Key Technology Research For Hybrid Storage Systems Based On Multi-media Device

Flash and SSDs have gotten much attention in recent years by the academia and industry. Flash memory is a solid-state electrical drive storage media with high per-formance, low power, compact and lightweight characteristics. It has been widely used in embedded systems and mobile devices. With the continuous improvement of flash memory storage density, flash-based SSD (Solid State Drive) is increasing pop-ularity. As electronic equipment, SSD breaks the limitations of traditional mechanical disks, and it is high-speed reading, low noise, low power consumption, etc. Its per-formance can narrow the performance gap between the storage devices and computing devices and help solve the performance bottleneck of a computer system. Therefore, SSDs become an important secondary storage devices, being widely welcomed indi-vidual PC and enterprise storage applications.However, SSDs also inherited the shortcomings of flash memory, including read and write asymmetry, erase wear, etc. The software systems of this stage are designed for disk, which can not take full advantage of the performance benefits of SSDs. Be-sides, the high storage prices further hindered the popularity of SSDs. All these factors have led to that SSDs can not completely replace the disk. The storage system having coexistence of SSD and disk is the development trend in the future. Therefore, hybrid storage technology research attracted academic attention.The hybrid storage technology with SSD and disk is new research topic. Under this background, this paper summarizes the research results about hybrid storage and flash indexing technology. We analysis the key issues in the research field, and carry out the research work on hybrid storage and flash indexing.Hybrid storage technology research is mainly for3kinds of storage model:a. main memory-disk-SSDs, b. main memory-solid state drive-disk, c. main memory-solid state drives&disks. A class storage model is built for low-end SSD by using disk as the write cache of SSD, to optimize the random writing of SSD. B calsee storage model is built for high-end SSD by using SSD as the cache of disk, to to improve the overall performance of the storage system. C class storage model takes SSD and disk on the same layer to manage. With the3kinds of storage model, this paper launches research work on the key technology of hybrid storage, such as data partitioning, migration, multi-level cache, hybrid indexing, etc. We propose a series hybrid storage algorithms to resolve these problems.In order to promote the high-performance of SSD in hybrid storage, this paper also takes research work on the flash related B+tree index. Firstly, this paper summs up the present work about flash B+tree. Then we analyzes the read/write feature of B+tree and points out traditional B+tree is a structure with much read/write ran-domness. Then this paper improves the B+tree on hybrid storage and flash device. In the hybrid storage, we present a new B+tree adapting the feature of SSD and disk. By improving the structure of B+tree leaf, we reduce the write operations and en-hance the the overall performance. On the flash device, with the policy "trade read for write", we use Bloom Filter to improve the mechanism of overflow node and successfully resolve the problem of high costs in reading.This paper also discusses the problem that the research lack flash and hybrid storage hardware to support which is summarized as hardware constraints. We try to resolve the problem from hardware implementation and software emulation. In terms of hardware implementation, this paper presents an experimental platform combining software and hardware to support research work on SSDs technology. In the software simulation, this paper proposes a simulation platform for hybrid storage based on DiskSim and achieves accurate hardware emulation.The main contributions of the paper include the following aspects.(1) Based on the architecture with SSD and disk at the same layer, we proposed a time-sensitive hybrid storage model. The model calculates the read/write tendency of data by the pages’ access information, and makes the impact of statistical information decay over time by the pages’access heat. Finally we get the optimal ratio of SSD and disk with the cost analysis.(2) Base on the architecture with disk being write cache for SSD, we propose a hybrid storage model with asymmetric processing mechanisms. This model proposes an asymmetric trigger mechanism and an asymmetric migration migration to achieve the optimization of the SSD.(3) Based on the architecture with SSD being cache for disk, we propose a new leaf structure for B+tree. By combining the characteristics of SSD and disk, the new B+tree not only reduces the overall write cost, but also weaks the B+tree relying on the buffer with the premise of guaranteeing the performance. This is the improvement of traditional algorithms on hybrid storage system. (4) On a single flash device, we propose a new B+tree based on the overflow node mechanism and Bloom Filter. The new B+tree uses the overflow node mecha-nism to reduce the write cost and uses Bloom Filter to compensate the shortcomings of overflow node to achieve controlling read costs. By this method, this paper not on-ly optimizes the read and write cost, but also builds the entire structure of B+tree on the basis of mathematical analysis. So that B+tree optimization problem is changed to a mathematical analysis problem with nothing of the device hardware.(5) For the hardware constraint problem of flash related research, this paper pro-vides solutions from both hardware design and simulation software to enable re-searchers to easily expand the research work on multiple flash devices. So this work lays the foundation for the development of research in this field...