The Research Of Distributed Software Defined Storage Area Network Basing On ISCSI

Nowadays, with the explosive growth of data quantity and data becomes moreand more important, the requirements for fast and accurate data storage become moreand more strict. Traditional、centralized storage structure can’t satisfy the requirementfor massive data. Thus, distributed network storage structure is deployed widely.Among of it, iSCSI is widely deployed by many companies due to the fact that it isbased on the mature Ethernet structure. iSCSI has such advantages as easily set-up,strong scalability and long distance, etc.But in practical use, iSCSI needs expensive and specific hardware, such as HBAcard to satisfy the requirement for bandwidth. The high budget has become the barrierof its development. Besides, distributed storage structure needs a lot of schedulingwork because data is stored on different storage points. So, storage switches fromhardware-drived to software-drived. Nowadays, software is recognized as the trend ofstorage technology. The term software defined storage (SDS) means that abstractinghardware drive from storage technology but using software drive. SDS associates allthe control work of storage with software. This software is not a firmware in thestorage hardware, but is a part of operating system or hypervisor level. SDS canprovide such features as storage virtualization, dynamic management of storageresources.This paper is based on the development of common compute、storage and network technology. We try to design a distributed software defined storage areanetwork based on iSCSI (iSCSI-DSDSAN). We want to optimize position of data anddecrease the migration cost. Besides, we want to improve the access performance ofSAN and substitute the hardware drive structure. This paper is mainly researched onthe following part: iSCSI protocol, iSCSI parameters and iSCSI-DSDSAN. Wedesigned a series of experiments of how different iSCSI parameters influence theperformance of it. Through these experiments, we find the best parameters suitable foriSCSI. We also designed the structure of iSCSI-DSDSAN and its component. Wepropose a distributed placement model based on genetic algorithms. This model isused when place data to different storage entities. Through Matlab simulation, wecompare it with Random model and Heuristic model to verify the efficiency ofstorage scheduling of the proposed model. In the meanwhile, we propose the processof optimized iSCSI read and write operation. This process is used when read andwrite data. Under Ubuntu operating system, we realize it using java programming andshell scripts. The experiments results indicate that the distributed placement modelreduced the access delay and migration cost by15%on the whole. In the meanwhile,IO performance of iSCSI-DSDSAN optimized largely especially for write bandwidthwith36%increase.