Research On Key Techniques Of Applying MEMS-based Storage To Computer Systems

Magnetic disks have dominated secondary storage for decades. The capacity and access speed of magnetic disks have reached a new level, and it has becoming more and more difficult to improve. The huge disparity between memory access times and disk access time has created a performance bottleneck in computer systems. Many new storage media and new techniques have always been coming out. MEMS-based storage is an interesting new technology that promises to bring fast, non-volatile, mass data storage to computer systems. MEMS-based storage devices MEMS-based storage devices have advantages in physical size, storage density, and energy conservation. It has been a new challenge to traditional secondary storage systems.In this paper, we study MEMS-based storage and the key techniques of applying MEMS-based storage to computer systems. We study the architecture and storage principles of MEMS-based storage, and describe three generations of MEMS-based storage models of CMU in details, including physical architecture, accessing feature, logical data placement, and so on. Then, we build the performance model of MEMS-based storage based on the CMU model.RAID is still the main way to improve bandwidth and reliability. In our work, we study the key techniques of using MEMS-based storage in disk arrays. What we have done are as follows:(1)We use MEMS-based storages to build disk arrays instead of disks. Through the parallelism within a single storage device and the parallelism among many devices, we can obtain high access performances. Also, we can obtain high reliability through data redundancies and spare spaces. The simulated results show that disk arrays built with MEMS-based storage have higher performance than traditional disk arrays.(2)We integrate MEMS-based storage into disks, acting as the cache of disks. Also, we integrate it into disk arrays, replacing the NVRAM cache, and acting as the disk array cache in the controller. We study the main techniques to implement MEMS Caching Disk, and do performance simulation and comparison.(3)Because of the high cost of MEMS-based storage, we study three kinds of hybrid disk arrays, built with MEMS-based storages and disks—MEMSMirror, LogDisk, and DualStripe. They can take advantage of the fast access feature of MEMS-based storage and the advantages of the sequential access feature of disks. They can not only improve the performance, but also decrease the costs. We can get better performance/cost ratios.(4) We study the reliability of MEMS-based storage enclosures, which organized as RAID5. We study deeply into different failure device replacement schemes, and different spare space strategies. The study shows that MEMS-based storage enclosures are more reliability than disks.We study four aspects of operating system management of MEMS-based storages: request scheduling, data placement, failure management, and power conservation. We study a new bipartite layout scheme, which is more suitable for the physical position feature of MEMS-based storages. The two dimensional data layout feature of MEMS-based storage makes it more suitable in relational database. In this paper, we study the application of MEMS-based storage in database, introducing a new data placement scheme for MEMS-based storage-TDL. The Analysis shows that TDL can improve query speed, memory and cache utilization.What we have done in our work has important effects in applying MEMS-based storage to computer systems. Some of the key techniques and design ideas are also helpful for applying other new storages into computer systems.