| Micro-Electro-Mechanical Systems (MEMS)-based storage is an emerging non-volatile secondary storage technology. Combining a movable non-rotating media sled and a stationary array with thousands of read/write probe tips into a single chip as small as a dime, MEMS-based storage provides up to ten gigabytes of storage with low access latency, high streaming bandwidth, low power consumption, and low entry cost. It is equally important that MEMS-based storage exhibits distinct low-level device-specific features, such as two-dimensional independent media positioning and fast power state transitions. The theme of this thesis research is to exploit the distinct properties of MEMS-based storage to deliver better performance, higher cost/performance ratios, lower power consumption, and higher reliability of storage systems.;We developed an analytical model of seek times for MEMS-based storage and identified seek time equivalence regions. This finding has great impacts on request scheduling and data layout on MEMS storage devices. We have developed MEMS-aware request scheduling algorithms that achieve superior service performance without sacrificing fairness. Data layout that leverages the knowledge of MEMS seek time equivalence regions also results in better performance under randomly-accessed workloads. Our MEMS-aware power conservation techniques reduces half of the power consumed by storage devices with negligible I/O performance penalties. With internal redundancy, sufficient online spares as well as optional maintenance, MEMS storage drives, which are built as disk replacement, are highly reliable in the economic lifetime. MEMS-based storage can also be used as fast front-end for disks to build high-performing, cost-effective hybrid MEMS/disk storage devices. |
