| With the continuous acceleration of the digitalization process of various industries,the amount of data stored in large-scale distributed cloud storage systems continues to grow.In order to deal with the failure of cheap storage devices,the cloud storage system needs to use redundant codes to ensure the reliability of data storage and the efficiency of reading.Local Reconstruction Code(LRC)has low number of connected nodes and low repair disk I/O overhead during the repair process.It has been widely used on cloud storage systems such as Microsoft Azure.The explosive growth of data volume has led to a diversified trend of data.The difference has led to the existence of hot data,that is in the frequency of reading different data.It has become a research hotspot,that the coexistence of cold and hot data in distributed cloud storage systems.However,because hot data needs to be accessed and updated frequently,storage systems that use LRC will be affected by excessive local read pressure,resulting in system overload,temporary failures,and large network bandwidth consumption.Generally,degraded reads are used to deal with temporary failures.At present,erasure-encoded storage systems that coexist with hot and cold data blocks still have some problems.First,the existing erasure codes(including LRC)mostly regard all data blocks as homogeneous and they cannot handle local hot data blocks.High local throughput and low bandwidth consumption during t emporary fault repair;Second,the existing storage system lacks performance metrics to accurately measure the ability of encoding and storing local hot data,and it is difficult to adapt the parameter selection of the encoding scheme according to actual n eeds.Based on the above two key issues,the Scalable Local Reconstruction Code(SLRC)scheme is proposed,and the code parameter selection algorithm is given.The research contents and main contributions include:This article introduces the basic concept of Read Arrival Rate to lay the foundation for establishing the required new performance indicators.By analyzing the process of each data block in the LRC encoding system to complete the read task with each other,this paper derives the relationship between the actual read arrival rate before and after degraded read and the direct read arrival rate in the LRC storage system.On this basis,this paper studies how to reduce the network bandwidth consumption of degraded read in repairing temporary faults,and provides a theoretical basis for proposing a scalable coding scheme.Aiming at the first problem,based on the coding principle of local reconstruction codes,this paper gives design guidelines for scalable local reconstruction codes.The design of the scalable local reconstruction code is based on the above-mentioned analysis of degraded read,which has higher throughput of local data blocks,and reduces the resource consumption of temporary failures through degraded read.This paper derives the mapping relationship between its storage overload and reconstruction cost,and compares RS code with LRC on the lower bound of the tradeoff between storage overload and reconstruction cost.Aiming at the second problem,this paper establishes two metrics: Local Maximum Throughput(LMT)and penalty based on the concept of read arriva l rate,which ensures reasonableness and solves the lack of measuring local hot data.We simulate the performance of LRC and SLRC with multiple coding parameters in LMT and implementation cost,and proves the superiority of SLRC to store local hot data blocks.In addition,in view of the lack of data reading requirements and specific parameter selection in the second question,parameter selections in different situations are given,based on the limitations of storage space,reconstruction cost,and degraded reading traffic consumption in actual application scenarios.There are the code parameter selection algorithms of minimum reconstruction cost,minimum storage overload,minimum implementation cost,and minimum comprehensive performance evaluation parameters are respectively given,and a scalable SLRC parameter design scheme is realized.The throughput and delay are simulated and compared for the typical parameters of LRC and SLRC. |
