1. CS 410/510 Sensor Networks Portland State University
Lecture 4
Multi-hop Routing

2. Why use multi-hop communication?
Optimize energy consumption
Transmission power proportional to d^n (n > 2); shorter links better
Enable spectrum reuse
Can simultaneously transmit at spatially distant parts of the network
11/19/2018
Nirupama Bulusu

3. What is Routing?
Act of moving information across a multi-hop network from a source to a destination
Includes
Path determination: Determining optimal routing paths
“Optimal” varies depending on the routing metric.
Packet switching: Transporting information groups, called packets through a network
11/19/2018
Nirupama Bulusu

4. Communication Models in Sensor Networks
Dissemination (1-many)
Eg. code updates
from base station to
sensors (eg. Trickle,
Deluge)
11/19/2018
Nirupama Bulusu

5. Communication Models in Sensor Networks
Data Gathering (many-1)
11/19/2018
Nirupama Bulusu

6. Communication Models in Sensor Networks
Any-to-any communication
Eg. Triggering
another sensor
To start sensing
(eg. GPSR)
11/19/2018
Nirupama Bulusu

7. Example Tree based routing for data collection
Reliable end-to-end packet delivery with minimum number of transmissions (link retransmissions)
Advocate stability
Simple
11/19/2018
Nirupama Bulusu

8. Routing Techniques Directed Diffusion, DSR, AODV Essence:
reverse path routing by flooding
non-shortest path routing (do switch if a shorter path is found)
DSR, AODV (On-demand routing)
but every node is a source
source initiated route request broadcasts
“local storm” of replies
11/19/2018
Nirupama Bulusu

9. Tree-based routing drawbacks
Root of the tree is unique point of failure
Paths become longer and longer
Increased end-to-end packet loss, congestion
Decreased network lifetime
Poor load balancing
=> Nodes near the root carry more traffic
11/19/2018
Nirupama Bulusu

10. Clustering: From tree to forest
server
WiFi
Micro-server
Low power radio link
(eg. Zigbee)
11/19/2018
Nirupama Bulusu

11. Outline One-many communication example: Trickle
Any-any communication example: GPSR
Many-one communication example: Directed Diffusion
Issues in sensor net routing
11/19/2018
Nirupama Bulusu

12. Trickle: To Every Node in a Network
Network membership is not static
– Loss
– Transient disconnection
– Repopulation
Limited resources prevent storing complete network
population information
To ensure dissemination to every node, we must
periodically maintain that every node has the data.
11/19/2018
Nirupama Bulusu

13. The Real Cost Propagation is costly
Virtual programs (Maté, TinyDB): bytes
– Parameters, predicates: 8-20 bytes
– To every node in a large, multihop network…
But maintenance is more so
– For example, one maintenance transmission every minute
– Maintenance for 15 minutes costs more than 400B of data
– For 8-20B of data, two minutes are more costly!
Maintaining that everyone has the data costs more
than propagating the data itself.
11/19/2018
Nirupama Bulusu

14. Solution: Trickle
“Every once in a while, broadcast what data you have, unless you’ve heard some other nodes broadcast the same thing recently.”
Behavior (simulation and deployment):
– Maintenance: a few sends per hour
– Propagation: less than a minute
– Scalability: thousand-fold density changes
Instead of flooding a network, establish a trickle of packets, just enough to stay up to date.
11/19/2018
Nirupama Bulusu

15. Detecting That a Node Needs an Update
As long as each node communicates with others, inconsistencies will be found
Either reception or transmission is sufficient
Define a desired detection latency, t
Choose a redundancy constant k
– k = (receptions + transmissions)
– In an interval of length t
Trickle keeps the rate as close to k/ t as possible
11/19/2018
Nirupama Bulusu

16. Trickle Algorithm Time interval of length t
Redundancy constant k (e.g., 1, 2)
Maintain a counter c
Pick a time t from [0, t]
At time t, transmit metadata if c < k
Increment c when you hear identical metadata to your own
Transmit updates when you hear older metadata
At end of t, pick a new t
11/19/2018
Nirupama Bulusu

17. Maintenance with Loss: The logarithmic behavior represents the probability that a single node misses a no. of transmissions.
11/19/2018
Nirupama Bulusu

18. Summary Trickle scales logarithmically with density
Can obtain rapid propagation with low maintenance
In example deployment, maintenance of a few sends/hour, propagation of 30 seconds
Controls a transmission rate over space
Coupling between network and the physical world
Trickle is a nameless protocol
Uses wireless connectivity as an implicit naming scheme
No name management, neighbor lists...
Stateless operation (well, eleven bytes)
11/19/2018
Nirupama Bulusu

19. Outline One-many communication example: Trickle
Any-any communication example: GPSR
Many-one communication example: Directed Diffusion
Issues in sensor net routing
11/19/2018
Nirupama Bulusu

20. Geographic Routing One way of supporting any-any communication
Use location of nodes as addresses for routing
Example:
GPSR (Greedy Perimeter Stateless Routing)
11/19/2018
Nirupama Bulusu

21. Algorithm : Greedy Forwarding
The next hop from a node is the neighbor that is geographically closest to the packet’s destination.
11/19/2018
Nirupama Bulusu

22. Algorithm: Greedy Forwarding
Beaconing mechanism
Provides all nodes with neighbors’ positions.
Beacon contains broadcast MAC and position.
To minimize costs
Piggybacking
Promiscuous mode
11/19/2018
Nirupama Bulusu

23. Algorithm: Greedy Forwarding
Drawback !!!
11/19/2018
Nirupama Bulusu

24. Algorithm: Greedy Perimeter Routing
When greedy forwarding fails, switch packet mode to Perimeter
Use Perimeter Forwarding
11/19/2018
Nirupama Bulusu

25. Perimeter Forwarding uses Right Hand Rule
When arriving at a node x from node y, the next edge traversed is the next one sequentially counterclockwise about x from edge (x, y)
11/19/2018
Nirupama Bulusu

26. Planarized Graphs
A graph in which no two edges cross is known as planar.
Right-hand-rule fails on non-planar graphs
So, planarize it! Eg. Locally compute
Relative Neighborhood Graph (RNG)
Gabriel Graph (GG)
Run RHR on planar graph
11/19/2018
Nirupama Bulusu

27. GPSR: Pros and Cons Pros: Cons:
Low per router state for large number of network destinations
Handles mobility very well
Small routing protocol message complexity
Cons:
GPS location system might not be available everywhere.
Overhead in location registration and lookup
Planarization affected if nodes within another node’s radio range
11/19/2018
Nirupama Bulusu

28. Outline One-many communication example: Trickle
Any-any communication example: GPSR
Many-one communication example: Directed Diffusion
Issues in sensor net routing
11/19/2018
Nirupama Bulusu

29. Directed Diffusion
Directed diffusion is a new data dissemination paradigm for sensor networks
Data-centric
Data propagation and aggregation is determined by localized interactions between nodes: hop-to-hop rather than end-to-end
Long-range communication requires more energy
Hop-to-hop communication provides link diversity, helps overcome obstacles
Incorporates application-specific semantics
11/19/2018
Nirupama Bulusu

30. Directed Diffusion Elements
Gradients
Network paths
Reinforcement
Manages gradients
Named data
Described with attribute-value pairs
Interests
Describe sensing tasks
11/19/2018
Nirupama Bulusu

31. Naming Task descriptions are named by a list of attribute-value pairs:
Intuitively, the task description specifies an interest for data matching the attributes. For this reason, such a task description is called an interest
11/19/2018
Nirupama Bulusu

32. Interest Dissemination
C’s Interest cache C
Interests Gradients
source B
sink
Sink disseminates interest for a four-legged animal (~36 bytes).
Initial interval is large.
11/19/2018
Nirupama Bulusu

33. Interest Dissemination
C’s Interest cache C
Interests Gradients
source B
sink
Every node contains an interest cache, with separate entries for distinct interests.
Entries do not contain info about sinks and therefore scale well.
Overlapping entries may be aggregated for efficiency.
11/19/2018
Nirupama Bulusu

34. Interest Dissemination
C’s Interest cache C
Interests Gradients
| B: 1s
Sink: 1s
source: 1s
source B
Each interest cache entry contains a list of gradients; events that match interest entries are propagated back to the sink via these gradients. Gradient entries contain locally unique neighbor IDs, data rates, and interval attributes.
sink
In the absence of information about which sensor nodes are likely to be able to satisfy an interest, interests are broadcasted to all neighbors.
11/19/2018
Nirupama Bulusu
However, a node may suppress a received interest if it recently re-sent a matching request.

35. Data Propagation C | B: 1s source B sink
C’s Data cache
C’s Interest cache
1 eps C
Interests Gradients
EVENT
| B: 1s
Sink: 1s
source: 1s
source
1 eps
1 eps B
1 eps
1 eps
Initial interests request data at slow rates (e.g. 1 event per second).
sink
A sensor node that is able to furnish a query-result searches its interest cache for a matching entry; if it finds one, it begins sending data messages (~64 bytes) towards the sink via its gradient list at the highest specified rate.
11/19/2018
Nirupama Bulusu

36. Data Propagation C | B: 1s source B sink
C’s Data cache
C’s Interest cache
1 eps C
Interests Gradients
EVENT
match
| B: 1s
Sink: 1s
source: 1s
source
1 eps
1 eps B
1 eps
1 eps
Upon receiving a data message, nodes check their interest caches. If no match is found, the data message is silently dropped.
sink
If a match is found, the node checks its data cache, which keeps track of recently seen data items. If no data cache entry matches the message, a new entry is made in the data cache and the message is re-sent to the node’s neighbors.
11/19/2018
If a data cache entry matches the data message, the message is silently dropped, thus, preventing loops.
Nirupama Bulusu

37. Reinforcement C | B: 1s source B sink
C’s Data cache
C’s Interest cache
100 eps
1 eps C
Interests Gradients
EVENT
| B: 1s
Sink: 1s
S: .01s
source: 1s
source
1 eps
1 eps B
100 eps
1 eps
1 eps
After the sink starts receiving these low data rate events, it reinforces one particular neighbor in order to “draw down” higher quality (higher data rate) events.
sink
It does this explicitly by re-sending the original interest message, but with a smaller interval value, to the empirically low delay path node.
Nodes update their caches and can then propagate reinforcement messages according to local policies. For example, the node might choose that neighbor from whom it first received the latest event matching the interest
11/19/2018
Nirupama Bulusu

38. Outline One-many communication example: Trickle
Any-any communication example: GPSR
Many-one communication example
Issues in sensor net routing
11/19/2018
Nirupama Bulusu

39. Theme: Why is sensor net routing challenging?
Explore underlying communication issues and their effects on multi-hop routing protocols
Single hop
Zhao and Govindan, SenSys 2003
Extra: SCALE (Cerpa et al. CENS TR 2003)
Network-level protocol
multi-hop routing for data collection. (Woo et al., SenSys 2003)
11/19/2018
Nirupama Bulusu

40. Motivation Why do we care about all these details?
we will experience it!
if we want to implement our protocols
actual deployment decision
11/19/2018
Nirupama Bulusu

41. Outline Network Services Link estimation Neighborhood management
Reliable multi-hop routing for data collection
11/19/2018
Nirupama Bulusu

42. Link Quality Estimation
Estimate rate of successful reception from neighboring nodes
RSSI may not work well
Neighbors exchange estimations to derive bi-directional link quality
2 Techniques: Passive vs. Active
Key decision factor: broadcast medium
Passive: snoop on neighbor packets
11/19/2018
Nirupama Bulusu

43. What is a good Estimator?
For a given error bound and agility,
yields the most stable estimation
with the smallest memory footprint
and the simplest algorithm
Agility and stability are at odds with each other
11/19/2018
Nirupama Bulusu

44. Agility and Error Bound
Simulation worst case: 10% error ~ 100 packet time
Assuming IID Binomial model, by the central limit theorem
Worst case (p = 0.5) requires
10% error with 90% confidence requires ~100 packet opportunities to learn
For example: at 30sec/packet
50 minutes for 100 packets
forwarding traffic helps to reduce this time but potentially a long delay
Major disadvantage
11/19/2018
Nirupama Bulusu

45. Infer Packet Loss Packet sequence number for inferring packet loss
Issue: cannot infer loss until hearing the next packet
E.g. dead node or mobility
Can infer losses based on time
Assume minimum data rate is known
Likely to be true in periodic data collection
11/19/2018
Nirupama Bulusu

46. WMEWMA Estimator
Compute an average success rate over time, T, and smoothes with an exponentially weighted moving average (EWMA)
Average calculation
Packets Received over T divided by
Max of
Number of packets expected over T
Number of packets sent over T suggested by sequence number
Tuning parameters:
T and history size of EWMA
Performance
Yields agile and stable estimations
Uses constant memory, and is very simple
11/19/2018
Nirupama Bulusu

47. Neighbor Table
Maintain link estimation statistics and routing information of each neighbor
How large should this table be?
O(cell density) * meta-data for each neighbor
Issue:
Density can be high but memory is limited
At high density, many links are poor or asymmetric
Question:
Can we use constant memory to maintain a set of good neighbors regardless of cell density?
11/19/2018
Nirupama Bulusu

48. Neighborhood Management
Question: when table becomes full,
should we add new neighbor?
If so, evict which old neighbor?
Neighbor Goodness
Basic one is link quality but it is unknown
Signal strength is a hint
Rely on frequency of packet reception
Assume periodic data packets or beacons
Similar to
frequency estimation of data streams, or
classical cache policy
11/19/2018
Nirupama Bulusu

49. Management Algorithm When we hear a node, if Eviction: (FREQUENCY)
In table: increment a counter for this node
Not in table
Insert if table is not full
down-sample if table is full
If successful, insert only if some nodes can be evicted
Eviction: (FREQUENCY)
Decrement counter for each table entry
Nodes with counter = 0 can be evicted
Otherwise, all nodes stay in the table
11/19/2018
Nirupama Bulusu

50. Key Results FREQUENCY algorithm can effectively Routing simulation:
utilize 50% to 70% of the table space to maintain a set of good neighbors
while being adaptive to neighborhood changes
Routing simulation:
Neighbor goodness is augmented to avoid maintaining sibling nodes based on routing cost difference
11/19/2018
Nirupama Bulusu

51. Ad Hoc Routing Protocol
creates a many-to-one spanning tree topology
self-organizing through local operations
simple and efficient
explore quality/reliable routing paths to the base station
11/19/2018
Nirupama Bulusu

52. Pitfalls Assumption that links are “inherently” good
assumes link layer abstraction provides good links and pay little attention below network layer
Reverse link quality is “as good”
routing (DSR, Diffusion, AODV) using reverse path routing
Fixed consecutive # of failures = link failures
“semi-good” links are easily treated as failures
create instability
11/19/2018
Nirupama Bulusu

53. Shortest Path is Good? Sink 90% 50% 90% Source
Exp num. trans. from A->B->Sink
Assume link reliability is symmetric
1/(90% * 90%) + 1/(90% * 90%) = 2.47
Routing objective:
Minimize (1/(pfi * pri))
where pfi and pri= forward and reverse link reliability at each i along the path
90%
Path
Hop
Count
Exp. Num. Trans.
A->Sink 1 4
A->B->Sink 2
2.47
50% B
90% A
Source
11/19/2018
Nirupama Bulusu

54. Lossy Links clear transitional silent Data = Connectivity
matrix of 20 Micas
on a line at a
particular trans.
power setting.
Each dot represents
reliability of a
directional link at a
given distance.
11/19/2018
Nirupama Bulusu

55. Hypothesis Shortest path routing can yield unreliable paths
Build reliability statistics of each neighbor through link estimations
Even coarse estimations are better than none
End-to-end reliability guarantee is unlikely
even if it exists, local actions are building blocks
Limited retransmissions per hop
Explore reliable paths to route packets
E( transmissions) along a path captures both “distance” and “reliability” for routing
ARPANET available bandwidth metric leads to congestion
11/19/2018
Nirupama Bulusu

56. Reliable Routing 3 core components for Routing Example
Routing protocol
Neighbor table management
Link estimation
Example
Tree based routing for data collection
Reliable end-to-end packet delivery with minimum number of transmissions (link retransmissions)
Advocate stability
Simple
11/19/2018
Nirupama Bulusu

57. Design Issues Shortest path alone yields poor end-to-end success rate
Multi-hop over bad links has exponential loss effect
Two approaches
Shortest Path over some link quality threshold
Minimum expected number of transmissions as routing cost
Route damping:
New route is not evaluated on every route update
Link failure detection using consecutive packet loss leads to instability
link quality characterization is better
Queuing Policy
Two queues
Fair allocation between forwarding and originating traffic
Cycles
Detection vs. loop-free
11/19/2018
Nirupama Bulusu

58. Shortest Path with Threshold (SP)
High threshold (e.g. 70%) fails to form a tree
Works fine in simulation!
link quality degrades when there is traffic
High threshold leads to network partition
Echo the observation made in Zhao’s work
Lower threshold (e.g. 40%) is also problematic
Tree prunes and rebuilds over time when traffic is high
11/19/2018
Nirupama Bulusu

59. Minimum Expected Transmissions (MT)
No predefined threshold is necessary
Captures both reliability and energy cost
Routing cost builds upon individual estimations along the path
Cost = hops + number of expected link retransmissions
if link quality = 100%, MT reduces to normal SP routing
11/19/2018
Nirupama Bulusu

60. Routing Techniques to Evaluate
Directed Diffusion, DSR, AODV
Essence:
reverse path routing by flooding
non-shortest path routing (do switch if a shorter path is found)
DSR, AODV (On-demand routing)
but every node is a source
source initiated route request broadcasts
“local storm” of replies
Evaluation
Sink initiated flooding to approximate reverse path routing
11/19/2018
Nirupama Bulusu

61. Routing Techniques to Evaluate
DSDV
Essence:
“fresher” path overrides “shorter” path
fixed number of failure = link failure detection
Distance vector based routing
Link estimations
Simple moving average link estimator
Sniff link seq. number in packets
Exchange link estimations via route messages
Two cost metrics
Shortest path
Only consider links with 75% reliability or better
Exp. number of transmissions along a path
11/19/2018
Nirupama Bulusu

62. Methodology Graph analysis Network simulation Empirical evaluation
Assuming packet loss are independent, following Binomial model
Empirical evaluation
On site connectivity vs. distance study
Find minimum transmit power
Transitional/gray area starts at average node distance
11/19/2018
Nirupama Bulusu

63. Simulation Platform Matlab simulator packet level simulation
incorporates loss model based on empirical traces
implements TinyOS network stack
visualization of the tree evolution
11/19/2018
Nirupama Bulusu

64. Simulation Results 49 nodes layout on a grid
grid interval = length of “effective” com. range
simulation time 1000s
data rate: 10s/msg
route rate:
10s/msg ( s)
50s/msg (200s – 1000s)
Simple, coarse link estimation is much better than none
Success
to B.S.
# Retrans.
/msg recv. by B.S.
Sink
Broadcast
35.8%
5.4
DSDV
84.8%
1.42
Shortest
Path (75%)
94.7%
0.7
Exp. #
Trans.
96.0%
0.54
11/19/2018
Nirupama Bulusu

65. Findings (I) Hop distribution and success rate
longer majority hop-count yields higher success
Hop distribution and distance
Evidence of long links, potentially reliable
Retransmissions are not too effective
MT yields ~80% success rate
packets delivered only experience 1 retransmission along the path
A maximum of 2 retransmission per hop can
Needs a maximum of 3 per hop to achieve over 90% end-to-end success rate
11/19/2018
Nirupama Bulusu

66. Findings (II)
Link failure detection with consecutive packet loss leads to instability
Stability and Congestion
link quality fluctuates at congestion period
creates global instability
BS can hear half the number of neighbors in the network even with a low power setting
MT metrics build upon link estimations are stable
No cycles are detected
11/19/2018
Nirupama Bulusu