Design And Implementation Of LRC Acceleration Optimization Based On Cheap GPU

In order to store diversified data more efficiently,Locally Repairable Codes(LRC)has been introduced into the storage system.LRC can ensure reliable storage while significantly reducing the cost of data reconstruction,so it is suitable for cold data storage.But with the development of storage systems,the performance of LRC has gradually become an important bottleneck limiting the development of storage systems.Therefore,it is urgent to make full use of the low-cost GPU used by the storage nodes and design an optimized LRC which can significantly improve the coding performance.In order to make use of LRC to improve the performance of storage system,an accelerated optimization design and implementation of LRC based on cheap GPU has been proposed.These following works have been done:(1)An accelerated optimization implementation of LRC based on cheap GPU is proposed.By directly exclusive or the original data block,the corresponding check block is generated,so as to reduce the resource consumption in the process of building generation matrix and executing matrix multiplication;(2)To ensure data between CPU and GPU and avoid data loss,a batch coding strategy is adopted for reliable transmission;(3)To reduce the time delay in data reading,GPU is used to share memory for data storage;(4)in order to accelerate the generation of local check blocks and ensure the rich information contained in the check blocks,a method of building check blocks based on local check blocks is proposed.The following conclusions can be drawn through multiple control experiments:(1)the coding throughput of low-cost GPU using XOR optimized local repair code is nearly 20 times higher than that of the same CPU;(2)in the case of low-cost GPU and less errors,the accelerated optimization scheme of local repair code based on low-cost GPU proposed in this paper and the RS code implemented by GCRs library are used in coding swallow In terms of throughput,this scheme is slightly better;(3)in the case of low-cost GPU and error tolerance,the accelerated optimization scheme of local repair code based on low-cost GPU proposed in this paper improves the coding performance by 3%-5% compared with LRC coding realized by using GCRs library.