Data Salmon:
A greedy mobile basestation protocol
for efficient data collection in WSNs
Murat Demirbas
Onur Soysal
SUNY Buffalo
Ali Saman Tosun
U. Texas @ San Antonio
Problems with static basestations
1. Static basestation (SB) approach ignores the
spatiotemporally varying nature of data generation
•
Most of the time the network remains idle, with burst of data generation
from a region upon event detection
2. SB approach leads to multihop relaying of high traffic data
•
Multihop relaying of high data-rate traffic consumes energy
•
Collisions result due to high data-rate traffic contending over multihops
2
Work on Mobile Basestations
• Data Mules:

MBs move randomly and collect data opportunistically from sensors

Sensors buffer data until mobile basestation (MB) is within range
• Predictable Data Collection:

Sensors are assumed to know the trajectory of MBs

Sensors buffer data until MB is within range
These work address problem 2
but also introduce latency
3
Work on MBs…
• Mobile Element Scheduling

MB visits sensors such that no sensor buffer overflow occurs

Problem is NP-complete, heuristic solutions provided
• Partition Based Scheduling

Algorithm partitions the network into regions according to data rates

Reduced overall complexity but still NP-complete
These work address problem 2, problem 1 is addressed
only for static/predetermined data generation rates
4
Our work: Data Salmon
• We address the spatiotemporal nature of data generation
by using a network controlled MB
• We achieve low-latency data collection
by maintaining a path to the MB for continuous data forwarding
• We reduce multihop relaying of high data-rate traffic
by devising an algorithm for relocating the MB to the regions that
produce higher data rates
• We prove that our local greedy algorithm is optimal
by showing the convexity of the cost function for our setup
5
Outline of this talk
• Tracking the MB
• Data Salmon algorithm for relocating the MB
• Proof of optimality
• Simulation results
• Extensions
6
Model
• A static WSN
• A mobile basestation

Suspended cableway mobility platform as in NIMS, SkyCam
• A spanning backbone tree over WSN

MB uses the backbone tree to navigate
7
Distributed arrow algorithm
• Assume initially all arrows point to the basestation
• When the MB moves, just flip the direction of traversed edge
Demmer, Herlihy (1998)8
Distributed arrow algorithm
• Assume initially all arrows point to the basestation
• When the MB moves, just flip the direction of traversed edge
Demmer, Herlihy (1998)9
Distributed arrow algorithm
• Assume initially all arrows point to the basestation
• When the MB moves, just flip the direction of traversed edge
10
Demmer, Herlihy (1998)
Distributed arrow algorithm
• Assume initially all arrows point to the basestation
• When the MB moves, just flip the direction of traversed edge
11
Demmer, Herlihy (1998)
Outline of this talk
• Tracking the MB
• Data Salmon algorithm for relocating the MB
• Proof of optimality
• Simulation results
• Extensions
12
MB relocation problem
• Minimize energy consumed for multihop relaying

d(i,j): hop-distance from node i to node j

wi: the data rate of node i

The energy spent for relaying when MB is at m :

The problem is to find optimal m* with minimum M(m*)
• Notation for the algorithm

Total data rate forwarded from subtree rooted at i is

Total data rate at WSN:
εi
13
Greedy algorithm
• Go to a neighbor b with a lower cost function M(b)
• It turns out b is unique if it exists!
M(b)=M(a)+ εa - εb
ε=εa+εb
14
Data Salmon algorithm
???
7
1
1
2
15
Data Salmon algorithm
7
1
1
2
16
Data Salmon algorithm
7
1
1
2
17
Data Salmon algorithm
3
2
4
18
Outline of this talk
• Tracking the MB
• Data Salmon algorithm for relocating the MB
• Proof of optimality
• Simulation results
• Extensions
19
Proof of optimality
B2
Bk
vk
A
B1
v2
v1
v0
• Let v0 be optimal position, vk be any node in tree
• We show that the path to v0 has decreasing cost
• Theorem 2: Path vk,vk-1,…,v0 satisfies M(v0)≤ M(v1)≤ …≤ M(vk)
20
Proof of optimality
B2
Bk
vk
A
B1
v2
v1
v0
When MB moves from v0 to v1

hop distance for all nodes in A increases by 1

hop distance for all nodes in B decreases by 1
≥0; since v0 is optimal!!
21
Proof of optimality
B2
Bk
vk
B1
v2
A
v1 v0
• When MB moves from v1 to v2

hop distance for all nodes in AUB1 increases by 1

hop distance for all nodes in B-B1 decreases by 1
≥0
≥0
22
Outline of this talk
• Tracking the MB
• Data Salmon algorithm for relocating the MB
• Proof of optimality
• Simulation results
• Extensions
23
Energy consumption for SB vs MB
24
Point difference between SB & MB
25
Outline of this talk
• Tracking the MB
• Data Salmon algorithm for relocating the MB
• Proof of optimality
• Simulation results
• Extensions
26
Tree reconfiguration problem
• Static backbone tree leads to hotspot problem & also do not
provide shortest path routing toward MB
• Is it possible/worthwhile to achieve an update-efficient
algorithm for dynamically reconfiguring the tree as the MB
relocates?

NB: Strictly local updating leads to deformed trees soon
27
Multiple MB extension
• Multiple MBs would mean multiple roots (DAG structure)
• When there are multiple outgoing edges in a node the
incoming traffic is equally divided among the outgoing edges

MBs calculate their movement in the same manner (local greedy)

Edge directions are maintained in the same manner
• How do we achieve an optimal multiple MB algorithm?
28
Other extensions
• Use of more general cost functions
• Investigation of buffering at the nodes for buffering/latency
trade-off
29
Summary
• We address the spatiotemporal nature of data generation
by using a network controlled MB
• We achieve low-latency data collection
by maintaining a path to MB for continuous data forwarding
• We reduce multihop relaying of high data-rate traffic
by devising an algorithm for relocating the MB to minimize the average
weighted multihop data traffic
• We prove that our local greedy algorithm is optimal
by showing the convexity of the cost function for our setup
30