GlusterFS Data Distribution Policy And Performance Optimization Research

With the rapid development of information technology, data usage is growinggeometrically, the capacity of the storage system, scalability, performance requirementscontinue to increase. Cloud storage generated associate with cloud computing, whichshields the differences in the way to access and management style and provide a unifiedstorage access service, storage scheduling mechanism adaptively adjusted according tothe application characteristics requirements, in order to achieve unified management oflarge-scale multi-level storage, efficient application and provide high-performancestorage and access platform to users. Distributed file system as the underlying enablingtechnology of cloud storage systems, for the cloud storage system scalability,transparency, fault-tolerant, management flexibility and other features provide thefoundation.From the technical requirements of cloud storage on the distributed file system, weanalyze and compare several typical distributed file system that support cloud storage,introduce the current status of research on the performance of distributed file system.We in-depth study typical for cloud storage distributed file system GlusterFSarchitecture, data distribution strategy, read and write operations processes, performancecharacteristics and optimization.Firstly, we analysis for GlusterFS modular stackable architecture, research itslinear scalability, flexible volume management, high reliability technical characteristics.GlusterFS three basic volume management mode(Distributed Hash, Automatic FileReplication, Stripe) data distribution strategy and implementation mechanism werestudied, focusing on the elastic hash algorithm mechanism to achieve in-depth analysis.Secondly, on the basis of theoretical analysis, set up an experimental environment,test GlusterFS features and performance. Including the system linear scalability, theread and write performance under the three basic volume management mode, in thedefault distributed hash volume mode large files and small files storage performance.Then analysis and compare the test results.Finally, poor storage performance for small files in GlusterFS were analyzed andimproved, we put forward to a small file prioritize consolidated block writing strategyand algorithm, experimental results demonstrate the effectiveness of the strategy.