NeTS-NOSS: STONES: Storage-Centric Networked Embedded Systems

提案编号：0626873PI: Deepak GanesanInstitution: University of Massachusetts AmherstTitle: NeTS-NOSS: STONES：以存储为中心的网络嵌入式系统摘要大量的传感器网络研究已经解决了能源效率的问题，主要是通过利用计算比无线电通信便宜许多数量级的事实。存储在传感器网络中的作用经常被忽视，因为闪存被认为是低效的能量方面。新的NAND闪存技术的出现极大地改变了这种权衡，测量表明，存储成本现在比通信成本便宜两个数量级，与计算成本相当。这一发展要求对传感器网络设计进行重大反思，并阐明了传感器节点将配备高容量本地闪存的愿景。STONES项目提出了基础的跨学科研究，旨在通过使存储在这些系统中的中心作用来重新设计传感器软件和平台。STONES提供了一个丰富的数据管理堆栈，它采用了类似于网络协议堆栈的分层架构。该堆栈包括三层：(i)对象存储层，支持针对能量和内存受限系统优化的灵活对象类型；（ii）本地数据库层，支持使用本地归档和索引数据以及多分辨率存储的功率受限查询处理；（iii）分布式数据管理层，负责确保节点故障时数据可用性。STONES系统将使用相机传感器网络和参考实现进行评估，原型的源代码将通过开源软件许可提供给其他研究人员。

Proposal Number: 0626873PI: Deepak GanesanInstitution: University of Massachusetts AmherstTitle: NeTS-NOSS: STONES: Storage-centric Networked Embedded SystemsAbstractA significant amount of sensor network research has addressed the problem of energy-efficiency, primarily by exploiting the fact that computation is many orders of magnitude less expensive than radio communication. The role of storage in sensor networks has often been overlooked since flash memory has been assumed to be inefficient energy-wise. The emergence of new NAND flash technologies significantly alters this tradeoff, and measurements show that the cost of storage is now two orders of magnitude cheaper than communication and comparable to that of computation. This development calls for a significant rethinking in sensor network design and articulates a vision where sensor nodes will be equipped with high-capacity local flash storage. The STONES project proposes fundamental cross-disciplinary research aimed at redesigning sensor software and platforms by making the role of storage central in these systems. STONES provides a rich data management stack that employs a layered architecture similar to a network protocol stack. This stack comprises three layers: (i) an object storage layer that supports flexible object types optimized for energy and memory-constrained systems, (ii) a local database layer that supports power-constrained query processing using locally archived and indexed data as well as multi-resolution storage, and (iii) a distributed data management layer responsible for ensuring data availability despite node failures. The STONES system will be evaluated using a camera sensor network and reference implementations and source code for prototypes will be made available to other researchers via an open-source software license.