Tenet: An Architecture for Tiered Embedded NetworksRamesh Govindan Eddie Kohler Deborah Estrin Fang BianKrishna Chintalapudi Om Gnawali Ramakrishna Gummadi Sumit RangwalaThanos StathopoulosAbstractFuture large-scale sensor network deployments will betiered, with the motes providing dense sensing and ahigher tier of 32-bit master nodes with more powerfulradios providing increased overall network capacity. Inthis paper, we describe a functional architecture for wire-less sensor networks that leverages this structure to sim-plify the overall system. Our Tenet architecture has thenice property that the mote-layer software is generic andreusable, and all application functionality resides in mas-ters.1 IntroductionOver the last five years, sensor network research hasseen significant advances in the development of hard-ware devices and platforms, and in the design of servicesand infrastructural elements such as routing, localization,and time synchronization. Deployed systems, however,have lagged behind. Existing deployments are small-to medium-scale continuous data acquisition systems inwhich all sensor data is collected at a central location.This is far from the vision enunciated in early sensor net-works work, in which sensor networks incorporate sig-nificant in-network processing for energy efficiency. Yet,attempts to move these deployments closer to the visionhave foundered.Our experience with sensor deployments has convincedus that the problem is with the vision. The constraints onprogramming low-power, mote-class systems—sensinguncertainty, wireless communication vagaries, and limitedenergy, processing, and memory—are difficult enough tohandle on their own; yet the “Application Specific” prin-ciple [6] on which much sensor research is based suggeststhat each application must tackle these problems com-bined with application-specific data fusion constraints.This is leading us to develop systems that are exceedinglycomplex, unmanageable, and not re-usable. A new archi-tecture is needed.This paper discusses the architectural foundations ofsensor networking. Our focus is on the functional archi-tecture, the principles that, based on cost and complex-ity arguments, state where functionality should reside ina network. Our arguments are modeled after the end-to-end principle [15], which states how functionality shouldbe placed in data communication networks. We call ourprincipleThe Tenet. Multi-node data fusion function-ality and complex application logic should beimplemented only in a tier of relatively high-powered Stargate-class nodes, which we callmasters. The cost and complexity of imple-menting this functionality in motes outweighsthe performance benefits of doing so.The tiered embedded networks built on this principle,which we also call Tenets, contain both small-form-factormotes and Stargate-class masters. Tiered organizationshave been discussed before [20]; our contribution is tosimplify the architecture by explicitly li miting mote func-tionality. Motes contain sensing and actuation functional-ity and enable infrastructure-less instrumentation of phys-ical spaces and artifacts, while masters are free of energyconstraints and provide increased network and compu-tational capacity, enabling large-scale deployments. Allmote sensor data is routed to computational elements run-ning on masters, or users and databases attached to mas-ters. Motes are tasked by applications running on mas-ters, and can implement s imple logical elements such asthresholds and compression, but any further computationtakes place only on masters.Excluding multi-sensor fusion and complex applicationlogic from motes will have two advantages: first, the ap-plication runs in a less resource-constrained environment,reducing development cycles and and improving overallsystem robustness; and second, the principle makes it pos-sible to conceive of a generic mote layer that need beimplemented once, and reused for a variety of applica-tions. The disadvantage—a potential loss of efficiency—is a small price to pay for increased robustness and man-ageability.We emphasize t hat the development of Tenet does notsupplant research on mote-class devices. Motes are es-sential for low-cost dense sensing, and ongoing researchon software architectures for the motes [1] and on various1mote subsystems, such as medium access, time synchro-nization, and localization, will remain highly relevant forTenets.Tenet opens up several novel research directions. Oneimportant area is the development of a generic master-to-mote interface that can be used by several applications.Another is the design of robust subsystems necessary foran operational Tenet: a robust routing system, reliabledelivery of data between masters and motes, effectivecongestion control for high data-rate applications, low-overhead network monitoring, and automated networkingmanagement and tuning. The design of these subsystemscan leverage masters, and the perspective they have of themote network, for simplicity and efficiency.The Tenet architecture can greatly accelerate the devel-opment of applications, and hence the adoption of thishighly-promising technology. This paper discusses theTenet architectural framework, and briefly discusses howit can simplify the design and development of sensor net-work applications.2 The Tenet ArchitectureIn this section, we first briefly review the progress of sen-sor networks research so far, then describe and justify thearchitectural principle guiding the design of Tenets. Wethen describe the Tenet architecture in a little bit moredetail, and discuss the relevant research challenges. Weconclude this section by outlining the implementation oftwo qualitatively different applications on the Tenet archi-tecture.2.1 MotivationRecent progress in sensor network research and develop-ment has been oddly nonuniform. Several groups havemade major advances in hardware development, leadingto two commonly-available classes of sensor platform: in-expensive motes, such as Crossbow’s Mica series and s im-ilar devices from Telos and Dust Inc.; and so-called gate-way nodes, which we call masters, such as Crossbow’sStargate. Masters have roughly an order of magnitudemore computational power, memory, and wireless com-munication bandwidth than motes. Moreover, the com-munity’s research output has been impressive and wide-ranging, from lower-layer services such as MAC, routing,localization and time synchronization [5, 14] to higher-layer