Research On Access Optimization Technologies Of Distributed Storage Based On Erasure Code

Erasure code has become a basic problem of researching on distributed storage system,also to be the critical issue for building distributed storage system.It is essential for improving system reliability,saving storage space and improving user experience.This paper concerns about key technologies of I/O based on Erasure Code in distributed storage system and make an in-depth study of data insertion and data reconstruction,obtain the contributions as follows:(1)Erasure code's transmission path optimization oriented data insertionTo solve the data insetion problem of locally repairable codes,existed methods are lack of the consideration about control of transmission structure and high randomicity,resulting in costing too much time and poor performance,so we propose a kind of transmission structure optimization algorithm based on tree and a transmission order of multitask scheduling algorithm based on outdegree.Experimental results show that,transmission structure optimization algorithm and transmission order of multitask scheduling algorithm improve performance significantly over traditional approaches.(2)Fine-grained parallel decode oriented data reconstructionTo solve the data reconstruction problem of locally repairable codes,existed methods usually only focus on serial decode,and are lack of the consideration about,parallel repair features when multiple original data blocks get lost and tradeoff of degree of parallelism and data downloads,resulting in serial decode and too much data downloads,so we propose a kind of parallel decode algorithm oriented maximum parallelism degree and a parallel decode algorithm oriented minimum data downloads.Experimental results show that,parallel decode algorithm oriented maximum parallelism degree and parallel decode algorithm oriented minimum data downloads can increase the decoding efficiency.Parallel decode algorithm oriented minimum data downloads gets better performance than parallel decode algorithm oriented maximum parallelism degree.