Impact of Sensor Networks on Future InterNet Design
David E. CullerUniversity of California, Berkeley
Arched Rock [email protected]
NSF FIND Info Meeting12-5-2005
12/5/05 NSF FIND 2
What does the Internet look like in 10 years?
0 2 4 6 8 10 12 14 16 18
-1
-0.5
0
0.5
1
Low resolution Sensor, Test4, Increasing frequency
Time (sec)
Acc
eler
atio
n (g
)
12/5/05 NSF FIND 3
In 10 years…
• 90% of the nodes on the “Internet” will embedded devices connected to the physical world
• Universal, host-host file-transfer and console access is the dominant usage pattern…..
NOT!
• So does it make sense to pay attention to the characteristics of these kind of nodes and applications in designing the future Internet?
12/5/05 NSF FIND 4
Transit Network (IP or not)
Access point - Base station - Proxy
Sensor Patch
Patch Network
Data Service
Intranet/Internet (IP)
Client Data Browsingand Processing
Sensor Node
GatewayGateway
Verification links
Other information sources
Sensor Node
Canonical Sensor Net Architecture Today
•An Analysis of a Large Scale Habitat Monitoring Application, Szewczyk, Polastre, Mainwaring, Anderson, and Culler, Sensys04
12/5/05 NSF FIND 5
The Next Tier
• Small sensors will be the most common nodes on the internet
• How will they be represented and accessed?
Clients Servers
Sensor Nets
12/5/05 NSF FIND 6
How will SensorNets and IP play together?
802.15.4, CC, …802.11Ethernet Sonet
IP
TCP / UDP
HTTP / FTP / SNMP
XML / RPC / REST / SOAP / OSGI
12/5/05 NSF FIND 7
Full IP stack throughout
802.15.4, CC, …802.11Ethernet Sonet
TCP / UDP
HTTP / FTP / SNMP
XML / RPC / REST / SOAP / OSGI
IP
12/5/05 NSF FIND 8
Beware “IP hype”
• Transmitting HTML over a wireless connection to a serial port attached to a PC is NOT running IP on the sensor network
12/5/05 NSF FIND 9
Where has Internet Research Reached and “struggled”?
• Aggregate communication => Multicast
• Resource constraints => QoS, DIFFSERV
• Communicate with data or logical services, not
just devices => URNs (DHTs?)
• Mobility => MobileIP, MANET
• In-network processing and storage => ActiveNets
• Intermittent connectivity => DTN ???
12/5/05 NSF FIND 10
What are the main characteristics of Sensor Networks?• Aggregate communication
– dissemination, data collection, aggregation
• Resource constraints– Limited bandwidth, limited storage, limited energy
• Communicate with data or logical services, not just devices
– Datacentric
• Mobility– Devices moving, tags, networks moving through networks
• In-network processing and storage– Really
• Intermittent connectivity– Low-power operation, out of range, obstructions
12/5/05 NSF FIND 11
Facing these challenges
• Today, we use a wide range of ad hoc, application specific techniques in the SensorNet patch
– Zillion different low-power MACs
– Many link-specific, app-specific multihop routing protocols
– Epidemic dissemination, directed diffusion, synopsis diffusion, …
– All sorts of communication scheduling and power management techniques
12/5/05 NSF FIND 12
Edge Network Approach
802.15.4, CC, …802.11Ethernet Sonet
IP
TCP / UDP
HTTP / FTP / SNMP
XML / RPC / REST / SOAP / OSGI
Pro
xy
/ Ga
tew
ay
12/5/05 NSF FIND 13
“Hacking it in” may not be so bad
• Security– No IP to the nodes, attacks have to get through the gateway or
be physically close
• Namespace management– Name nodes, networks, services– Hosts, URLs, …
• Mask intermittent connectivity– Terminate IP on the powered side– Loosely couple, energy aware protocols on the other
• Distillation proxies– Small binary packets where constrained– Expanded to full text, XML, HTML, web services
• Rich suite of networking techniques in the Patch unimpeded by the “ossification” of the rest
12/5/05 NSF FIND 14
Rethinking at Layer 7
IP Overlay NetworkGateways
SensorNet Patch
12/5/05 NSF FIND 15
Opportunity to rethink more deeply
• No dusty-decks yet
• Not a bunch of laptops running around with their sockets open trying to route through other laptops running around…
• Meaningful set of applications and associated traffic loads
– Environments, individual objects, interactions
• Chance to think through control as well as monitoring
• Physical embedding matters
• Techniques are likely to apply to the rest of the Internet
12/5/05 NSF FIND 16
Traditional Analysis
Delivered Performance
Offered Load
12/5/05 NSF FIND 17
Analysis that really matters
Reliability
Energy Expended
Delay
Traffic Load
Traffic Variability
Environmental variability
Bandwidth
Changes in network population
Mobility
12/5/05 NSF FIND 18
SensorNets need the Wisdom of the “Internet Architecture”
• Design for change!• Network protocols must work over a wide
variety of links– Links will evolve
• Network protocols must work for a variety of applications
– Applications will evolve
• Provide only simple primitives– Don’t confuse the networking standard with a
programming methodology
• Don’t try to lock-in your advantage in the spec
• Open process• Rough consensus AND running code
12/5/05 NSF FIND 19
XETF (Xternet Engineering and Technology Forum) ???
• Mission– Foster an open, innovative, and technically sound ecosystem around
interconnecting the physical world with modern networking and information technology through the creation of technical documents, protocols, reference implementations and APIs.
• Structure– Lean. Volunteer: BOD, steering comm., working groups.
• Membership– Individuals, corporate, academic, and gov’t
• Participation– Open. Role determined by contribution.
• IP Policy– Non-confidential. Disclosure and Contribution process. – Companies can develop own implementation.– BSD? Apache-like credit? MPL? LGPL?
• Output– RFC-like documents, reference implementations, forum for exchange and viz.– “Rough consensus AND running code”
12/5/05 NSF FIND 20
Uniting long-lost relatives
General Purpose Computing Instrumentation Computers
Mainframe
Minicomputer
Workstation
PC
VME
Dedicated Controllers
Home Automation
Building Automation
12/5/05 NSF FIND 21
Tides of Change
Log Stuff
Time
The successor emerges when prior regime is at its apex of strength – not at a point of weakness.
What was previously hard becomes easy, but its successor becomes possible…
Integration
Innovation
The Future Internet probably exists today; go find it
12/5/05 NSF FIND 22
Discussion