AI Lab logo
menu MENU

Faculty Candidate Seminar

Tracking and Reducing Energy Usage in Networked Embedded Systems

Prabal DuttaPhD CandidateUniversity of California, Berkeley
SHARE:

Energy is a critical resource in many wireless embedded systems but
profiling energy usage in resource-constrained systems, like sensor
networks, has proven challenging. Answering even basic questions
about energy usage is difficult. For example, how much energy do
individual operations, such as sampling sensors, receiving packets, or
using the CPU cost? What is the energy breakdown of a node, in terms
of logical activity, hardware subsystem, or time? Network-wide, how
much energy do distributed services such as routing, time
synchronization, and localization consume?

In this talk, I will outline three technical challenges to energy
profiling — measuring energy consumption, breaking down aggregate
usage by power domain, and grouping causally connected actions within
nodes and across the network — and present iCount and Quanto, our
solutions to these problems. iCount is new energy meter design that
measures energy usage by counting the cycles of a switching regulator
and offers high-speed, high-resolution, and low-overhead snapshots of
system-wide energy usage. Quanto builds on iCount, and combines
well-defined interfaces for hardware power state visibility,
regression for per-hardware subsystem energy breakdown, and causal
activity tracking using labels, to map how energy and time are spent
on nodes and across a network.

Using iCount and Quanto, we have been able to precisely quantify the
effects of low-level system implementation decisions, such as using
DMA versus direct bus operations, or the potentially dramatic effect
of external radio interference on the power draw of a low duty cycle
radio. These new insights have guided the design and implementation
of a novel link layer primitive, called Backcast, and a clean slate,
receiver-initiated link layer above it. This new link is conceptually
simple, provides asynchronous neighbor discovery, unicast, broadcast,
wakeup, pollcast and other services, and is well-suited to emerging
sensornet applications like low-power mobile sensing, bursty
contending flows, or ultra-low duty cycles.

Prabal Dutta is a Ph.D. candidate in the Computer Science Division at
the University of California, Berkeley, where he is advised by David
Culler and Scott Shenker. His research interests straddle the
hardware/software interface and include embedded systems, networking,
and architecture, with a focus on low-power, wireless embedded
systems. His research has enabled some of the largest sensor networks
deployed to date including the 1,000+ node ExScal project and the 557
node DARPA NEST Final Experiment. His work has been commercialized by
Aginova, Arch Rock, Crossbow, and Moteiv, three of which are venture
funded UC Berkeley spinoffs. His work has also won several awards
including an IPSN Best Paper, an ISLPED Design Contest Winner, and a
COMDEX Best of Show. He is a National Science Foundation Graduate
Research Fellow and a Microsoft Research Graduate Fellow.

Sponsored by

EECS - CSE Division