tolerance requirements, and allow more error patterns;3)When using uniform random matrix and compared with RS and CRS,the random RAID can improve the speeds of encoding and decoding greatly thanks to XOR operations,especially in large scale coding constructions;4)With the growth of the scale,the random RAID tends to approximate MDS,thus realizing highly efficient storage.Due to the properties of efficiency,reliability and scalability,the random RAID can realize the general construction of RAID storage fault-tolerance system.Moreover,it can be combined with other fault-tolerance technologies to construct