1. Intelligent Fluid infrastructure for Embedded Networks Aman Kansal, Arun A Somasundara, David D Jea, Mani B Srivastava, Deborah Estrin. University of California, Los Angeles (UCLA)
MobiSys ‘04, June 6-9, ACM 2004

2. Outline Overview: Hardware Prototype Control Primitives
The Idea
Advantages
The system – components of the infrastructure
Hardware Prototype
Control Primitives
Communication Protocol
Experimental Results
Future Work

3. Overview – General Idea
Fluid Infrastructure: mobile components built into the system for enabling specific functionality that is very hard to achieve using other methods
Built-in intelligence to adapt to achieve objectives
Energy constraints, security, delay, sparse networks
Founded on real world behavior of wireless devices; not based on radio models; Real time test
Focus on Network Layer Functionality
Sensor networks
Mobility:
Random: whales, zebra
Predicted: a network node mounted on a bus and acts like a base station for collecting data
Controlled: introduce an element of deliberate control over mobility; controlled by the infrastructure
Increase performance

4. Overview - Advantages System lifetime 
Battery capacity of embedded static nodes. Replace/Recharge not feasible: Physically not reachable, cost
A mobile agent helps conserve energy at static nodes:
Reduce number of hops
Reduce number of packets transmitted by static nodes
Mobile nodes are not energy-constrained
Static nodes are energy-constrained
Assume Data gathering protocol:
Directed Diffusion: broadcast INTEREST,
Data in response via nodes from which INTEREST heard.
1-5 hops ; 1-2 hops
- INTEREST Transmission = 60s.
- INTEREST Expire = 75s
- Sample Generation Period = 5s
- Samples Per Data Packet = 4
- 16 minutes simulation period, TOSSIM
- Stop at each node 60s

5. Overview - Advantages Quality of data  Reduce number of hops
Probability of errors ↓
Reduce retransmissions at static nodes: energy conserved

6. Overview - Advantages Data Rate ; or Latency ↓
Network capacity increased; physically carrying data in mobile router
Collisions reduced
Less hops: error rate ↓
Mobile device has connection with the base; used only when at Base
mica2 motes
Each mote: 512KB flash M; 500KB
stored
38.4Kbps baud rate
Max speed of router: 388 cm/s; 200 cm/s
used
Tideal = T(A,A,B) + T(A,B,base) + T(B,B,base)
Tmobile = D(base,A) + T(A,A,mobile_router) + D(A,B) + T(B,B,mobile_router) +
D(B,base) + T({A,B},mobile_router,base).

7. Overview - Advantages
Handle Sparse (low density) & Disconnected Networks
Reduce transmission range to lowest value to reach the mobile router; energy conserved
To ensure connectivity [14]:
Communication range = 2*Sensing range
Use mobile agent instead of relay nodes that are just for connectivity; energy conserved

8. Overview - Advantages Others:
Time synchronization error decreases; less hops
Security enhanced; data travels across less hops

9. Overview – The System Prototype hardware Control primitives
A sensor network
An autonomous mobile router
Real hardware for test; usable methods
Control primitives
Control the motion of mobile routing components
Communication protocol
Changes with the nature of the system; presence of mobile router

10. System Architecture Motes: Static embedded sensor nodes; mica2 [14]
A mobile router: the fluid infrastructure connecting the static nodes to the base
(The base could be a human
user based system, or
controller)

11. System Architecture A traction platform: Packbot
A rugged terrain robot
Unmanned ground vehicle
interface to get commands
SIR (Simple Interface for Robots):
only 2 API functions roboAct(), roboAdd() to move, rotate, and flip
Runs only on Linux, and TinyOS
Supports 2 robots: Packbot, and Amigobot
robotAct(id, ‘f’, 30): robot of ID ‘id’ perform command ‘f’ (move forward) with parameter ’30’ (30 cm/s)
robotAct(id, ‘m’, 30, 1.0, 1.0): robot ‘id’ move to coordinate (1.0, 1.0) at 30 cm/s; 2D.

12. System Architecture Stargate [20] Processing board; xScale processor
card:
Communicate with the Base
Send commands to the Packbot
A mote network interface card (NIC):
Communicate with static motes
Software:
Localization, Navigation not complete
Assumption: obstacle free environment

13. System Architecture The mobile router’s Mote
Relays data received from radio to Stargate
Relays data from Stargate to radio
Can be used to transfer data to the Base

14. Adaptive Motion Control
2 factors:
Constraints due to the terrain
Depending on terrain: router can visit all nodes or a subset
1st Fluid Infrastructure is for gathering data from an ecosystem
Mobility is restricted, and could lead to a disturbance to the habitat
Motion control algorithms designed with the path traversed by the mobile node agent is fixed
The data collection performance parameters

15. Adaptive Motion Control
2 factors:
Constraints due to the terrain
The data collection performance parameters
Application priorities can be considered in the algorithm
Max life time: visit each node, least amount of energy in transmitting data should be used by embedded nodes
Total amount of data collected:
static nodes generate data at rapid rate, router transfer max possible data to Base.
Mobile router could use a hybrid topology (Multi-Hop & mobile infrastructure)

16. Adaptive Motion Control
2 factors:
Constraints due to the terrain
The data collection performance parameters
Latency: maximize number of data collected within a period; high speed of router
Limited buffers of static nodes :
router to collect data before overflow
Multi-hope may be needed for part of transmissions
Higher speeds for mobile router, frequent traverses
Delay needed to charge the router’s battery:
added to data latency if data is continuously collected,
otherwise it is not a factor and intervals of no activity will exist

17. Adaptive Motion Control
Speed Influence on Data Collection
No account for any influence of speed in collision free radio channel for the motion control algorithm
2 motes out of range, continuously transmitting
The influence of speed only
Speed s, time t to cover the trail, n packets
2s, t to cover the trail twice, n/2 packets per round
results of each speed averaged over 3 rounds of trail

18. Adaptive Motion Control
Speed Influence on Data Collection

19. Adaptive Motion Control
Latency Sensitive Data Collection
T = The max latency of the time for reporting sensor readings
T = The max time the mobile router spends to complete one round
Objective:
maximize amount of data collected within T
Naïve Approach: get the speed of the mobile router from T, the length of round.

20. Adaptive Motion Control
Latency Sensitive Data Collection
Network conditions:
Certain nodes in range of the router for longer duration; closer to the path
More than one node in range of the router;
Communication BW will be divided among nodes
MAC collisions
Wireless channel may introduce errors at particular nodes at certain times; data rate reduced for such nodes
Router moves slower/ stops when more time needed; to increase the data collected
Router speeds up when higher data rates achieved

21. Adaptive Motion Control
Latency Sensitive Data Collection
An algorithm that:
Is founded on real world behavior of wireless devices; not based on radio models which is not practical; Real time test
Does not use geographical info about location of static nodes:
Location could be not available
Error in location estimate can be significant
Does not require the router to know number of static nodes
The number changes: adding nodes, node damage, battery death.
Router estimates this number using node ID in received data
The router to adapt by learning about network conditions (regions of congestion, poor transmission, or high data rates)

22. Adaptive Motion Control
Latency Sensitive Data Collection
T: latency for round traversal time
Naïve speed s.
At speed 2s, router can complete the round at T/2
Extra T/2 for collecting data in regions that is congested, or have poor transmission of data
2 approaches to manage the extra time
SCD: Stop at locations where nodes are found waiting with data
ASC: Move slower in regions where data collection is poor and stop in regions where data loss is severe

23. Adaptive Motion Control
SCD: Stop to Collect Data Algorithm
No need to know locations
No need for time synchronization
No deadlocks
SCD is proposed in case of latency sensitive communication in Sparse Networks

24. Adaptive Motion Control
ASC: Adaptive Speed Control
Stargate is sufficient for processing; not energy-constrained
The router decides when to slow down, stop, or speed up depending on communication characteristics of the deployed network.
Router keeps State Info that is based on received data from static nodes
Every packet’s header from any static node includes:
The remaining number of data samples it wishes to transfer
Using first and last packets from a nodes: the percentage d of total samples that were received
** 4 Sample sent per packet

25. Adaptive Motion Control
ASC: Adaptive Speed Control
d = n/(s1 + (s2 - s1) + n) ; % of received samples
= n/(s2 + n)
Where:
n = number of unique samples received from a node
s1 = number of remaining samples count as per 1st pkt
s2 = number of remaining samples count as per last pkt
** Also used for number of samples collected after sending 1st pkt
** di = d for a node; Router keeps d for every node it hears from

26. Adaptive Motion Control
ASC: Adaptive Speed Control
2 thresholds are selected 0 < ω1 < ω2 < 100
2 classes of nodes:
N1: di < ω1
The router will stop if encounters N1 node
N2: ω1 < di < ω2
The router will slow down to speed s if encounters N2 node
Otherwise: move at 2s.
If n1 from N1, and n2 from N2:
∆ = T/2 / ( n1 + n2/2)
Gives extra time to N1 nodes (double):
Moving at 2s then stops for ∆ : lost time is ∆
Moving at 2s then slows down to s for : lost time is ∆/2

27. Adaptive Motion Control
ASC: Adaptive Speed Control
Router can be in one of 3 states:
SL: encounters a node of class N2
ST: encounters a node of class N1
TE: current state timer expired
M: the timer,
m: elapsed time since timer was reset
δ: duration to which timer reset
can be different at different resets

28. Communication Protocols
Based on Directed Diffusion
No interest in addresses of nodes, but in data elements
Designed to collect data from a large number of sensor nodes
Interest message
Data constraints
TTL: number of hops for which Interest will be forwarded
Each node getting the Interest: store it in Interest-cache, decrease the TTL, and rebroadcast if not TTL is 0
Reverse path toward the Router is built and used to forward data from sensor nodes to router
Whenever a node has data:
check the cache for data constraints,
If constraints met, forward data to the node from which it heard Interest
Repeated till data delivered to sink (the router)
Expiration timer is used to handle different changes and errors in net.

29. Communication Protocols
Modification is needed
Requirements:
Static nodes needed to send data directly to router if possible; otherwise use multi-hop communication to reach the nearest node to the router
1st modification:
1st round is used to learn about connectivity of mobile router
For every Interest received, each node records if it is from the router or other static node
If heard from the router, the node will not respond to or rebroadcast any subsequent Interest from other static nodes.
Only rebroadcast Interest from the router

30. Communication Protocols
The mobile router may move out of range while the Interest didn’t expire; data sent in this duration is lost
2nd modification:
Use of ACK retransmit scheme
Mobile router sends ACK to the node sent the data
The node will not send next packet till it gets the ACK
Retransmit timeout, resend data
Also, this recovers from lost data
When Interest expires, no data is sent
After hearing new Interest, the node may complete from where it was or begin new data transmission depending on the Interest message

31. Communication Protocols

32. Communication Protocols
Forwarding nodes can pre-fetch data from nodes for which they forward data
From 2nd round.
If a node hears a response to an Interest transmitted by it, it knows that others forward data through it
The node starts local diffusion, Interest with TTL = 1
Nodes receiving this, may begin local diffusion (they have node that forward through them)
No need to know exact topology of net, or length of path

33. Communication Protocols
Sleep mode:
Radio is a major power consuming component, and must be on when the router is in range and there is data to deliver
While ACK is received, router is in range
Time synchronization is needed: T for one round could be made known to sensor nodes at design, using Interest message, or run time using

34. Experiments Dense Network
T = 72s to keep s and 2s within limits of the robot
Speeds: Naïve 50cm/s, Fast 100cm/s, Slow 50cm/s
ASC: ω1 25%, ω2 75%
Results averaged over 7 rounds

35. Experiments Sparse Network
2 groups out of range; nodes of same group are in range

36. Experiments Sparse Network
Connectivity does not exist without mobile router
SCD is used

37. Future Work Adding navigational support Several paths for the router
Multiple mobile routers/nodes are available
Enhancements of the communication protocol
Interaction between sleep management and local diffusion with motion control primitives

38. Thank You