Font Size: a A A

Hypergraph-Based Data Block Placement Strategies For Erasure Code Storage System

Posted on:2023-04-26Degree:MasterType:Thesis
Country:ChinaCandidate:B QiFull Text:PDF
GTID:2568307070984499Subject:Engineering
Abstract/Summary:PDF Full Text Request
Currently,distributed storage systems are developing rapidly.As a redundancy mechanism to guarantee data reliability,the erasure code model can achieve a good compromise between storage overhead,data reliability and data access performance.How to improve the efficiency of data update is a key issue of the erasure code storage system,which has important research significance.The thesis presents a comprehensive analysis of the update process of data blocks in a erasure code storage system,establishes an I/O-oriented data update efficiency evaluation method,constructs a data block access sequence hypergraph model through the correlation between data blocks,and designs a data block placement strategy based on the hypergraph model.The problem of inaccurate super-edge weights during the construction of the hypergraph model is solved by designing a deep learning-based hypergraph model parameter optimisation algorithm.The main contributions of the thesis are as follows.(1)A data block placement strategy based on hypergraph partitioning for erasure coding storage systems.In response to the need to improve the data update efficiency of the erasure code storage system,the thesis designs a performance evaluation method based on the number of update operation I/Os,builds a hypergraph model of the data block access sequence,and obtains the data block placement with correlation by hypergraph k-way partitioning.The experiments verify that the hypergraph-based data block placement strategy can reduce the I/O counts of data updates in the erasure code storage system compared with other placement strategies under different access sequences and different erasure code environments.(2)Optimization algorithm of hypergraph model parameters based on deep learning model.To address the problem of inaccurate weights assigned to hypergraph vertices and hyperedges in the hypergraph modelling algorithm,the thesis proposes a hypergraph model parameter optimisation algorithm based on a deep learning model.The algorithm transforms the logical address of a data block into a distributed representation by means of a word vector method,and uses LSTM to extract the features of the access sequence.The experimental results show that the algorithm can improve the accuracy of vertex and hyperedge weights in the hypergraph model under both modelling approaches based on the difference between LBA and LD,thus optimising the hypergraph model parameters and reducing the number of I/Os during data updates in the erasure code storage system.
Keywords/Search Tags:erasure coding, small-write, distributed storage, hypergraph, deep learning
PDF Full Text Request
Related items