Mass Storage System Based On Network Attached Disk Array

Recent developments in computer systems and high-speed networks have stimulated the emergence of mass information applications. It's a difficult problem to store and manage the mass information effectively, which propelled research on the mass storage system. Network storage technology is a good method to solve the problems of mass storage system. Different connection ways between storage devices and network compose different topology and result in different impacts on the system performance. It's a significant research project to choose a suitable topology, construct a mass storage system with big capacity, high bandwidth, scalability and manageability.Centralized storage systems can reduce management spending and improve management quality because of the single control point, but system performance and scalability is limited. Distributed storage system can improve system performance and scalability, but the manageability and Usability is poor. Based on DAS (direct attached storage) architecture, centralized file Management and Distributed data Storage Mass Storage System (CMDS-MSS) connects multiple storage nodes of DAS to network respectively and strikes a good balance between a centralized file management and a distributed data storage. Storage nodes of CMDS-MSS exchange data with network clients directly through network channels, the architecture separates control and data flow to avoid file server bottleneck. CMDS-MSS achieves capacity scalability by expanding the system storage capacity incrementally with additional Net-RAIDs along with associated network interfaces that expand data transmission rate proportionally, parallel virtual transmission mechanism is used to implement parallel transmission of multiple network channels. Because of the scalability, storage management of mass storage system will be more and more complicated with increase of heterogeneous storage nodes. Virtual storage technology consolidates various heterogeneous storage resources into one transparent single logic view. Clients can allocate the logic view with their owner requirements. According to the realization position in storage system architecture, virtual storage technology is divided into three categories, based on host, based on storage devices and based on storage network. Host based scheme constructs the virtual abstract layer on file server, the virtual abstract layer can be implemented on application layer and device layer. CMDS-MSS implements the single logic view between file system and device driver of file server. The method retains compatibility with existed applications and architectures, and also takes advantage of the block buffer of file system to enhance system performance. Because the single logic view replaces physical devices by virtual abstract layer of storage space, CMDS-MSS supports storage subsystems with heterogeneous interfaces.The parallelism of storage system is usually developed from two sides, namly space and This research work was supported by the National Nature Science Foundation of China under grant 60273074 and 60303032time, pipeline technology adopts time parallelism. Net-RAID adopts virtual SCSI commands encapsulated by operation type (read/write), start sector, sector number and other information to execute I/O operation. There are multiple virtual SCSI commands in Net-RAID (Network Redundant Array Independent Disks) command queue within system resources. Pipeline producer/consumer policy divides the I/O cycle of virtual SCSI command into different phases and uses buffer technology to smooth work speed of different function components, the policy overlaps disk I/O and CPU computation to improve system performance.RAID organizes multiple disk drives into one logic disk. Requested data of multiple network clients is distributed on the logic disk in disorder, decomposed data on disk drives is more random. Sequential data on logic disk is decomposed to every disk drive with sequence too. Queue the multiple I/O requests of network clients and make them distribute on the logic d...