| Downsizing Sensor Software
David
Pescovitz photo
Professor
David Culler demonstrates how a camera mounted above a grid
of sensor nodes could track the motion of an object in a
room. (Click for larger image.)
|
Hidden in the
lush habitat of nesting seabirds are dozens of tiny electronic sensors
measuring the humidity and temperature to gain insight into these
coastal creatures' environment. Meanwhile, in the Intel Research
Laboratory in downtown Berkeley, the same sensors are embedded in
office chairs to track their movements as they are pushed together
for meetings and then apart as the attendees return to their individual
workspaces. While seemingly disparate locales, both the seabirds'
natural habitat and a bustling research facility are ideal testbeds
for the distributed, dynamic, and adaptive software brains behind
tiny wireless sensors.
The sensors — such as Berkeley's 10 cubic mm Smart Dust
motes — are equipped with microscale devices that measure
light, temperature, motion, or other conditions and are cheap
enough to deploy en masse. For example, in December Berkeley civil
engineers instrumented an experimental building with 50 sensors
to determine its structural integrity after a simulated earthquake.
The main problem encountered when designing sensor networks,
however, is scarcity — from processing power to storage to
energy. Quite simply, Windows XP won't fly. But TinyOS will.
Built from the bottom up by a team led by David Culler, a computer
science professor and director of the Intel Research Laboratory
at Berkeley, the tiny operating system (TinyOS) and its related
networking infrastructure support the myriad applications promised
by Smart Dust and other wireless sensors-on-a-chip.
For one, TinyOS enables the sensors to form their own networks,
bouncing bits of data from neighboring node to neighboring node
until the information reaches its desired destination for processing.
At last year's Intel Developers Forum, Culler and his team demonstrated
a self-organizing wireless sensor network of 800 nodes distributed
in an auditorium. TinyOS is now at the heart of a collaboration
between Intel and the University. While ad-hoc networking is essential
for sensorwebs, the researchers' latest coup gets to the core
of what makes these networks so novel: they're dynamic by definition.
And the software that drives them must account for that.
David
Pescovitz photo
Outfitted
with a sensor node, this mobile robot can build a digital
map of the terrain it traverses. (Click for larger
image.)
|
Take the case of the seabirds, Culler says. Perhaps a scientist,
while looking at the environmental big picture, notices a discrepancy
in one nest. Ideally, if the sensor nodes in and around the nest
could quickly shift from monitoring climate to tracking motion
and sound.
"Oftentimes, scientists don't know what the most useful application
is until the network is deployed," Culler says. "You can't just
put all possible software you might need into each sensor node.
How do you go about re-tasking them?"
It's not unlike installing software in a small network of desktop
computers when a new task demands additional functionality. The
big difference though, Culler explains, is that instead of a handful
of PCs in an office, he's faced with thousands of sensor nodes
that may be hanging in trees or mounted behind drywall in a building.
"Ironically, these little devices need to be more manageable
than your PC," he says. "Our expectations are high and you can't
go around and power cycle them on and off if they're not working.
They need to be robust in a noisy world."
Fortunately, the applications his group develops for the TinyOS
are on the order of 24 bytes long. That's even shorter than this
sentence. So it doesn't take long to teach each node something
new.
"Much of the structure might be common to a whole class of applications
resident on the device," Culler says. "The rest is then dynamically
controlled by capsules that you toss into each device."
This
image shows the topology of a self-organizing network of
sensor nodes at a recent demonstration in San Francisco's
Moscone Center. (Click for larger
image.)
|
Still, it's absurdly inefficient to program each node individually.
Instead, Culler introduces the equivalent of a computer virus
into the network. As the nodes communicate, they infect their
kin with the new operating instructions. The programs then run
inside an easily-accessible and manageable "virtual machine,"
software inside the node's main operating system that simulates
a separate computer.
While the Intel/Berkeley team tackles system architecture, networking,
and programming issues, it's also developing efficient ways to
ask the self-assembled networks questions about the data collected.
"In this world you're not interested in getting a file from
a single device, but rather information from a collection of devices,"
Culler says. "It's much more like a database query."
For example, he explains, the plan is to continue instrumenting
the Intel laboratory along with facilities on the main Berkeley
campus. The goal is to weave the two locations together using
the wireless nodes.
"Part of the virtue of an open lab environment is that you know
who is around and can get together as needed," Culler says. "If
you are in one of the open labs and so am I, the network can help
connect the spaces and bring us together. But when we are not
in those well-defined locations, the technology does not step
into our lives."
Indeed, much like at the Berkeley Wireless Research Center across
the street, real-world applications — explored through collaborations
with the likes of Berkeley's Center for the Built Environment
and the Center for Information Technology Research in the Interest
of Society are the fuel that drives the technological development
at the Intel Research Laboratory.
"Low-power wireless networking has gained a lot of credibility
because there's industrial interest in the idea," Culler says.
"At the same time, nobody really knows what the market is. And
it wouldn't really exist if there wasn't an academic force saying,
'We know there's this interesting thing out there and we're going
to explore it.'"
Intel
Research Laboratory at Berkeley
David Culler's
home page
CITRIS
Center for the Built
Environment
Smart
Dust
Lab Notes is published online by the Public Affairs Office of the UC Berkeley College of Engineering. The Lab Notes mission is to illuminate groundbreaking
research underway today at the College of Engineering that will dramatically change our lives tomorrow.
Editor, Director of Public Affairs: Teresa Moore
Writer, Researcher: David Pescovitz
Designer: Robyn Altman
Subscribe or send comments to the Engineering Public Affairs Office: lab-notes@coe.berkeley.edu.
© 2002 UC Regents. Updated
4/1/02.
|
