Design And Implementation Of A Highly Concurrent Client For Blueocean Mass Storage System

With the rapid growth of information, how to efficiently store andprocess large amounts of data has become a huge challenge faced by today’scomputer science. The traditional file system has reached its bottleneck inthe improvements of capacity and performance. Distributed mass storagesystem is becoming an inevitable trend of the future storage. In such asystem, a good design and implementation of client has an important impacton the overall performance of the storage system.BlueOcean is a large-scale distributed storage system based onobject-oriented storage technology, with the advantages of high performance,large capacity, high reliability and easy management. The system consists ofa single metadata server and several data storage servers. It provides POSIXinterfaces for the application software via clients, and can support mostexisting applications.In this thesis, we made an in-depth research on the critical technologiesof the implementation of the mass storage system. On this basis, wedesigned and implemented a highly concurrent client for BlueOcean system.The main work in this thesis includes:(1)Describe the evolution of the storage system architecture. Throughthe study of several typical mass storage systems, analyze the criticaltechnology on the design of the mass storage system.(2)Analyze the weaknesses of clients in the current mass storage systemsand implement a highly concurrent client using the event-driven model. (3)Describe the detailed realization of the modules in the highlyconcurrent client.(4)According to characteristic of the high concurrent client, improve thewrite performance through three ways: optimize the FUSE framework,reduce data copy and parallel the write processes.(5)Design performance tests of the highly concurrent client.The results of the experiments show that when the BlueOcean storagesystem is outputted by the highly concurrent client, it has good writeperformance, and with the increase in the number of concurrent access files,the data read/write performance, metadata operations performance andtransactions performance of the overall system has a significantimprovement. The research of this thesis could be a valuable reference forthe design and implementation of highly concurrent client in other massstorage system.