CAJ | 학술논문

在云存储中，数据可以分为三种类型：读频繁、写频繁以及冷数据。对不同类型的数据设置不同的编码机制：对读频繁数据采用多副本技术，频繁写的数据采用单文件编码方案，冷数据采用跨文件编码方案。在云存储系统中提出了一种新的纠删码技术的可靠性机制-魔方码，该编码易于实现，具备高的容错性并且相比多副本机制可以减少资源的消耗。以HDFS上的一个RAID包为基础，实现魔方码的原型，并添加BlaumRoth、RDP以对比测试。虽然各种编码各有优缺点，通过单机和集群测试RAID-5、Blaum-Roth、RDP和魔方码的编译码速度，魔方码译码速度比现有最优RAID-6编码RDP码提高了一个数量级。同时，针对大规模分布式系统提出了一种三维纠删码，可以容任意7个错误。总之，测试结果显示魔方码相比三份副本技术，在减少一半资源消耗的同时可以保证高的可用性和可靠性。
재운존저중，수거가이분위삼충류형：독빈번、사빈번이급랭수거。대불동류형적수거설치불동적편마궤제：대독빈번수거채용다부본기술，빈번사적수거채용단문건편마방안，랭수거채용과문건편마방안。재운존저계통중제출료일충신적규산마기술적가고성궤제-마방마，해편마역우실현，구비고적용착성병차상비다부본궤제가이감소자원적소모。이HDFS상적일개RAID포위기출，실현마방마적원형，병첨가BlaumRoth、RDP이대비측시。수연각충편마각유우결점，통과단궤화집군측시RAID-5、Blaum-Roth、RDP화마방마적편역마속도，마방마역마속도비현유최우RAID-6편마RDP마제고료일개수량급。동시，침대대규모분포식계통제출료일충삼유규산마，가이용임의7개착오。총지，측시결과현시마방마상비삼빈부본기술，재감소일반자원소모적동시가이보증고적가용성화가고성。
In cloud storage,data are divided into three types:the heavy-read,the heavy-write and the cold.For data in different types, different coding mechanism is configured,namely,the multi-replication for heavy-read,the single-file coding for heavy-write,and the cross-file coding for cold data.We present in cloud storage system a new reliability mechanism-the Cubic code,as it is called,for erasure coding technology.Such code is easy to implement,has high fault tolerance,and can cut down the resources consumption comparing with the multi-replication mechanism.We implement the prototype of Cubic code by taking a RAID packet on HDFS as the basis,and add BlaumRoth and RDP for contrastive test.Although each code has its own pros and cons,by stand-alone and cluster testing the encoding and decoding speeds of RAID-5 ,ClaumRoth,RDP and Cubic code,the last is faster by one order of magnitude in decoding speed than that of the currently best RDP code of RAID-6 coding.Meanwhile,we also present a 3D erasure coding for large-scale distribution system,it can tolerate any 7 faults.In summary,the test results show that the Cubic code can guarantee high availability and reliability while reducing half resources con-sumption compared with three-way multi-replication technology.