HYPHOS... a wireless, self-organizing network
In the near future, many everyday objects will be connected to
digital networks, giving rise to Things That Think. In becoming
integrated into larger networks, common objects can acquire
personalized behavior and memory.
But how can you can you make a thousand network connections in one
room? How can you interconnect these objects without a tangle of
wires, burden of battery packs, or Ph.D. in network administration?
Hyphos (from the Greek word for web), is a wireless, self-organizing
digital network. Each node in the network communicates only with its
immediate neighbors. Neighbors relay messages to their neighbors in
turn until the message reaches its destination.
Each node in a Hyphos Network:
* is autonomous, acting as its own router
* consumes very little power, designed to run on a watch
battery or parasitic power
* transmits over a localized area, conserving bandwidth in the airwaves
* is homogenous; there's no distinction between "mobile nodes"
and "land stations"
* is mobile; can move freely in the environment
* is low cost; can be constructed on a single silicon chip
The overall Hyphos Network:
* is self-organizing, dynamically sets up routing
* is self-configuring, requiring no pre-existing wiring or infrastructure
* is self-healing, adapting automatically as nodes move or vanish
Robert D. Poor
MIT Media Laboratory
Last Edit: Aug 1996
Research in the Things That Think consortium at the MIT Media Lab
focuses on the boundaries between atoms and bits. The work of the
Personal Information Architecture group, of which I am a member,
examines how everyday objects will find expression on digital
networks. As an example, there are now Coke machines with 2 way
pagers that can announce their inventory to the soda delivery van.
This saves the delivery guy a huge amount of time.
As part of this investigation, we're developing "Hyphos," a
self-organizing wireless network. Hyphos has been created with the
premise that high-density, low-cost networking will become an
important market in the next few years.
The technologies that are most important to us are those that touch
our lives on a daily basis. In thinking about the future of computer
networks, its instructive to consider another major network: the
Telephones have made a huge difference in our lives, but the change
didn't happen all at once. The advent of the first telephone in your
home town might have made front page news in the local newspaper, but
didn't significantly change the way you conducted business or lived
your life. But now that the telephones are essentially ubiquitous,
they've become part of the way we work and live.
The point is that it's the density, not the number of connections,
that makes a real difference. The same is true of computer networks.
Now I'm going to spout some numbers. I can't prove that the numbers
are absolutely right, but you can't prove that they're wrong, so
let's call it a draw.
In 1984, there are about ten thousand network connections in the
United States. By a "network conntion", I mean a place where you
could walk up to a terminal and type "telnet stanford" (or "telnet
mit") and establish a connection to a remote computer.
Note that there are over 120K towns in the United States. This means
that on average, only one of out twelve towns would have a network
connection -- you'd have to travel a long way to get online.
But check out that flashing square...
A mere 6 years later, 1990, we don't have to travel so far to arrive
at a live network connection: there are about ten thousand network
connctions in the greater Boston area.
Zoom ahead another 6 years, the year is 1996. There are, in fact,
well over ten thosand network connections on the MIT campus alone.
We're now at a point of density similar to telephones. The current
model is that there's one person, one computer, one network
But what happens when we zoom ahead six more years?
It's reasonable to expect that there are ten thousand network
connections in the MIT Media Lab alone. But this does NOT mean that
there will be ten thosand workstations or PCs in the building. The
model of one person / one computer / one network connection no longer
Instead, we'll see an entirely new category of "everyday objects"
connected to networks. Computers, certainly, but clocks, thermostats,
appliances, even lowly light switches will be networked. And the
density of connections will increased by several orders of magnitude.
What's required to make a network that will support this kind of density?
Where there is currently a single network connection, there will be
hundreds or thousands. And these most of these connections will
terminate with everyday objects, not high-priced computers.
Imagine that we have a network to interconnect "everyday objects".
The primary characteristics of this network are that each node is
really dirt cheap -- less than ten dollars -- and that there may be
hundreds or thousands of these nodes in the space of a single room.
My recent research has focused on designing a new kind of network,
which I've dubbed "Hyphos", which means "web" in Greek. From the
start, the design of Hyphos has been driven by the dual contraints of
high-density and low cost. The result is a network architecture
suitable for interconnecting everyday objects.
So what are the technical requirements of a Hyphos network?
First off, a Hyphos node will be wireless. You can't go stringing
hundreds of copper or glass filaments between all the nodes.
Secondly, a Hyphos node will be low power, since you can't assume
that there will be a power source handy. If the node runs on
batteries, it should last for years, not hours. Ideally, the node
will consume so little power that it can be run off parasitic power
sources, such as ambient EMI.
Third, the overall Hyphos network must be self-organizing. Nodes are
associated with everyday objects, and some everyday objects tend to
move around. You can't have some poor fool editing IP routing tables
every time an object is carried into the room. You also want a
network that's robust even when individual nodes fail or are removed
from the environment.
Since the network nodes are to be wireless, low-power, and
self-organizing, some corollaries follow.
Corollary one is that an architecture that uses base stations won't
work. In a system that uses base stations, each network node must
have sufficient transmitter power to reach all the way to the base
station, which violates the low-power requirements.
Corollary two is that the responsibility for routing messages and
maintaining the network must be fully distributed. Since each node
has limited transmit range, there's no one node that can "hear" all
the other nodes, therefore, no centralized control is possible.
Corollary three is that in order to meet cost and power constraints,
each node has limited storage and computational resources. In
particular, this means that nodes cannot maintain routing tables or
routing cost graphs of the entire network -- that would take too much
storage and require too many compute cycles to maintain.
The power of radio signals tends to fall off with the square of
distance. A popular wireless LAN system advertises a transmit range
of 200 meters. In contrast, a Hyphos node has a transmit range of
about 6 meters. This requires one thousand times less power than its
its LAN counterpart. Instead of milliwatts, a Hyphos node requires
If you think of limited transmit range as a limitation, consider
this: airwaves are a limited commodity -- they're not making any more
of them. When the aforementioned wireless LAN transmitter speaks, it
blankets over 125 thousand square meters. This is not a viable
approach for high-density networks.
Another way to think of this is in terms of fiber optic cable. One of
the key advantages to fiber optics is that you can run a lot of data
through a cable with a very small cross section. It wouldn't do much
good if a fiber optic cable was a meter in diameter.
So by localizing transmit range, you save power and conserve bandwidth.
In a dense network, bandwidth is limited if one transmitter dominates
the available airspace. Therefore, a better figure of merit is how
tightly localized the transmitters are, not how far each transmitter
A typical wireless network is comprised of wireless nodes and wired
base stations. As I mentioned earlier, this forces all nodes to have
sufficient transmit power to reach the base station. But a base
station architecture has other problems as well.
It's not always practical to install wired base stations, due to
environmental or even political considerations. A steel mill is one
example of a hostile environment. A jungle controlled by a
totalitarian government is another.
Very dense networks
It's hard to manage high density effectively with a base station
architecture. If there are many nodes clustered around a base
station, the base station becomes a bottleneck and network throughput
Very large networks
Cell-splitting is the technique of installing more base stations and
reducing the transmit power of all the nodes. This works to a point,
but there's a central switch that manages communication between nodes
and base stations. As the density grows and more base stations are
added, the central switch becomes a bottleneck.
So what it comes down to is that to create a high-density network,
the routing of packets and the maintenance of the network must be
fully distributed with no distinction between a "router" and a
One reward of this architecture is that the creation of the initial
infrastructure is almost free. All you need to create a network is to
scatter enough nodes around so that they can hear each other. If you
don't have sufficient coverage at some place in the network, you
simply reach into your bag of nodes and stick another on the wall.
I'd like to describe four different scenarios in which these
high-density, low-cost networks would be a win.
The first application, and perhaps the least astonishing, is simply
another wireless LAN. In a corporate environment, you could connect
your computer or printer to the network simply plugging a Hyphos node
into it. This eliminates the junction boxes, internal wiring, wiring
closets, and the cost associated with the person that maintains all
the cross connections.
Active Inventory: manufacturing, warehousing, shipping
The second application sector is manufacturing, warehousing, and
shipping, and something I've started to call "active inventory
systems." One of the Media Lab sponsors is Steelcase. Aside from the
ubiquotous filing cabinet, a large piece of their business is
manufacturing custom office furniture. They'd like to use Hyphos
networking in their manufacturing process.
For example, to make a chair, they'd embed the particulars of the
custom order in a Hyphos node and send the node through the
production line. The Hyphos node would specify the make, model,
fabric color of the chair. As the Hyphos node moves through the
production line, the chair would literally be built around the node.
At the end of the production line, the chair would be put in a
warehouse for shipment. A truck entering the warehouse would
communicate with the Hyphos nodes embedded in all the furniture, and
could learn which goods are to be carried on that truck.
Industrial controls and sensors
Hyphos networks can also be used to connect industrial controls and sensors.
Sometimes wire just isn't practical. An industrial sensor that costs
only $50 may require $5000 in conduit, code compliance inspection,
and union electricians to install. A low-cost, wireless connection
would be popular in this setting.
Another Media Lab sponsor is United Technologies. Their subsidiary,
Carrier Systems, builds the big air conditioning units that sit on
top of office buildings. The bane of their world is wires on
thermostats. It's already enough of a problem to locate thermostats
at the optimal location for sensing room temperature.
But every time there's a corporate reorganization, the walls inside
office buildings move around, so Carrier spends lots of time
relocating thermostats. They'd like a low-cost, wireless connection
between their thermostats and the air conditioning units on the roof.
Of course, the right place to place thermostats is directly on the
bodies of the people in a building, but the we'll save the topic of
"Body Area Networks" for another talk.
Home and "last meter" delivery
Perhaps the most insteresting application area for Hyphos networks is
in the home, where we can really start to connect everyday objects to
Why should you ever have to set a clock? When a clock is on the
network, it can contact the local Network Time Protocol server and
always be within a few milliseconds of the cesium clocks at the
National Bureau of Standards.
A smoke detector that beeps in the basement isn't effective if you're
asleep on the third floor. When it's connected to a network, it will
be able to alert you regardless of where you are, even if you're
driving on your way to work.
Your washing machine, if it's feeling ill, can contact www.maytag.com
and download some diagnostic software. And if it detects a problem,
it can alert the local Maytag repairman that he finally has a job.
All the major appliances in your house could negotiate with the local
power company to cut back on usage a few percent during peak hours.
By reducing the peak loads, this could save the utility companies
billions of dollars.
My local telephone company charges $75 to do any internal wiring
inside a house. Using a Hyphos network, the phone line terminates
ouside the house and the "last meter" connections are made wirelessly.
In fact, some enterprising service provider could subsidize the cost
of a "hyphos gateway" for use in the home. For a monthly fee, all of
the everyday objects on a Hyphos network now have access to the
internet. Your VCR will have the entire TV guide available to it. A
child's toy can be in contact with www.disney.com to download new
learning activities every day.
So maybe Sun Microsystems had it right: the network is the computer.
But for those of those who still think that computers are the
important component, consider this: when everyday objects become
connected to digital networks, these same objects can become input
and output devices for computers. Among other things, this will give
computers much richer means of interaction with people and with the
environment. Before long, we'll think it strange that people ever sat
at a screen with a keyboard and a mouse.