IEEE DCOSS’12Mitigate Funnel Effect in Sensor Networks

with Multi-Interface Relay Nodes

Jorge MenaUniversity of California, Los Angeles

Mario GerlaUniversity of California, Los Angeles

Vana KalogerakyAthens University of Economics and Business

17 de Mayo de 2012

2

Network Research Lab (NRL)

Vehicular Network (VaNet) Traffic congestion modeling and car sensor network

projects WiMax/WiFi testbed that spans the UCLA campus. Network Coding

Cognitive Networks (CogNets) Application, protocol stack and Physical sensing

spectrum allocation research Mobile Health Networks

PAN with UVA/UVB sensors research Radioactive particles sensor research (coming soon)

3

Sensors Networks

Transceivers adapted with a sensor devices (temp, motion, particles, etc.) Disposable, low-cost, low-performance

Sink node Powerful, reusable, resourceful Final destination of all data flows

Sensors establish ad-hoc networks with the goals to sense their surrounding environment and generate update data flows to the sink.

Mica2 sensor

4

The problem we address

Funnel effect problem Sensor networks with high activity may generate large

data flows. Flows from remote areas converge at the region that

surrounds the sink Consuming resources: bandwidth, power Duty cycles conserves energy and saves bandwidth but

may disconnect the network Decreases the number of available paths to sink Available paths consume the little bandwidth Sensors compete for bandwidth and scarse

Intense usage of spectrum resources and energy at the areas that surround the sink due to remote flows.

5

Preliminary view for a solution

If the problem is with resources…

… add more…

… or even more.

6

Relay Node Network

Relay nodes Mobile devices adapted with multiple radio interfaces

and larger power and computational resources They establish overlay networks using orthogonal

(non-interfering) channels on top of the sensor network.

More resources are added since there is no competition for the bandwidth

Using a second interface, a relay may pick up and drop off packets from the sensor network

Solution approach: Use an network of relay nodes that uses orthogonal

channels to provide additional resources to the sensor network that experiences the funnel effect.

7

Problem Statement

Now that we have an idea of what to do, the main question is:

How do we deploy such a network with the minimum number of relay nodes that

mitigate the funnel effect? Our contribution

A placement algorithm Minimum number of relay nodes

8

Network Model

A standard 2-D wireless network n nodes A unidirectional graph G(V,E)

V is the set of vertices (nodes) E is the set of edges (links)

A subset of available channels Cu for each node u from the total set C.

A transmission range r Eucledian distance d(u,v) between two nodes u and v Edge e = (u,v) belongs to E if

d(u,v) < r (the two nodes hear each other) Intersection of Cu and Cv is not empty (there is at least one

common channel)

9

Assumptions

A node u is able to tune its interfaces to any of its channels c in Cu

A sensor has only one interface that is tuned to a common channel c

A relay node has at least two interfaces that may be tuned to any channel in C

Static assignment or some channel assignment algorithm Relay nodes may be potentially connected to several

networks at the same time. SDR, AODV routing for within the intra-net; static routes

for inter-nets. A decision is made at a relay node to accept traffic or

not (policy based through route announcement).

10

Funnel Effect Mitigation – Congested Region Congested Region

A group of nodes characterized by their proximity that experience a high demand of their resources.

A node is considered congested if its local statistics surpass some threshold ETD, packet drops, jitter, etc.

A message exchange protocol with high priority on control packets determine this region and forms a cluster with a well-defined boundary (its Convex Hull)

11

Funnel Effect Mitigation – Basic Intuition Two remote sensor nodes, S1 and S2 may

be connected by a relay node R provided that it is placed within the intersection of the sensors’ ranges:

r < d(S1,S2) ≤ 2r

12

FEM – Relay Placement

Given a congested region and our basic intuition, cover the region with relays

In a circular-manner, cover the convex hull until all the edge nodes are connected to at least one relay node

Repeat with the inner ring of relay nodes

Stop when the inner-most ring is connected to the sink.

13

Placement Condition

To approach the minimal number of relay nodes, we use the placement condition:

The geographical location of a new relay node is that which covers the largest amount number of elements and it is closest to the sink

14

Placement Condition

Node proximity:The minimum allowed proximity distance for two nodes to be covered by one relay is r√(3)

This guarantees that a relay node is placed at a location at least r/2 units closer to the sink from the midpoint of the segment of two nodes being covered

extent is the parameter that controls up to what node proximity the new relay will cover.

15

Node Extent

d(S1,S2) ≤ r × extent < 2r

16

Placement Theory

Theorem 1If the point c (slide 18) lies within the

placement area, then it is the only geographical location that satisfies the

placement condition. Corollary 1

If the point c lies outside the placement area, then the closest point a or b to c

has the best placement.

17

Three Scenarios (First)

18

Three Scenarios (Second)

19

Three Scenarios (Third)

20

The Placement Algorithm Input: Congested region C, sink, extent Output: List R of relay node placements

R = 0 if all elements in C are covered, return R C’ = Convex Hull(C) Sort C’ so we can visit each element in clock-wise order (make a ring) for each element e in C’ not covered:

if e reaches the sink, mark it as covered and continue the loop find the simple best placement p for e and mark it as covered for each e’ in C’ after e that is not covered:

if d(e,e’) > r x extent, break this loop find the best placement q for the triangle (e,e’,sink) and mark

e’ as covered R = R U Closest(sink,{p,q})

return Algorithm1(R, extent, sink)

21

The Placement Algorithm

22

Algorithm Analysis

The complexity time to calculate the Convex Hull of a set C of size m is O(mlog(h)), for h being the number of elements in the hull.

Sorting C’ takes O(hlog(h)) The nested loops visit every single element in the hull (h)

exactly one time, so it’s linear Since m>h, the dominating factor is the repeated

calculations of the convex hull Due to the placement condition, for every recursive call,

the algorithm advances at least r/2 units closer into the sink. rC’ is the radius of first Convex Hull. There are rC’/r/2 =

2rC’/r number of recursive calls (constant) So the overall complexity is dominated by O(mlog(h))

23

Experiment Settings

Simulation QualNet 5.0 1000m by 1200m flat terrain with

Rayleigh fading model (forrest, debris) 40 sensors, 1 sink, 7 feeding sensors WiFi 2Mbps Tx rate, 200m Tx range, 32-

Byte packet size at 4/sec (local traffic) Foreign sensors simulate the rest of the

network by generating packets of size 1KB at a rate of 2/sec.

24

Metrics Observed

Throughput observed per node E2E Delay observed per node Jitter of the data flow

inter-packet arrival gap of two consecutive packets sent from the same source.

25

Naïve Placement Strategy

26

Results (Number of Relays)

Placement Constraint:

Our strategy tries to maximize the coverage of nodes by choosing the location that covers the most elements in the convex hull AND is the closest to the sink.

27

Results (Avg. Throughput)

Funnel Effect.

Without relays, we observe little throughput, mostly from the nodes inside the Congested Region. As a consequence, the foreign nodes starve.

28

Results (Avg. E2E Delay)

With relays the network stabilizes.

29

Results (Jitter Observed)

The relay network improves the availability of data due to the addition of new path resources.

30

Conclusions

Placement condition guarantees minimum number of relay nodes used 43% less relays used

O(mlog(h)) algorithm Improves observed throughput and delivery

ratio It stabilized the transmission delay

Less oscillations of data flows Decreases the jitter, making packets more

readily available

31