Energy-based Metric for the Routing Protocol in Low-power and Lossy

Network

Patrick Olivier Kamgueu

, Emmanuel Nataf

, Thomas Djotio

and Olivier Festor

University of Yaounde I, LIRIMA - Masecness Project, Yaound´e, Cameroon

Universit´e de Lorraine, INRIA - Madynes Project, Nancy, France

Keywords:

Routing, Energy Efﬁciency, RPL, Wireless Sensor Network.

Abstract:

Saving power while ensuring acceptable service levels is a major concern in wireless sensor networks, since

nodes are usually deployed and not replaced in case of breakdown. Several efforts have recently led to the

standardization of a routing protocol for low power and lossy network. The standard provides various metrics,

which can be used to guide the routing. Most protocol implementations use expected transmission count as

routing metric, thus focus on the link reliability. To our knowledge, there is no protocol implementation that

uses nodes remaining energy for next hop selection. This paper discusses the usage of the latter as a routing

metric for the Routing Protocol in Low power and Lossy Networks (RPL). We design an objective function

for that metric and compared experiments result with the most popular expected transmission count scheme.

1 INTRODUCTION

Wireless Sensor Networks (WSN) consist of up to

hundreds or thousands nodes scattered in an environ-

ment of interest, where nodes and their interconnec-

tion are constrained. Nodes discover their neighbors,

self-organize to build a topology and route sensed

data towards a central point: the sink. To deal with

challenge presented by low power and lossy net-

works, the IETF Roll Working Group has recently

published several standards related to RPL (Winter,

2012; Vasseur et al., 2012).

RPL organizes network as one or more Directed

Acyclic Graph (DAG), each one rooted at a single

point : the DAG root. Topology construction begins

at this node, which periodically sends a Destination

Oriented DAG Information Object (DIO) via link lo-

cal multicast. DIO carries necessary informations to

build the topology, including root unique identiﬁer,

routing metrics, originating router’s depth called rank,

and other network parameters. Nodes in the vicinity

receiving DIOs, join the DAG by selecting their par-

ents (one or more) as next hop upwards to the sink.

Parent selection process is governed by an Objective

Function (OF), which uses routing metrics to select

node’s preferred parent among neighbors. Different

criteria also called routing metrics (Vasseur et al.,

2012) are deﬁned to capture node or link character-

istics on the path for parent selection. They could

be a node attribute: hop count, node residual energy,

or a link attribute : throughput, latency, link quality

level or expected transmission count (ETX). In this

paper, using an online real time battery level estima-

tion model, we design an OF for RPL that used node

remaining energy as metric. The proposed OF is com-

pared against the existing that rely on ETX.

The remainder of the document is organised as

follows. In the next section we describe energy-based

OF characteristics in terms of node battery level es-

timation, path cost and node rank computation. Sec-

tion 3 presents some related work on energy aware

routing. Implementation parameters, simulations and

results are discussed in section 4. Section 5 concludes

our work and discusses future directions.

2 ENERGY-BASED OBJECTIVE

FUNCTION

2.1 Node’ Battery Level Estimation

To predict the lifetime of the node, we use a well-

known battery model proposed by (Rakhmatov and

Vrudhula, 2003). It uses the current consumption dur-

ing each node state and its duration to estimate the

battery remaining energy. The model is very accu-

rate and cannot be implemented on real sensor nodes

145

Kamgueu P., Nataf E., Djotio T. and Festor O..

Energy-based Metric for the Routing Protocol in Low-power and Lossy Network.

DOI: 10.5220/0004313401450148

In Proceedings of the 2nd International Conference on Sensor Networks (SENSORNETS-2013), pages 145-148

ISBN: 978-989-8565-45-7

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

due to its complex computations and the memory size

requirements. (Rahm´e and Fourthy, 2010) have ap-

proximated the latter by simple computations on low

memory to ﬁt into sensor nodes, while maintaining

the original model accuracy. Based on these approx-

imations, we implemented this model on real sensor

nodes, with the possibility to predict their lifetime on-

line (Nataf and Festor, 2013). Following RPL metrics

recommendations(Vasseur et al., 2012), node resid-

ual energy is estimated on a scale of 255 (full) to 0

(empty).

2.2 Energy-based Path Cost

Computation

Path cost is a scalar value representing link or node

characteristics for which it expresses some quality

level along end-to-end path. Its calculation depends

on the metric chosen by the network operator. After a

node has calculated the path cost for all its neighbors

and chose the best parent in regard of the relation or-

der for the selected metric, node updates its metric

container (by computing its path through that parent)

and starts to send its proper DIO. We consider the path

cost PW

from a node i to the sink as the minimum

value between the preferred parent path cost and its

own energy. At the sink node, this cost is set as the

energy value of node. A node selects the neighbor

that advertises the greatest path cost value as parent.

More formally:

= min[max

j∈N

(PW

),E

] (1)

Where N

is the set of node i’s neighbors toward the

sink, and E

represents the energy of node i. For a

given path, the cost value is also the same as the min-

imum node’s energy level encountered on that path,

since this energy is critical for the route lifetime. Net-

work topology shown in ﬁgure 1 depicts the proposed

energy-based path cost. Node 1 is the sink, and is

Figure 1: Path cost and DAG Rank calculation.

main powered, other nodes at a given time are sup-

posed to have residual energy as shown. Dashed lines

and arrows represent neighbor reachability. Consid-

ering the ﬁgure, node 6 receives respectively values

= 200, PW

= 210, PW

= 212, and PW

= 235.

So node 6 selects 4, as best parent. Similarly, other

nodes select their next hop to the sink.

2.3 Energy-based DAG Rank

To avoid cycle in the network, every node uses a

scalar value: the rank, to record its relative position

to other nodes with regard to DODAG root. Rank

value must monotonically decrease as we move up-

wards to the sink, but it does not necessarily change

as fast as some link or node metrics would. For this

reason, rank value is thought as a ﬁxed-point num-

ber where the position of the radix point between the

integer part and the fractional part is determined by

the

MinHopRankIncr

parameter. When rank is com-

pared for parent relationships or loop detection pur-

pose, only the integer part is used, but OF computes

entire ﬁxed-point value (16-bit). Once a node (say N)

has chosen its preferred parent (P), node computes its

own rank from preferred parent’s rank as deﬁned in

(2) where step = MAX

energy

− Node

energy

Rank(N) = Rank(P) + Rank

incr

with Rank

incr

= step+ MinHopRankIncr

(2)

This formula ensures the monotonicity property of

the rank which increases by at least one point

(

MinHopRankIncr

) between node and its preferred

parent, when child node has a full battery level. The

increment is even greater as node consumes its bat-

tery, because of penalty of step which feeds the frac-

tional part of ﬁx-point rank value. By cumulative ef-

fect of penalties in the node’s parentage, node’s rank

can grow to more than one point as shown in ﬁgure

1 (rank increase between node 5 and 7). Root rank is

set to the same value as

MinHopRankIncr

(256 in this

example).

3 RELATED WORK

Among techniques used to maximize the network life-

time, energy aware routing protocols appear to be

suitable for multi-hop wireless sensor networks, since

they explicitly take into account node residual en-

ergy for route establishment. EESP (Shivaprakasha

and Kulkarni, 2011) follows shortest path algorithm

by combining distance and node residual energy as

cost. Similarly, (Mohajerzadeh and Yaghmaee, 2009)

proposed an algorithm which considers both energy

and delay metric to ﬁnd an optimal path. (Chiang

et al., 2007) proposed Minimum Hop (MH) routing

SENSORNETS2013-2ndInternationalConferenceonSensorNetworks

146

protocol which organize routing topology based on

nodes hop counts and battery power levels. For a

given node, neighbors are classiﬁed into three cate-

gories: parent, sibling and child node, on the basis

of their vicinity in hop count to the sink, respectively

one less, same, and one more than that of the send-

ing node. MH ﬁrst tries to reach sink by path through

a parent node, which guarantee a min hop path. In

case of more than one parent, the protocol uses the

one with the highest energy level. If there is no parent

node available, the sender forwards data through the

sibling node with the highest energy level. MH uses

a local (parent or sibling) energy view of the sender

for next hop selection, and does not always reﬂect the

real energy distribution of node in the path. On con-

trary, (El-Semary and Azim, 2010) proposed the path

energy weight protocol that improve MH by using an

energy-weightedfunction, to indicate how balanced is

the energy among all nodes along a given path. This

path weight takes into account all nodes energy along

the path, although greatly disadvantaging lower en-

ergy nodes. In our approach only the lowest node

energy constrained the path. All previous protocols

improve the average energy consumption in the net-

work compared to a solely hop count-based protocol.

Their major difference is that, they use the later as

main criterion for next hop selection, node’s energy

level is usually used to break the tie. We do present

another scheme, namely one that favors node energy

on a routing standard for WSN, while using the rank

notion to avoid routing loops.

4 SIMULATIONS AND RESULTS

4.1 Environment Setup

Experiments were carried out using Cooja simulator

(Osterlind and Dunkels, 2006). Network topology is a

300×300m

2D-grid of 20 sensors, the sink is located

at the upper left corner. Each sensor node acts in a

120m maximum transmission range with 140m inter-

ference range, and periodically sends data to the sink

using UDP as the transportlayer with a Tx/Rx success

ratio of 80%. The layer 2 medium access control is

ContikiMAC (Dunkels, 2011) that provides power ef-

ﬁciency by the node keeping their radios turned offfor

roughly 99% of the time. All nodes have full battery

charge at the beginningof the simulations, with an ini-

tial power level set to 880mAh. The hardware charac-

teristics for the simulation computer are 3.2Ghz Dual

Core Intel XEON processor board, with 8GiB Mem-

ory size, on Ubuntu 11.10 operating system.

4.2 Results

Simulations were performed for one month network

activities (corresponding to 13 real days on our sim-

ulation computer). We deﬁne the network lifetime

as the date on which the ﬁrst node has completely

exhausted its battery (Dietrich and Dressler, 2009).

The energy aware RPL implementation was com-

pared against the ETX implementation. For both,

the sink collects data generated at various through-

put expressed as the number of application packets

per minute (pkts/min), each having 87 bytes of size.

Then, we evaluate energy depletion and packet deliv-

ery ratio for both scenarios, one at 1pkt/min, the other

at 6pkts/min.

4.2.1 Remaining Power Distribution

Energy aware routing aims to use nodes with higher

remaining power level, thus these nodes drain their

battery more quickly and further become less attrac-

tive to relay data. The network should be reorganized

to ﬁnd more interesting nodes for routing and so on,

thereby a balancing on all nodes battery levels should

occur. This can be seen in ﬁgure 2 which presents

the proportion of nodes in the network with the cor-

responding percentage of remaining energy at the end

of the simulation. In ﬁgure 2a at 1pkt/min, 85% of

nodes have the power level between the range 54%

to 56%, whereas the ETX-based routing spread the

energy distribution unequally among the nodes. At

a higher rate (6pkts/min) in ﬁgure 2b, this observa-

tion is much more pronounced, since the trafﬁc ﬂow is

more important and nodes exhaust their battery much

faster. At the same time, in both illustrations the

ETX-based scheme presents much less-power nodes

(around 20%) than the energy aware scheme, the lat-

ter delaying the ﬁrst nodes that will completely ex-

haust their battery and the possibility to create net-

work holes. This is an important point, because the

network integrity can be affected when some nodes

are stopped. We estimated by a linear regression when

ﬁrst nodes drain completely their energy. Computa-

tions indicate a network lifetime of 35 days for ETX-

based RPL, while 40 days for energy aware scheme,

thus the increase in network lifetime is around 14%.

4.2.2 Transmission Accuracy

We also evaluated the accuracy of routing to collect

the application data. The table 1 highlights the to-

tal number of received packets at the sink for both

rates. ETX-based routing promotesroutes with higher

packets delivery ratio, while energy aware routing

don’t care on that. It is therefore not surprising that

Energy-basedMetricfortheRoutingProtocolinLow-powerandLossyNetwork

147

51 52 53 54 55 56 57 58 59 60

Nodes (%)

Remaining Energy (%)

Throughput: 1pkt/min

ETX

Energy

22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51

Nodes (%)

Remaining Energy (%)

Throughput: 6pkts/min

ETX

Energy

(a) (b)

Figure 2: Nodes Remaining Power Distribution.

the number of received packets with ETX is slightly

greater than energy aware scheme as outlined by the

table, but this delivery ratio difference between these

two schemes is minimal (only around 3%).

Table 1: Transmission Accuracy.

Throughput Sent Pkt Received (ETX) Received (Energy)

6 pkts/min 4488635 4390282 (97.80%) 4251962 (94.72%)

1 pkt/min 748105 735680 (98.34%) 722394 (96.56%)

5 CONCLUSIONS AND FUTURE

WORKS

In this paper, we presented an instantiation and im-

plementation of the routing protocol for low power

and lossy network that uses the node’s remaining en-

ergy as the main routing metric. The implementation

makes use of a well-known battery theoretical model

from which we estimate at runtime the node battery

lifetime for routing. Experiments reveal that, com-

pared to the popular RPL ETX-based scheme, the pro-

posed implementation increase the network lifetime

and distributes energy evenly among nodes without

an appreciable lack of the transmission accuracy.

Our future works aims to combinate these both

metrics (energy and ETX), in accordance with (Za-

hariadis and Trakadas, 2012). We expect to leverage

the strengths of each, and obtain a better compromise.

Furthermore we seek to provide additional decision

criteria in order to better guide the routing decisions

in WSN.

REFERENCES

Chiang, S., Huang, C., and Chang, K. (2007). A minimum

hop routing protocol for home security systems using

wsn. IEEE Trans. on Cons. Elec., 53:1483 – 1489.

Dietrich, I. and Dressler, F. (2009). On the lifetime of wsn.

ACM: Trans. on Sensor Networks, 5.

Dunkels, A. (2011). The contikimac radio duty cycling pro-

tocol. Technical Report ISBN 1100-3154, SICS.

El-Semary, A. M. and Azim, M. M. A. (2010). Path energy

weight: A global energy-aware routing protocol for

wireless sensor network. In Proc. of IFIP WD, Venice.

Mohajerzadeh, A. and Yaghmaee, M. (2009). An efﬁcient

energy aware routing protocol for real time trafﬁc in

wsn. In Proc. of ICUMT, pages 1 – 9, St-P, Russia.

Nataf, E. and Festor, O. (2013). Online estimation of battery

lifetime for wireless sensors network. In Proc. of 2nd

Sensornets, Barcelona, Spain.

Osterlind, F. and Dunkels, A. (2006). Cross-level sensor

network simulation with cooja. In Proc. of 31st IEEE

Conf. SenseApp, pages 641 – 648, Tampa, Florida.

Rahm´e, J. and Fourthy, N. (2010). Energy management for

battery-powered embedded systems. In Proc. of IEEE

WCNC, pages 277 – 324, Sydney.

Rakhmatov, D. and Vrudhula, S. (2003). Energy manage-

ment for battery-powered embedded systems. ACM

trans. on Embed. Comput. Syst, pages 277 – 324.

Shivaprakasha, K. S. and Kulkarni, M. (2011). Energy efﬁ-

cient shortest path routing protocol for wsn. In Proc.

of 3rd CICSyN, pages 333 – 337, Bali, Indonesia.

Vasseur, J., Kim, M., Pister, K., Dejean, N., and Barthel, D.

(2012). Routing metrics used for path calculation in

low-power and lossy networks. Number 6551 in RFC.

Winter, T. (2012). Rpl: Ipv6 routing protocol for low-power

and lossy networks. Number 6550 in RFC.

Zahariadis, T. and Trakadas, P. (2012). Design guidelines

for routing metrics composition in lln. IETF Work in

progress.

SENSORNETS2013-2ndInternationalConferenceonSensorNetworks

148