Record:   Prev Next
作者 Greenstein, Benjamin Morrison
書名 Compile-time optimization and runtime flexibility in tiered sensor networks
說明 140 p
附註 Source: Dissertation Abstracts International, Volume: 67-12, Section: B, page: 7164
Adviser: Deborah Estrin
Thesis (Ph.D.)--University of California, Los Angeles, 2006
Distributed sensing applications can be complicated and inflexible. This thesis investigates three concrete ways to make sensor-application programming more accessible
We produce the first flexible system for the collection of high-rate data, such as acoustics, on motes; the key was designing a unified filter architecture allowing easy adaptation to environmental conditions and user interest. VANGo, our software system, provides filters to reduce high-rate data at its source, lessening communication and power overhead. Microservers, with their greater resources, calibrate and configure data reduction software running on motes. Filter adaptation, in our experiments, improves the yield of interesting data by a factor of two over standard techniques and results in a 78% decrease in radio transmissions
Using these insights, we designed a more general task library for all mote applications and implemented it for TENET, an architecture for tiered sensor networks developed in collaboration with Govindan et al. at USC. This library allows for the dynamic composition of applications from simple pieces. As tasks are issued by capable microservers in a tiered network, they remove the responsibility of interpreting complex application logic from mote devices; the mote tier thus becomes flexible and generic, supporting a wide range of application services. The management of tasks is fairly lightweight; mote-resident task descriptions usually require fewer than 200 bytes of RAM. Therefore, our mote runtime can support the concurrent execution of up to 26 of our most complicated tasks
In those cases when very high efficiency is required, the SNACK component library and composition language compiles high-level specifications of applications into statically minimal code. SNACK makes it easy to write and use application-level sensor network services (e.g., routing trees and periodic sensing;) and to express them in ways that improve efficiency by sharing substructure. In our analysis, we present a simple SNACK-based multi-hop data collection application. It uses SNACK language features to provide both simplicity (excluding reusable service definitions, its description is three lines long) and efficiency (it performs comparably to the well-known Surge application)
These contributions separately tackle three important areas of sensor network system design: providing flexible node-local processing, dynamic application composition and control, and efficient execution. In all cases, these efforts focus on a single goal, that of simplifying application development
School code: 0031
Host Item Dissertation Abstracts International 67-12B
主題 Computer Science
Alt Author University of California, Los Angeles
Record:   Prev Next