MEMS-Based Storage Devices: Integration in Energy-Constrained Mobile Systems

The digital era in which we are living today requires our increasing awareness of energy efficiency to reduce the negative effects on our lovely environment. We, people, are increasingly dealing with digital contents to facilitate our deals, which increases the demand for larger storage capacities than ever before. The environmental considerations and the data explosion worldwide are calling for green and ultrahigh-density storage technologies. A storage technology, based on MEMS, promises to deliver green and high-capacity storage systems. Storage densities up to 4 Tb $in^2$ have already been demonstrated. With such a technology, a storage device with a capacity of 1 Tb can be mounted in a package smaller than a thumbnail, dissipating one Watt of power. Disruptive technologies take, however, a significant amount of time to materialize into commercial products. One key reason for this delay is the difficulty of integrating and adopting new technologies. Integration, in a broad sense, involves the investigation of roles a new technology can play, and solutions to its impediments. Early solutions to the integration problem help to reduce the time to market, and most likely contribute to the success of the technology. Flash memory, for instance, was invented in the eighties. The large demand for Flash in mobile systems has drawn the attention of researchers to investigate Flash. Just recently, researchers started looking into ways to construct storage systems based on Flash that are reliable, have low latency, and consume little energy. Flash memory has not found its way to enterprise storage yet, whereas their kin, hard disk drives, are sitting there wasting a significant amount of energy. That kind of late response to Flash is costing data centers millions of dollars every day in energy and cooling cost. We would like to avoid such a late response for MEMS-based storage devices by being proactive in how we can get this family of devices successfully integrated as early as possible. Like any other technology, MEMS-based storage demands optimization, and has challenges that need to be tackled. In this work, we optimize MEMS-based storage, tailor it to mobile battery-powered systems, and compare it to Flash memory and Hybrid (Disk--Flash) storage. The research of this dissertation looks mainly into the energy and cost aspects of MEMS-based storage with the following two contributions. We devise policies to reduce the energy consumption of MEMS-based storage devices. We also propose to exploit knowledge of the expected workload in configuring the data layout of a MEMS-based storage device in order to increase the effective capacity. Both contributions target at satisfying the increasing demand for green and inexpensive storage devices. In addition to the energy and cost aspects, we make sure that the response time and the lifetime of MEMS-based storage devices are competitive. The data-layout and energy-saving policies account for the timing performance by looking at configurations that do not compromise on the response time. With respect to lifetime, we devise probe wear-leveling policies that increase the lifetime of MEMS-based storage devices with minimal influence on the energy consumption and the response time. The dissertation incorporates the conclusions from the study of the policies, and investigates the employment of MEMS-based storage devices in important types of mobile application. For predominately streaming applications, we also investigate the influence of buffering on the energy consumption, response time, and capacity of MEMS-based storage devices. We put the technology into perspective by comparing it to Flash memory and Hybrid (Disk--Flash) storage. Our system-level research in this dissertation identifies potential points of enhancement of MEMS-based storage devices. Enhancements are targeted at reducing the energy consumption, decreasing the response time, cutting down the per-bit cost, and increasing the lifetime of the device. Most importantly, we show that the per-bit cost of MEMS-based storage is crucial to its success. Our system-level contributes to reduce the cost, while reduction on the device level is still needed. We provide methods and means to configure MEMS-based storage devices to prepare them to serve in different environments as a viable storage technology. Following our research findings, designers can craft storage systems based on MEMS-based storage that are reliable, energy and performance efficient, and cost effective