Abstract:Array codes are widely used to improve the fault tolerance ability of redundant arrays inindependent disks(RAID) storage system.With the in-depth understanding of RAID storage system,the update efficiency of array codes has gradually become an important performance metric.In order to deal with the low update efficiency problem of current triple fault tolerant horizontal array codes,a new kind of low-density and triple fault tolerant horizontal array code called inverse code was proposed.Unlike the traditional array codes whose encoding process was determined by using special geometric methods,the encoding process of inverse codes were determined by constructing coding distribution matrix in generator matrix.Firstly,a matrix structure based on GF(2^w) called inverse structure matrix was proposed,which is a super-regular matrix and contains only three rows of elements.Secondly,bit square matrices of size w×w were used to represent the elements of GF(2^w).Finally,the coding distribution matrix with low density property was produced by a new optimized algorithm and used to determine the encoding process of inverse code.Theoretical analysis showed that the inverse code was maximum distance separable and obtained the optimal storage efficiency.Moreover,the parameters of inverse code were not constrained to be prime numbers and more continuous compared with other triple fault tolerant horizontal codes such as STAR code and RDP code.Experimental results showed that inverse code had the advantages in sparse degree,update efficiency,encoding and decoding efficiency over CRS code whose encoding process was determined by constructing generator matrix.Compared with STAR code and RDP code,the update efficiency of inverse code was improved by 20 % on average.Besides,different kinds of ‘XOR scheduling’ techniques were used to improve the decoding efficiency of inverse code.