1 
 
 
 
 
 
 
Reliability, Scalability and Security in Smart Utility Networks 
 
by 
 
Gopalakrishnan B Iyer 
 
 
 
 
A dissertation submitted to the Graduate Faculty of 
Auburn University 
in partial fulfillment of the 
requirements for the Degree of 
Doctor of Philosophy 
 
Auburn, Alabama 
May 4, 2014 
 
 
 
 
Keywords: Wireless mesh networks, Smart utility networks, Performance analysis, Directed 
acyclic graphs, MMPP, OMNeT++  
 
 
Copyright 2014 by Gopalakrishnan B Iyer 
 
 
Approved by 
 
Prathima Agrawal, Chair, Ginn Distinguished Professor, Electrical and Computer Engineering 
Shiwen Mao, McWane Associate Professor, Electrical and Computer Engineering 
Thaddeus Roppel, Associate Professor, Electrical and Computer Engineering  
Alvin Lim, Associate Professor, Computer Science and Software Engineering 
 
 
 
 
 
 
 
ii 
 
 
 
 
 
 
Abstract 
 
 With the rapid increase in deployment of smart meters across North America, resource 
consumption data of utilities such as water, gas and electricity is being collected at a much higher 
granular level. The huge amount of data is being used to make demand-response applications 
smarter. A large part of this deployment of smart utility networks and Advanced Metering 
Infrastructure (AMI) is being done using wireless mesh architecture due to the low deployment 
costs offered by this method. Although wireless environment parameters such as fading and path 
loss differ widely from home, outdoor to industrial, in-building scenarios, efficient protocols at the 
Medium Access Control (MAC) and Physical (PHY) layers and lower layer agnostic data routing 
protocols can be employed to overcome these challenges. For such routing protocols to be 
applicable to smart utility networks, reliability, scalability and security are vital metrics which will 
determine their performance in such networks.  
Our objective in this dissertation is to present a Reliability, Scalability and Security (RSS-
SUN) suite of recommended protocols most applicable for Smart Utility Networks. We present 
the performance analysis of wireless mesh routing protocols most applicable to smart utility 
networks.  New and emerging routing protocols being proposed as an alternate standard by the 
Internet Engineering Task Force (IETF) community have been chosen for performance 
comparison. Herein, we analyze the reliability of three routing protocols, viz. RPL (IPv6 Routing 
Protocol for Low power and Lossy Networks), LOAD (6LoWPAN Ad Hoc On-Demand Distance 
Vector Routing) and a proprietary flavor of Geographical Routing developed by Landis+Gyr.  
Routing metrics of packet success probability, end-end delay, hop count, link quality and 
packet delivery ratio have been considered for this performance analysis. Further, realistic network 
topologies, obtained from actual smart meter network deployments in North America have been 
modeled in simulations to derive at results pertinent to the applicability of these wireless mesh 
routing protocols to Advanced Metering Infrastructure comprising of smart utility networks.  
iii 
 
With respect to scalability, we address the issue of scalable routing protocols being vital to 
its successful deployment in smart utility networks. To that effect, we present two approaches to 
modeling the wireless mesh network with the goal of analyzing the scalability of routing protocols 
in such networks with respect to large scale deployment.  
Approach I models the network as being connected with a Poisson distribution with density 
and transmission range as parameters. We quantify an upper bound to the network size at a 
maximum total expected successful packet transmission as determined by the packet success 
probability of each link. We validate these results with simulation data from a large scale network 
using the supercomputer infrastructure at the Alabama Supercomputing Authority located in 
Huntsville, AL. 
In Approach II, we model the wireless mesh network traffic arrival process as a Markov 
Modulated Poisson Process (MMPP) with two distinct modes. Further, an MMPP(2)/M/1/N 
queuing model is analyzed with the same goal of finding a network size upper bound, such that 
stability is maintained in the network. Verification of the model with analytical and simulation 
data is presented with conclusive scalability bounds affecting performance 
With respect to Security, we present the urgent need to implement state-of-the-art 
cryptographic schemes to devices in smart utility networks. Further, we entail the security risks 
and possible mitigation practices. Two popular industry adopted public-key cryptographic 
schemes of RSA (Rivest-Shamir-Adleman) and Elliptic Curve Cryptography (ECC) are compared 
for performance on the Tmote-Sky hardware platform. We present the impact and trade-offs 
involved in implementing security, an indispensable requirement in smart utility networks.  
 
        
 
 
 
 
 
 
 
 
 
 
 
 
iv 
 
 
 
 
 
 
Acknowledgments 
 
 
There are several people in Auburn that I owe a debt of gratitude for assisting me in making 
my doctoral dreams a reality. I would like to express my sincere gratitude to my advisor, Dr. 
Prathima Agrawal, for convincing me to take the Ph.D. qualifying exams at the end of my Master?s 
program here at Auburn. She has been a constant source of encouragement during the highs and 
lows of this program. Without her support, this research work and dissertation would not have 
been possible.  
I would like to thank Dr. Shiwen Mao, Dr. Thaddeus Roppel and Dr. Alvin Lim for serving 
as members on my advisory committee. They have been kind, patient and provided guidance, 
corrections and suggestions to keep me steering the right way.  I would also like to thank Dr. Saad 
Biaz for agreeing to be the external reader of this dissertation. 
I would like to acknowledge the efforts and support of Ms. Shelia Collis, Ms. Jo Ann 
Loden, Ms. Penny Christopher and Ms. Jennifer Jackson for helping me with paper work related 
to graduate studies, immigration and travel matters.  
My thanks go out to my present and former colleagues in Broun 405 for keeping the office 
lively and engaging in cerebral discussions. I would like to specially thank the group of Santosh 
Kulkarni, Vijay Sheshadri, Hui Zhou, Vignesh Sivakumar and Saketh Reddy for their excellent 
company. My thanks to former colleagues Yogesh Kondareddy, Nida Bano, Nirmal Andrews, 
Santosh Kulkarni, Pratap Prasad, Suryakant Bhandare and Mithun Raghav for making my time in 
Auburn filled with fun.  
v 
 
 A special thanks to my friends Neeraj Naik, Vikas Iyer, Disha Vira, Roshan Pandey and 
Gurudutt Telang who have been there for me, both during the most testing times and enjoyable 
ones. 
Above all, I am grateful to my parents Chitra and Balasubramanian, my sister Ranjani 
Kumar and my brother in-law Kartik Kumar for their encouragement and motivating talks. They 
have been a constant reminder of my purpose during these long years. Without their moral and 
emotional support, this research work and dissertation would not have been possible. I lovingly 
dedicate this work to them.  
This work was supported by the NSF Wireless Internet Center for Advanced Technology 
(WICAT), the NSF Broadband Wireless Access and Applications Center (BWAC) and the 
Wireless Engineering Research and Education Center (WEREC) at Auburn University. We also 
acknowledge the support of Landis+Gyr, Alpharetta, GA as industry members of WICAT and 
BWAC.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
vi 
 
 
 
 
 
 
Table of Contents 
 
 
Abstract ............................................................................................................................... ii 
Acknowledgements? ........................................................................................................ iv 
List of Tables ................................................................................................................... viii 
List of Figures? ................................................................................................................ ix 
Chapter 1   Introduction .......................................................................................................1 
Chapter 2   Reliability of routing protocols in Smart Utility Networks...............................5 
              2.1   Radio and Wireless Environment ..................................................................6 
              2.2   Real Network Topology .................................................................................9 
              2.3   Routing Protocols.........................................................................................10 
              2.4   Performance Analysis ..................................................................................17 
              2.5   Summary ......................................................................................................26 
Chapter 3   Scalability of routing protocols in Smart Utility Networks ............................28 
              3.1   Approach I (Poisson distributed Connectivity Model)  ...............................28  
              3.2   Approach II (Markov Modulated Poisson Process Traffic Model) .............35 
3.2.1 Raw data for 2-state MMPP model ...................................................39 
3.2.2 Fitting raw data to MMPP model......................................................42 
3.2.3 LAMBDA algorithm to determine average rate and states N ...........46 
3.2.4 Fitting data to Markov Chain ............................................................50 
3.2.5 Mean queue length ............................................................................50 
vii 
 
  3.3   Packet loss and Blocking Probability ..........................................................51 
  3.4   Interpreting MMPP model results ................................................................54 
  3.5   Summary ......................................................................................................54 
Chapter 4   Cost of Security in Smart Utility Networks  ...................................................56 
              4.1   NIST data on Cyber-Physical Security ........................................................56 
            4.1.1   Security Risks to the Smart Grid ....................................................58 
            4.1.2   Mitigating security risks in the Smart Grid.....................................58 
              4.2   Public-Key Cryptographic Schemes ............................................................59 
            4.2.1   RSA Public-Key Cryptography ......................................................60 
            4.2.2   ECC Public-Key Cryptography ......................................................61 
              4.3   Power consumption and Processing delay of RSA and ECC ......................62 
              4.4   Summary ......................................................................................................65 
Chapter 5   Conclusions and Future Work  ........................................................................66 
 5.1    Summary of Research Work  .......................................................................66 
             5.2    Future Work  ................................................................................................68 
Bibliography  .....................................................................................................................71 
   
 
 
 
 
 
 
 
 
 
 
 
 
 
viii 
 
 
 
 
 
 
List of Tables 
 
 
Table 2.1   IEEE 802.15.4 PHY Parameters  .....................................................................19 
Table 2.2   Wireless Environment Parameters  ..................................................................19 
Table 2.3   Simulation Parameters  ....................................................................................22 
Table 3.1   LAMBDA Algorithm ......................................................................................47 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ix 
 
 
 
 
 
 
List of Figures 
 
 
Figure 1.1   Current and Projected Smart Meter deployment data ......................................2 
Figure 2.1   Smart utility mesh based architecture ..............................................................6 
Figure 2.2   Temporal fading in a wireless channel  ............................................................8 
Figure 2.3   Geographical distribution of a 7500 node network ........................................10 
Figure 2.4   Instance of RPL after DODAG formation .....................................................13 
Figure 2.5   Exchange of LOAD control messages ...........................................................15 
Figure 2.6   Geographical routing mechanism  ..................................................................17 
Figure 2.7   Composite node module in Castalia  ..............................................................18 
Figure 2.8   Packet success rate histogram for validation  .................................................20 
Figure 2.9   Plot of links with PSR greater than 60% in the network  ...............................22 
Figure 2.10   RSSI values Real vs. Simulated over 24 hr. period  .....................................22 
Figure 2.11   Delay histogram  ...........................................................................................23 
Figure 2.12   Hop count histogram  ...................................................................................24 
Figure 2.13   Packet Delivery Ratio  ..................................................................................24 
Figure 2.14   Control messages  .........................................................................................25 
Figure 3.1   Spatial distribution of nodes for Scalability model  .......................................29 
Figure 3.2   E[pf] vs. u .......................................................................................................32 
Figure 3.3   N vs. u for varying r  ......................................................................................33 
Figure 3.4   N vs. u for varying R  .....................................................................................34 
x 
 
Figure 3.5   Calculated and Simulated N and E[pf] vs. u (Scaled)  ...................................35 
Figure 3.6   State transition diagram for MMPP(2)/M/1/N queue  ....................................38 
Figure 3.7   Raw data arrival rates for 20 node network ...................................................40 
Figure 3.8   Raw data arrival rates for 200 node network ..................................................41 
Figure 3.9   Raw data arrival rates for 6000 node network................................................41 
Figure 3.10   Arrival Process at Collector from metering nodes .......................................42 
Figure 3.11   Raw data fitted to MMPP process for a 20 node network ...................................44 
Figure 3.12   Raw data fitted to MMPP process for a 200 node network .................................44 
Figure 3.13   Raw data fitted to MMPP process for a 6000 node network ...............................45 
Figure 3.14   Raw data fitted to MMPP process for a 10000 node network ..............................45 
Figure 3.15   Fitted data showing 2 state MMPP process for 20 node network .........................48 
Figure 3.16   Fitted data showing 2 state MMPP process for 200 node network .......................48 
Figure 3.17   Fitted data showing 2 state MMPP process for 6000 node network .....................49 
Figure 3.18   Fitted data showing 2 state MMPP process for 10000 node network ..................... 49 
Figure 3.19   Mean packet loss rate for two-state MMPP process ....................................52 
Figure 3.20   Mean delay for two-state MMPP process ....................................................53 
Figure 3.21   Blocking Probability g1842g3029with varying network size ......................................53 
Figure 4.1   Power consumption vs. encryption scheme ...................................................63 
Figure 4.2   Time taken by RSA and ECC for signature generation and verification .......64 
Figure 4.3   Power consumed by RSA and ECC for signature operations ........................64 
Figure 5.1   Simulated Network size versus Simulation Execution Time .........................69 
1 
 
 
 
 
 
 
Chapter 1 
Introduction 
 
 Smart meters are devices which measure utility resource consumption, usually over a 
period of time or in units specified by the utility provider. These devices are comparable with 
sensors which are deployed in large scale to measure data that is important in analyzing a particular 
metric. The value in large scale collection of sensor data is greatly impacted by the collection 
method [1]. In the case of legacy utility meters, historically, manual collection of consumption 
data has been the norm followed by management and billing. With advancement in radio frequency 
technology and miniaturization of transceivers, it has been possible to aggregate such sensor data 
remotely for several decades. Large scale deployment of such remote data aggregating systems is 
now possible due to rapidly declining costs and mass manufacturing of such devices in accordance 
with Moore?s law for silicon technology. Advanced Metering Infrastructure (AMI) comprising of 
smart meters, collectors, transmission and distribution devices and automated aggregation systems 
are now being widely deployed as a part of the Smart Grid initiative [2]. Figure 1.1 shows that 
over 36 million smart meter devices have been deployed as of 2012 in the United States alone. 
This is a 33% increase over 2011 deployments and is projected to be at 65 million by 2015 [3]. In 
a majority of deployments, devices in the AMI have wireless interfaces for communication to 
collector devices. In addition, smart utility networks are characterized by harsh wireless 
environment conditions in which they are deployed. Wireless fading and path loss have spatial and 
2 
 
temporal components which require complex models to study performance in such environments. 
The enormous scale of the network also makes scalable solutions an important part of achieving 
efficiency in such networks.  
A typical smart meter network can range from a few thousand metering nodes reporting to 
a single collector to millions of metering end points reporting to the head-end systems. These 
networks can be deployed using a plethora of evolved communication technologies that are 
available today. Physical layer technologies such as wireless cellular, power-line communications, 
fiber-optical networks and wireless mesh networks are under consideration for deploying AMI. 
The most popular technology yet has been low power wireless systems used in the wireless mesh 
architecture for deploying these networks. The wireless mesh architecture provides several 
advantages over the others including flexibility, minimal infrastructure, extensive network 
coverage and low deployment cost. 
 
 
Figure 1.1: Current and Projected Smart Meter deployment data [3]. 
 
3 
 
In addition, the usage of license free spectrum for deploying wireless mesh networks is 
highly lucrative. It is hence evident that an efficiently deployed smart utility network or AMI is 
vital to several key applications at the utility hierarchy. An efficient data routing protocol, which 
is agnostic to lower layer technologies is imperative to the efficiency and reliability of the smart 
utility network in general. In addition, inter-operability requirement is inherent to all devices in 
the AMI and must be the same with a routing protocol used in such a deployment. Three such IPv6 
based routing protocols developed by the IETF (Internet Engineering Task Force) community viz. 
RPL [4] [5], LOAD [6] and LOADng [7] have been studied herein. Additionally, a proprietary 
flavor of Geographical routing protocol developed by Landis+Gyr has been studied for comparison 
[8]. Previous work done in [9] entails performance analysis using the metrics of hop-count and 
end to end delay, comparing RPL and Geographical routing for a real deployed 500 node smart 
utility network. We extend the performance analysis to LOAD routing protocol in smart utility 
networks. In addition, varying network sizes of up to 7500 nodes with a single collector have been 
used herein, allowing a high-level scalability analysis of these routing protocols. 
Due to the enormous scale of deployment, scalability is a vital requirement of protocols 
deployed in smart utility networks. Here, we analyze scalability by modeling the smart utility 
network using two approaches. One approach aims to determine the scalability according to the 
link packet success probability by modeling the node connectivity using a Poisson distribution. In 
this analysis, we arrive at an analytic result showing the dependency of scale on packet success 
probability. Further, we verify the analytic results by using a large scale simulation of such 
networks. A second approach models the network as a single server queue with a Markov 
Modulated Poisson arrival process, exponential service and finite buffer. We present an analytic 
result in which the scale is dependent on the mode of traffic in the network. The scalability 
4 
 
relationships thus derived from both the approaches can be used in different smart utility metering 
scenarios to estimate a viable scale in such networks.  
Security is an inherent requirement in smart utility networks. Since these networks are tied 
to physical systems, a compromise in confidentiality, integrity and availability of control in such 
networks will be far more consequential than in traditional internetworks. Addition of security in 
smart utility networks presents an inevitable trade-off in performance and power consumption. 
The impact of security overhead is analyzed herein. Two popular public-key cryptographic 
schemes, viz. RSA-1024 bit and ECC-160 bit are implemented on the Tmote-Sky hardware 
platform and their power consumption in a test mesh network are compared versus a node with 
no-encryption for their impact on power consumption in smart utility networks.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5 
 
 
 
 
 
 
Chapter 2 
Reliability of routing protocols in Smart Utility Networks 
 
 Smart utility networks can be deployed using various wireless network architectures. These 
networks can be implemented as wireless cellular networks or as a wireless mesh networks. The 
goal is to provide complete connectivity to all metering nodes in the network. High reliability and 
low end to end latency combined with 100% packet success and delivery ratios are the 
requirements of an AMI network. In such networks, the quality of service is of utmost importance. 
Cellular network architectures in general are providers of greater quality of service as more often 
than not, they use licensed radio spectrum for their service and are fairly immune from the effects 
of interference. This type of deployment bears a significant infrastructure cost and complexity of 
deployment. The cellular deployment also requires extensive radio planning in order to optimize 
installation and use of infrastructure. On the other hand, a wireless mesh network does not involve 
use of complex infrastructure. An electric utility AMI network consists of the electric meters 
themselves, equipped with wireless communication capabilities and several collection points. In 
this mesh network, the electricity meters form an ad-hoc self-organizing mesh and communicate 
with the collection points known as collectors, thus requiring minimal deployment effort and 
network management. Wireless mesh network deployment is an optimal choice for deploying an 
AMI. Figure 2.1 shows the hierarchical architecture of Smart Utility Networks, in which several 
6 
 
metering devices with routers and collectors in a mesh network form the Neighborhood area 
network and report to the head end systems though backhaul wide area networks.  
 
2.1  Radio and Wireless Environment 
The wireless environment places fundamental limitations on the performance of communication 
networks based on radio frequencies. The path between a pair of transmitter and receiver can vary 
between plain line of sight to tall buildings, forest areas and mountainous terrain. These variations 
in geographical topology have given rise to the challenging problems of signal fading due to path 
loss and fitting of complex signal propagation models based on statistical data.   
 
 
Figure 2.1: Smart Utility Mesh based Architecture [8]. 
 
In addition, the wireless channel is plagued by the ubiquitous problem of interference from a 
variety of sources ranging from cosmic background noise, to the electromagnetic noise generated 
by household microwaves. Even though there is a big corpus of theoretical results and 
experimental measurements for wireless communication in general, they tend to practically affect 
areas with more commercial impact such as cellular communication. 
7 
 
 In Smart utility networks and the AMI, the wireless environment varies from indoor for 
home automation application to outdoor and industrial, highly noisy conditions. One important 
aspect of this wireless channel modeling is to estimate the average path loss between two nodes, 
or in general, two points in space. For smart utility networks, where the separation of nodes is from 
a couple of meters to a hundred meters, the lognormal shadowing model has been shown to give 
accurate estimates for average path loss. Equation 2.1 returns path loss in dB as a function of the 
distance between two nodes. 
 
                                      g1842g1838g4666g1856g4667= 	g1842g1838g4666g1856g2868g4667+	10	g2015logg4672g3031g3031
g3116
g4673+	g1850g3097                               (2.1) 
  
PL (d) is the path loss at distance d, PL (d0) is the known path loss at a reference distance d0, ? 
is the path loss exponent, and X? is a Gaussian zero-mean random variable with standard 
deviation ?. 
 Another very important aspect of the wireless channel is the temporal variation of channel 
quality. This is especially pronounced in rapidly changing environments as those experienced in a 
smart utility network. Figure 2.2 shows a typical profile of channel variation. Notice the sharp 
drops (measured up to -50dB) and that most of the time the channel is below the average path loss 
(0dB). There are several theoretical models to describe temporal variation, taking their names from 
corresponding complex distributions (e.g., Rayleigh, Weibull, Nakagami, Gamma and lognormal) 
[10].  
 The choice of radio for any hardware mainly depends on the factors of power, data-rate 
and operating frequencies. Proprietary radios generally operate in a licensed spectrum for which 
the user is billed. On the other hand, IEEE standards based radios operate in the license free 
8 
 
spectrum allocated for target applications by the FCC. Wireless Local Area Network (WLAN) and 
Worldwide Interoperability for Microwave Access (WiMAX) are the most popular examples 
operating on IEEE 802.11 and IEEE 802.16 standards, respectively.  
 Smart Utility Networks are comprised of low power devices communicating at a low data 
rate. The maximum allowed transmission power for such devices is 30dBm (1000 mW) and can 
communicate at a maximum data rate of 250 kbps. Typically, devices in Smart Utility Network 
operate at 0dBm and a variable data rate between 25 kbps and 250 kbps. The IEEE 802.15.4 
specifications for MAC and PHY are built for such target devices, operating at low power, low 
data rates and in highly lossy environments. Moreover, these devices operate in the license free 
Industrial, Scientific and Medical (ISM) radio bands centered around 900 MHz and 2.4 GHz 
frequency bands. The Zigbee alliance has adopted these standards for all Zigbee enabled devices. 
The IETF working group 6LoWPAN (IPv6 over Low power Wireless Personal Area Networks) 
has adopted this standard for building a wireless embedded internet [11], [12], [13].  
 
Figure 2.2: Temporal fading in a wireless channel [14]. 
 
9 
 
2.2   Real Network Topology 
 In the study of wireless mesh networks, a playground grid size is chosen for the simulation. 
In this playground, nodes are placed at random locations based on a chosen seed or they are 
distributed according to a uniform distribution with chosen mean and standard deviation. Although 
these distributions are good for simulations, they do not represent the way in which smart metering 
nodes are distributed in a real-world network.  
 In order to study routing protocols for smart utility networks, forming a wireless mesh, 
incorporating real world topologies are essential. Trace files containing geographical locations of 
real world deployed smart metering nodes were obtained and used in simulations to arrive at 
simulation results and conclusion herein. Varying network size ranging from 25 nodes to 7500 
metering nodes with the collector located at the center of the deployed area derived from a large 
AMI deployment were used as network topologies for the simulation of data traffic in smart utility 
networks. In order to obtain real world comparable performance metrics from the routing 
protocols, it was necessary to use real deployed network topologies instead of a randomized 
topology.  In general, connectivity and performance are directly related to the network topology 
in which a routing protocol is used. Figure 2.3 shows the real geographical distribution of a 7500 
node network used for the performance analysis. 
 
10 
 
 
Figure 2.3: Geographical distribution of a 7500 node network 
 
2.3   Routing Protocols 
 In order for the smart utility network to function as designed for its application, an efficient 
routing protocol is imperative. A single routing protocol is designed with target applications, 
according to which the routing algorithm optimizes routes and controls network functions. The 
requirements for routing in a smart utility network are diverse, as it has to be compatible for a wide 
range of devices, from home appliances to metering devices to collectors. Routing Protocols 
designed for the smart utility networks have to adhere to the requirements spelled out in 
[RFC5826], [RFC5673], [RFC5548] and [RFC5867] which are home automation routing, 
industrial routing, urban routing environments and in-building routing requirements respectively. 
 The nodes participating in these networks are low power devices and have severe 
constraints in terms of CPU power, memory, battery and operating environments. Utility meters 
11 
 
are designed to last for approximately 20 years on battery. Due to these constraints, restrictions 
are placed on routing requirements. For Home automation applications, the duty cycle of devices 
is less than 1% and the end to end delay requirement for the routing protocol is 500 milliseconds. 
 The routing protocol has to perform device aware routing, and must avoid battery 
constrained devices or devices under strain for frequent routing of packets. For industrial 
automation applications, more focus is on the reliability which has to be as close to 100% as 
possible. Reliability is measured by the number of successful transmissions over number of 
attempts in a given period of time. Also minimal network re-configuration is required when new 
devices enter or leave the network.  
Wireless mesh routing protocols in smart utility networks, viz. RPL, LOAD and 
Geographical routing have been analyzed and compared for performance herein. The following is 
a brief overview of each of the routing protocols. 
RPL is based on construction of a DODAG (Destination Oriented Directed Acyclic Graph) 
for the network which has the collector or data aggregating device as the root of the directed acyclic 
graph. It is a distance vector routing protocol for Low Power and Lossy Networks (LLNs). The 
route is constructed with a goal of optimizing an objective-function. The objective function is 
constructed using a combination of metrics and constraints to compute the best path for routing 
data packets to the destination. Thus, there can be multiple DODAGs for the same network 
depending on the objective-function. Applications such as outage reporting may require hard 
deadlines on delivery and can use an objective function to minimize latency as an example. 
Network state information such as connectivity and degree of connectivity of each node to its 
parent node up to the root is maintained and periodically updated using the RPL control plane. The 
root node initiates formation of the graph using the DODAG Information Object (DIO) control 
12 
 
message. Consider node j receiving a DIO (DODAG Information Object) from node i. Upon 
receiving the DIO message the jth node will compute its own rank Rj according to eq. (2.2) and 
determine its position in the DODAG with respect to other nodes. Parameter ? defines the 
granularity of the network topology. It is defined as minimum rank increase and can be adjusted 
as required during deployment. ? varies depending on the network density. For a sparse network, 
? is small and for a dense network ? is large. Figure 2.4 illustrates a RPL instance after formation 
of the DODAG. DIOs are used to form up-routes for MP2P (Multipoint to Point) traffic.  
 
																																																																									g1844g3037 =	g1858g1864g1867g1867g1870	g4672g1844g3036 g2010g3415 g4673                                         (2.2) 
 
RPL also creates down-routes for P2MP (Point to Multipoint) traffic. DAO (Destination 
Advertisement Object) messages are periodically transmitted by the root requesting reverse route 
to a destination node in the tree. DAO messages are used to advertise prefix reach-ability towards 
the leaf nodes in support of the down traffic. RPL forms acyclic graphs and inherently does not 
allow nodes at the same hierarchy level to form an edge. A modification in RPL supports point-
to-point (P2P) traffic instead of data packets traversing all the way up to the root and then down-
route. RPL makes use of the trickle-timer [15] in order to control flooding at the control plane. 
Trickle timer governs the DIO conditions and DAO intervals. Flooding of control packets in smart 
utility networks would be particularly costly where node resources such as energy and computing 
power are scarce and highly constrained. The adaptive trickle-timer is crucial to the efficiency of 
RPL. As the routes form and the nodes are connected (a rank is established) the network stabilizes 
and control plane messages decrease. When an inconsistency arises (such as a loop or a change in 
the DODAG parameters) as local repair is initiated by the node which detected the inconsistency, 
13 
 
the trickle timer is reset and DIOs are exchanged to quickly restore connectivity. In the worst case, 
a global repair is initiated where the graph is built starting at the root. The metric used to make the 
rank change decision is Expected Transmission Count (ETX). It is calculated using the link packet 
success probability. Packet success probability of each link k, denoted pk is randomly assigned for 
each link to a parent. The metric ETX of link k, X (k), is computed as the inverse of the packet 
success probability of each link according to eq. (2.3). 
                                                              g1850g4666g1863g4667=	 g2869g3043
g3286
                                                                 (2.3) 
The initial value of ETX of any link k is set to be 1.0. The ETX is updated and calculated from 
Equation 4.1 from a randomized set of packet success probability values when a DIO is received 
by a node from a parent. Hence, the ETX value of each link keeps changing over time. This 
introduces the temporal nature of link characteristics observed in Low power and Lossy Networks 
(LLNs). The rank is computed based on the updated ETX count and is transmitted in the DIO. 
Figure 2.4 shows an instance of the Directed Acyclic Graph of RPL after DIO control packets 
exchange. Node 1 is the root of the graph with the least rank and node 5 is the leaf node with the 
maximum rank. 
 
 Figure 2.4: Instance of RPL after DODAG formation 
14 
 
LOAD was originally designed to be a light-weight modification of the popular Ad-hoc 
On-demand Distance Vector routing protocol (AODV) for 6lowpan networks [16]. It has been 
simplified by not requiring the intermediate nodes to reply a RREQ (Route Request) control packet 
with the RREP (Route Reply). Instead, only the destination for which the RREQ was intended 
replies to the route-request. In addition, a table maintaining RREQ for each node has been 
eliminated. Also, LOAD does not use the precursor list of AODV in order to simplify the routing 
table structure.  LOAD does not use the destination sequence number as gratuitous RREPs are no 
longer permitted. The accumulated route cost such as Link Quality Indicator (LQI) and the number 
of hops from the source to the destination are the preferred routing metrics in LOAD. A route is 
preferred if the number of weak links along the way is smaller and least number of hops from the 
source to the destination. In LOAD, whenever there is a break in the link, the upstream node of 
the broken link may try to repair the route locally by using the route discovery mechanism. LOAD 
uses periodic HELLO messages, governed by a timer to maintain route state information. The 
process of LOAD control plane message exchange is illustrated in Figure 2.5. Source node S 
initiates a route request destined for D and is forwarded to D by intermediate nodes N1 and N2. 
Destination D replies to the RREQ by sending a route reply (RREP) via the reverse route in the 
RREQ message (i.e. via N2 and N1 to S). After S receives the RREP, data exchange begins. 
Additionally, if the data packet fails at any node, a route error (RERR) message is issued to the 
preceding node. In the LOAD control plane, there is no separate mechanism in comparison to RPL 
for up and down routes. 
15 
 
 
Figure 2.5: Exchange of LOAD control messages 
 
Each source sending RREQ to a particular destination receives a RREP with the reverse route to 
the destination node contained in the message. This reverse route is used to reach the destination. 
Thus, a special hierarchy cannot be established in LOAD wherein the root node, participating 
nodes and leaf nodes, all discover each other by the same mechanism and maintain routes using 
periodic HELLO messages. Discovery and formation of routes is on-demand. This implies that 
routes will not be discovered until there is an application packet in the buffer to be sent to a 
particular destination. 
The proprietary flavor of Geographical routing protocol developed by Landis+Gyr is built 
around P2P traffic exchanges. Nodes poll each other using two mechanisms called TICKLE 
(similar to request) and ACK (acknowledgement of the TICKLE). These control exchanges 
contain geographical information in the form of co-ordinates (latitude, longitude and altitude) of 
the nodes. In addition, they signal if a node is ready to receive a data packet. In order to make 
routing decisions, the nodes are aware of the geographical location of the collector which is the 
upstream destination. The collector is also aware of the geographical location of target nodes for 
downstream communications. The node initiating route discovery builds a scan table to select a 
16 
 
neighbor which is geographically proximate to the destination node and computes its distance 
using eq. (2.4).  
																																					g1856g3037g3036 =	g3495g3435g1876g3037 ?g1876g3036g3439g2870 +g3435g1877g3037 ?g1877g3036g3439g2870 +g3435g1878g3037 ?g1878g3036g3439g2870	                                         (2.4) 
 
Node j receiving the geographical information from node i computes distance dji and stores 
in a table. Sorted according to shortest distance, each node is then polled to transmit data if 
available at buffer. This selection method also considers the dynamically changing link quality in 
a wireless environment and combines geographical proximity with a weighted link quality 
indicator metric in order to select the least cost neighbor. A heuristic approach to routing is 
employed in which active links are strengthened, ensuring stability in the network topology. Figure 
2.6 illustrates a metering source S selecting next hop metering node A to forward data to the 
collector (destination) D by geographical proximity. The altitude or z component is important in 
the realistic perspective, since collectors or specially designated ?Routers? are typically mounted 
on top of electric poles or a higher altitude for better line of sight with meters in a dense radio 
environment. If a designated router is in the scan table, it is selected as a preferred neighbor by 
default due to better connectivity with the collector. This implementation of geographical routing 
allows for a minimal state routing approach and can adapt to the highly temporal characteristics 
wireless medium. It has been deployed in large scale networks while maintaining the same level 
of complexity. To reflect changing link quality in routing decisions, accumulated packet success 
rates are considered after choosing the nearest node for the next hop. 
17 
 
 
Figure 2.6: Geographical routing mechanism 
 
These values are evaluated continuously in every exchange and the link table keeps track 
of the metric for routing decisions. Large scale deployment of nodes with geographical routing 
initially required feeding GPS co-ordinates into the nodes manually. This method has now been 
deprecated and a geographical co-ordinate system based on the node?s link layer address has been 
developed for rapid deployment. 
 
2.4   Performance Analysis 
In order to compare the performance of the wireless mesh routing protocols a simulation 
setup was developed using the discrete event simulator OMNeT++ [17] and the wireless sensor 
network framework Castalia [18]. The 802.15.4 MAC/PHY modules provided with the Castalia 
framework were used as lower layers under the routing layers of RPL and Geographical routing 
developed at a conceptual level for this performance analysis. As Castalia was developed for BANs 
(Body Area Networks) it included temporal channel variation and fading models which were close 
to real world low power and lossy networks. These models made Castalia a right choice over other 
frameworks available for wireless sensor network simulations. Geographical routing protocol, 
18 
 
although proprietary, has been widely deployed in smart utility networks and AMI systems and is 
currently the protocol running in over 2 million metering end-points. 
 
 
Figure 2.7: Composite node module in Castalia 
 
Figure 2.7 shows the architecture of a node in the Castalia framework comprising of an 
application module, communication module (Routing, MAC and PHY) and resource management 
module. Simulation parameters for the IEEE 802.15.4 radio and wireless environment are 
summarized in tables 2.1 and 2.2 respectively.  
 
 
 
 
 
 
 
 
19 
 
Table 2.1: IEEE 802.15.4 PHY parameters 
Parameter Value 
Radio Data Rate 250 kbps 
Modulation Scheme Frequency Shift Keying (FSK) 
Bandwidth 2 MHz 
Receiver Sensitivity -95dBm 
Transmission Power 20dBm 
Operating Frequency 900 MHz 
 
Table 2.2: Wireless environment parameters 
Parameter Value 
Path loss at distance of d0 60dB 
Reference distance d0 1m 
Path Loss Exponent (?) 2.7 
Standard Deviation of Gaussian 
distribution (?) 
4.0 
 
In order to analyze multiple routing protocols, the underlying link layer, physical layer and 
the channel conditions have to be optimized according to the network topology in use and the 
environment in which such networks are deployed. The choice of IEEE 802.15.4 MAC/PHY 
standards for such networks is optimal as it is similar to the future standard 802.15.4g for 
Neighborhood Area Networks. This standard is a likely candidate for deployment in smart utility 
networks and has been adopted by the Zigbee Alliance for home automation networks. With the 
20 
 
aim of quickly validating the link layer model for analysis of the routing protocols, packet success 
probability rates were gathered from a 400 node real network comprising of smart meters and 
network devices. The packet success probability distribution in the network was obtained as shown 
in Figure 2.8. These values represent the real world channel characteristics reflected in the packet 
success probabilities at the link layer.   
 
Figure 2.8: Packet Success Rate Histogram for Validation 
 
After having computed the packet success probability distribution for a 400 node real 
network for comparison, a simulation experiment was conducted to determine the packet success 
probability yielded by the simulator?s link, physical and channel models. This was done for a 100 
node subset derived from a large rurally deployed real network. This experiment set each node to 
broadcast a total of 100 packets in a sequential order. The received packets were counted and the 
source was recorded. The computed ratio of total transmitted packets to those received by each 
21 
 
node was collected as the link packet success probability. The packet success probability values 
obtained from the experiment were used to plot the connectivity map of the nodes with the 
collector. Figure 2.9 shows the topology of the 100 node network used for the experiment and all 
the possible links between nodes. The nodes with links to the collector in figure 2.9, have a packet 
success rate of more than 60%. The values obtained from the experiment with the 100 node 
network and the on-field data from the 400 node network showed a fair correlation in packet 
success probability. This experiment quickly validated the simulation model for IEEE 802.15.4 
MAC/PHY to be yielding link characteristics similar to those obtained from field measurements. 
This validation allowed the use of the IEEE 802.15.4 MAC/PHY simulation models to be used 
under the two routing layers of RPL and Geographical routing for performance analysis. 
In addition to the validation with field data provided by Landis+Gyr, we ran an experiment 
with 9 nodes in a grid topology in both the simulator and on the Tmote-sky hardware platform. 
The Tmote-sky radio is a Texas Instrument CC2420 radio conforming to IEEE 802.15.4 standards. 
The Received Signal Strength Index (RSSI) was obtained over a 24hr period. Figure 2.10 shows 
that the simulation environment is finely tuned to the radio and wireless characteristics of the real 
world. This validates the simulators close emulation of IEEE 802.15.4 radio and the wireless 
channel characteristics used in the simulation parameters. 
 
 
22 
 
 
Figure 2.9: Plot of links with PSR greater than 60% in the network 
 
 
Figure 2.10: RSSI values Real vs. Simulated over 24 hr period. 
 
Table 2.3: Simulation Parameters 
 
 
23 
 
Having validated the simulation model, simulation experiments with varying network size 
from 25 to 7500 were run. The simulation parameters are listed in Table 2.3. The statistics of End-
End delay, hop count and packet delivery ratio were collected from the simulator. The results are 
presented in figures 2.11, 2.12 and 2.13 respectively. The delay and hop count results are for a 
network size of 500 nodes. In addition, statistics from the control plane of RPL, LOAD and 
Geographical routing were collected. The control plane results are presented in figure 2.14. 
 Simulations of RPL, LOAD and Geographical routing protocols were run with the 
simulation parameters for varying network sizes. The traffic scenario simulated was MP2P 
(Multipoint to Point), wherein all meters send periodic data packets to the collector. This traffic 
scenario is the most prevalent in smart utility networks. The application packet rate was set at 1 
packet/second. For a total simulation time of 100 seconds, the application layer generated a total 
of 100 packets. 
 
Figure 2.11: Delay histogram 
24 
 
 
Figure 2.12: Hop Count Histogram 
 
Figure 2.13: Packet Delivery Ratio 
25 
 
 
Figure 2.14: Control Messages 
 
Each node transmitted 100 packets with the collector as the destination. The statistics of 
hop count, delay and packet delivery ratio were gathered for the entire network. At each node, the 
total number of control packets corresponding to each routing protocol was measured.  
Figure 2.11 shows the delay histogram for the 500 node real network for RPL, LOAD and 
Geographical routing protocols. Overall the packets recorded the least delay with RPL. The data 
shows that on an average, packets suffer 160msecs, 173msecs and 339msecs of delay for RPL, 
Geographical routing and LOAD respectively. Reliability, in terms of packet delivery ratio was 
calculated as the ratio of total packets received over total packets transmitted at the root node.  
Figure 2.12 shows the hop count histogram for a 500 node intermediate network size. A 
maximum of 7 hops was observed for each of the three routing protocols, RPL, LOAD and 
Geographical routing. RPL constructs a route in which 190 nodes are 1 hop from the root. In 
26 
 
contrast 110 nodes are 5 hops away from the collector as constructed by LOAD. 175 nodes are 3 
hops away from the collector as constructed by Geographical routing. Among the three RPL places 
maximum number of nodes 1 hop away from the root which is a good metric of its performance. 
Overall, RPL has an average hop-count of 2.3 hops, LOAD of 3.9 hops and Geographical routing 
has an average of 2.6 hops.  
Figure 2.13 shows the packet delivery ratio (PDR) comparison of RPL, Geographical 
routing and LOAD.  RPL and Geographical routing perform consistently between 95%-98% for 
all network sizes. LOAD has the lowest PDR of 58% for the largest network size. It performs 
reasonably well for smaller networks in terms of reliability. Figure 2.14 shows the control packet 
comparison of the three routing protocols. RPL shows a linear increase in the control packets with 
network size while both LOAD and Geographical routing have an exponential increase trend. 
LOAD generates a significantly more number of control packets throughout all network sizes in 
comparison to RPL and Geographical routing. These results show that RPL performs better in 
terms of delay and hop count. It is also the most reliable routing protocol among the three, 
providing high packet delivery ratios consistently across network sizes. RPL shows a linear 
increase in control messages with network size while LOAD and Geographical routing show an 
exponential increase. This is suspected to be the cause of the increased end-end delay and lower 
reliability statistics for LOAD and Geographical routing. 
 
2.5   Summary 
 In this chapter, we described the wireless channel conditions in which smart utility 
networks are typically deployed. We modeled the path loss and fading in the wireless channel and 
introduced a temporal component to model a dynamic wireless channel. We also described the 
27 
 
wireless mesh architecture in which smart meters are typically deployed. Further, we describe the 
routing protocol mechanisms of RPL, LOAD and Geographical routing and their implementation 
in the simulation framework of OMNeT++ for performance analysis. The simulation parameters 
were provided and the results from the simulation of a large scale smart utility network were 
presented. The results show RPL performing better than Geographical routing and LOAD in terms 
of hop count, delay, packet delivery ratio and low control overhead. Thus, we choose RPL as the 
most applicable routing protocol to be deployed in smart utility networks under the conditions 
simulated.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28 
 
 
 
 
 
 
Chapter 3 
Scalability of routing protocols in Smart Utility Networks 
 
 Scalability of routing protocols is an important problem in wireless mesh networks [19] 
[20]. The large scale deployment of smart utility networks poses challenges in determining the 
maximum capacity and scale which each collector or root node can support [21]. Very little field 
deployment data is available on the maximum number of nodes supported by each collector. In 
this work, we attempt to statistically model the smart utility network in order to determine a 
theoretical limit on the maximum number of nodes supported by each collector. In addition, we 
verify these results with extensive simulations of large scale networks. Here, we use two distinct 
approaches to model the smart utility network. One is based on the idea that the network size is 
limited by the packet success probability of each link in the mesh network. The other analyzes the 
mesh network as a single server queuing system in which the packet arrival process at the collector 
is according to a Markov Modulated Poisson Process, service is exponential and the buffer size is 
limited.  
 
3.1 Approach I (Poisson distributed Connectivity Model) 
 This approach models the smart utility network as spatially distributed according to a 
Poisson distribution [22]. Figure 3.1 shows the smart utility network comprised of a collector at 
the center of a geographical area of radius R. Each node in the network has coverage range of 
29 
 
radius r. This radius r is derived from eq. (2.1) from Chapter 2 section 2.1. The transmission radius 
is dependent on the radio transmission power and the wireless channel characteristics. Since the 
wireless channel model has been validated by comparing with real world experiments in Chapter 
2, the r thus calculated provides realistic values for the transmission range of each node in the 
smart utility network. The total number of nodes in the network is N. Typically, the collector node 
has a higher transmission power than the participating nodes in real deployments. For the sake of 
simplicity in this analysis, we assume that all the nodes including the collector transmit at the same 
power, thus having the same transmission range r. In addition we define the density parameter ? 
calculated by eq. (3.1). The parameter ?d is the deployment density and can vary between densely 
and sparsely deployed smart utility networks. The nodes are connected to each other and the 
collector at degree k by a Poisson distribution whose parameter is nf. Here, nf is the number of 
forwarding neighbors available to each node at the kth hop.  
 
 
Figure 3.1: Spatial distribution of nodes for Scalability model 
 
30 
 
Thus the probability of finding a node at kth ring from the collector is given by eq. (3.2). 
Consequently, the probability that a node is detached from the network is at k=0, given by eq. 
(3.3).  
																																	g2025 =		g3006g4670g3041g3278g3280g3287g3287g4671g3015 =		g3096g3279.g3095	.g3045g3118g3096
g3279.g3095	.g3019g3118
=	g4672g3045g3019g4673g2870                                         (3.1) 
  
  																																g1842	g3435g1863;g2019 = g1866g3033g3439=	g3032
g3127g3289g3281.	g3041
g3281g3286
g3038!                                                       (3.2) 
 
																																g1868g3031g3032g3047g3028g3030g3035 = g1842g3435g1863 = 0;	g1866g3033g3439= g1857g2879g3041g3281																																		(3.3) 
 
Let the packet success probability for each link be u. The probability of a successful 
transmission over n hops is given by eq. (3.4). The expectation of successful transmission of 
packets over k hops according to the connectivity model in eq. (3.1) is given by eq. (3.5). Here, 
the parameter ncell is the number of forwarding neighbors found in the ring of the kth hop.  
																																																														g1868g3033 =	g1873g3041                                                        (3.4) 
 
                                   																	g1831g4674g1868g1858g4675=	? g1857?g1866g1855g1857g1864g1864.	g1866g1855g1857g1864g1864g1863g1863!g1863=g1866g1863=1 	g4666g1873g1863g4667                                   (3.5) 
 
g1866g3030g3032g3039g3039 = 	g2025g1840 
 
 
31 
 
Limit of E[pf] as  	g1866 ?	?  yields, 
 
																																						g1831g3427g1868g3033g3431	?	g1857g2879g3096g3015	g4666g1857g3048g3096g3015 ?	1g4667                                           (3.6) 
 
Let the differential of E[pf] w.r.t. N be zero to find maximum; 
 
g1856g1831g3427g1868g3033g3431
g1856g1840 = 0 
 
																																															g2025g1840 =	g2869g3048	.g1864g1866g4672 g2869g2869g2879g3048g4673																		                               (3.7) 
 
As n ? 	? , the expectation of successful packet forwarding can be approximated to eq. 
(3.6). To find the number of nodes N at which we get the maximum expected successful packet 
transmission, we let the differential of E[pf] with respect to N be zero. It can be easily verified that 
a maximum exists as the second order differential of E[pf] with respect to N is negative. 
Substituting the result of eq. (3.7) in eq. (3.6) for the maximum expected successful packet 
transmissions yields eq. (3.8). Here, we have a relationship between the maximum expected 
successful packet transmissions and the link packet success probability in the network for a 
network of N nodes. Figure 3.2 shows the relationship of E[pf] and u. 
 
																																			g1831g3427g1868g3033g3431 =	g1857g2879g4674
g3117
g3296	.g3039g3041g4672
g3117
g3117g3127g3296g4673g4675	g3436g1857g4674g3039g3041g4672
g3117
g3117g3127g3296g4673g4675 ?	1g3440                                      (3.8) 
32 
 
 
Figure 3.2: E[pf] vs. u 
 
Figure 3.2 gives a general relationship, allowing for geographical size R and transmission range r 
as variable parameters to determine the value of u at which we can obtain a maximum expected 
successful packet transmission for network size N. Thus we can determine the scalability in smart 
utility networks by simulating different network sizes and determine the average link packet 
success probability yielded by each network size N. Figures 3.3 and 3.4 show the various 
theoretical values of N as a function of the link packet success probability as calculated using eq. 
(3.7). This is the value of N which will yield the maximum E[pf]. In figure 3.3, we vary the 
transmission range r from r=100m to r=450m in steps of 50m and obtain the corresponding values 
of N against the link packet success probability u. In practice, this can be done by varying the 
transmission power of the node. In figure 3.4, we vary the geographical area R, keeping the 
transmission range r constant at r=450m. This yields a new set of values of N against u and by 
33 
 
utilizing this data, we can calculate N for maximum expected successful packet transmissions as 
in figure 3.2. In order to verify these theoretical results, we ran simulation scenarios with 
transmission range set at r=100m and geographical area R=3536m. Average link success 
probabilities (u) were calculated for varying network size using the same radio and wireless 
environments described in Chapter 2 section 2.1.  
 
 
Figure 3.3: N vs. u for varying r 
34 
 
 
Figure 3.4: N vs. u for varying R 
 
Figure 3.5 shows the maximum E[pf] versus u for varying network sizes N both calculated 
and simulated. The values are scaled to fit in a multiple y-axis grid. The graph can be read by 
choosing a point on the black (calculated) or blue (simulated), projecting it on the x-axis for a 
value of u. Further, we can project this value of u onto the red (maximum E[pf]) curve to get a 
maximum expected successful packet transmission for the network size N. The simulated results 
were obtained for a maximum network size of 5000 nodes before terminating. The difficulties and 
challenges in simulating a network of size beyond 5000 nodes for this experiment is outlined in 
chapter 4.  
35 
 
 
Figure 3.5: Calculated and Simulated N and E[pf] vs. u (Scaled) 
 
3.2 Approach II (Markov Modulated Poisson Process Traffic Model) 
 Approach II is based on the idea that, given ideal wireless channel conditions, the scale of 
the smart utility network is limited by the traffic. In this approach, we model the network as a 
Single Server Queue (SSQ) [23], the collector node or the root being the server. The Poisson arrival 
and service process is exponentially distributed. This allows for simplification in calculations due 
to the fact that we can add the departure rates at each node and generate a cumulative arrival 
process at the server which is also exponentially distributed [24]. In smart utility networks, the 
traffic is of two distinct types. One is the metering data which is periodically transmitted to the 
collector as per the requirements of the utility demand-response applications [25]. This can be 
modeled as Constant-Bit-Rate (CBR) traffic. The second type of traffic is the emergency data 
36 
 
during outages and anomalous power fluctuations. This type of traffic takes precedence over the 
metering data. Thus we can model the two types of traffic as two modes of a Markov Modulated 
Poisson arrival Process (MMPP) [26].  
 In our model the MMPP has two modes m0 and m1 with different arrival rates ?0 and ?1. 
The MMPP(2)/M/1/N queue with buffer size N, has mode parameters ?0 and ?1 for modes m0 and 
m1. The server has an exponentially distributed service time ?. The service times are mutually 
independent and independent of the arrival process. Since, the inter-arrival times are not 
independent in this queue, it affects queuing performance, packet loss and utilization. The queue 
process is a continuous-time Markov chain but its states are two dimensional non-scalar vectors. 
Each state has two scalar quantities; the mode m and the queue size. Let pim for i=0, 1, 2? N be 
the probability that the arrival process is in mode m and that there are i packets in the system. By 
obtaining the values of ?im the steady-state queue probabilities can be obtained by using eq. (3.9). 
																																																										g2024i =	g2024i0+	g2024i1								for		i	=	0,1,2,?N                                      (3.9) 
The probability of arrival in mode m can be calculated by solving the set of equations (3.10) and 
(3.11). Solving these equations gives the steady-state probabilities of arriving in mode m as a 
function of mode parameters ?0 and ?1 shown in equations (3.12) and (3.13). 
																																																							g1842g4666g1865 = 0g4667g20120 = g1842g4666g1865 = 1g4667g20121                   (3.10) 
																																																							g1842g4666g1865 = 0g4667+	g1842g4666g1865 = 1g4667= 1                              (3.11) 
 
																																																							g1842g4666g1865 = 0g4667 =	 g30831g3083
0g2878g30831
                                    (3.12) 
 
																																																							g1842g4666g1865 = 1g4667 =	 g30830g3083
0g2878g30831
                     (3.13) 
 
37 
 
The probability of the arrival process to be in mode m, is equal to eq. (3.14). This follows from 
equations (3.12) and (3.13).  
	
    ? g2024img3036g2880g3015g3036g2880g2868 = g30831-mg3083
1-mg2878g3083m
					for	m=0,1                         (3.14) 
 
Thus, the average arrival rate of both the modes is ?av given by eq. (3.15). Also, we let utilization 
of the combined arrival process as ?* and is given by eq. (3.16). 
 
																																					g2019av =g1842g4666g1865 = 0g4667g20190	+g1842g4666g1865 = 1g4667g20191	                                                                     (3.15) 
 
																																											=	 g20121g2012
0+g20121
g20190	+	 g20120g2012
0+g20121
	g20191		 
 
																																														g2025* =	g3090avg3091	                                                                                                                            (3.16) 
 
 
We also define a parameter ? called the mode duration parameter. In general, ?m =  ? ?*m. 
The mode duration parameter ? can be varied according to how bursty the traffic is. For example, 
if we receive frequent outage and emergency information from the meters, the value of ? can be 
large. Lower value of ? indicates that the traffic in each mode is for a long duration or less bursty 
traffic. 
38 
 
 The values of ?im can be calculated using the finite set of steady state equations given by 
eq. (3.17). Q is the infinitesimal generator whose size is 2Nx2N, where N is the size of the buffer. 
Non-zero entries of Q for [i,j] two-dimensional vectors are given herein.  
																																																																0= 	?g1843                                                               (3.17) 
 
Here, the generator matrices Q and ?	are given by equations (3.18) and (3.19). 
 
? = g3427g2024g2868g2868,g2024g2868g2869,g2024g2869g2868,g2024g2869g2869,g2024g2870g2868,g2024g2870g2869,?,g2024g3015g2879g2869,g2869,g2024g3015g2868,g2024g3015g2869g3431                         (3.18)   
 
g2173 =	g3427g1843g3036,g3037g3431	  ; i,j are two-dimensional vectors ; Size of Q is 2Nx2N                                       (3.19) 
 
The non-zero entries of Q for buffer size N>i>0 are as follows: 
 
g1843g18610,g18610 =	?	g20190?g20120??	;      	g1843g18610,g18611 =	g20120 ;   g1843g18610,g4666g1861+1,0g4667 =	g20190 
g1843g18611,g18610 =	g20121 ;    g184301,01 =	?	g20191?g20121?? ;   g1843g18611,g4666g1861+1,1g4667 =	g20191 
 
 
Figure 3.6: State Transition diagram for the modeled MMPP(2)/M/1/N queue 
39 
 
Figure 3.6 gives the state transition diagram for the above modeled MMPP(2)/M/1/N 
system. The normalizing equation (3.20) and the steady-state equation (3.17) need to be solved in 
order to obtain steady state probabilities. We use the Gauss-Seidel method of successive 
approximations or the iterative method to derive these values. The blocking probability, Pb can be 
obtained using equation (3.21).  
 
																																														? ? g2024g3036g3040 = 1g3040g2880g2869g3040g2880g2868g3036g2880g3015g3036g2880g2868                                                           (3.20) 
 
      																																														g1842g3029 =	g3090g3116g3095g3263g3116g2878g3090g3117g3095g3263g3117	g3090
g3276g3297
                                                         (3.21) 
 
Having constructed the model, we aim to simulate the smart utility network as the modeled queuing 
system to calculate the blocking probability for different buffer sizes N and network sizes. The 
goal is to find a maximum network size supported by the collector for the smart utility network 
traffic arrival rate at which blocking probability approaches the value of 1.  
 
3.2.1 Raw data for 2-state MMPP model 
 Using the validated OMNeT++ simulation model described in Chapter 2, packet arrival 
rates are collected for varying network sizes from 20 nodes to 6000 nodes at a single collector 
node. Each network size is simulated for a time of 100 seconds and the number of packets arriving 
at the collector per second is recorded. This will allow us to approximate the packet arrival rate ? 
at the collector for each network size. Subsequently we use the LAMBDA algorithm presented in 
[27] [28] to fit the raw data into the MMPP modeled herein.  
40 
 
 In addition to the fitted arrival rates, the LAMBDA algorithm gives the number of states 
in the MMPP model that the traffic can be classified into. To this effect, we generate two types of 
traffic while generating data. Periodic metering information sent by each node and random outage 
information packets. We estimate this type of traffic in smart utility networks to fit into the two-
state MMPP model. Figures 3.7, 3.8 and 3.9 show the raw data obtained from varying network 
sizes of 20 nodes, 200 nodes and 6000 nodes respectively to indicate the scale of data collected 
form the simulator.  
 
Figure 3.7: Raw data arrival rates for 20 node network 
41 
 
 
Figure 3.8: Raw data arrival rates for 200 node network 
 
Figure 3.9: Raw data arrival rates for 6000 node network 
42 
 
3.2.2 Fitting raw data to MMPP model 
 An important application of queuing models in smart utility networks is to size the number 
of links and buffer sizes at routers. The fundamental part of the queuing model is the arrival 
process. We have collected raw data from the simulation in order to fit the raw data to the modeled 
arrival process. Using this data, we present the case that smart meters generate packet arrivals in 
the form of an MMPP. Moreover, the nature of traffic is closely related to an Interrupted Poisson 
Process (IPP) in which the arrival rate of one type of traffic is abruptly made 0 when the other type 
of traffic is active. This accurately models the traffic generated by smart meters. 
When an emergency or outage information packet arrives, the collector stop processing the 
periodic meter reading traffic and prioritizes the outage packet. The arrivals pass through a 
constant-rate device, or a collector. When the smart meter data generation rate exceeds the rate of 
the collector, we assume that the excess is either buffered or retransmitted after a short delay. Then 
the output from the collector is a modification of the original MMPP with a peak rate equal to the 
rate of the collector. We present a method to capture this effect and develop equations to describe 
this rate-limited process. Fig. 3.10 shows the different arrival rates and switching of processes by 
the collector.  
 
 
 
 
  
 
Figure 3.10: Arrival Process at Collector from metering nodes 
g2019g2868 
g2019g2869 
? 
g2019g3028g3049 
Collector 
43 
 
In [29] the first proposal to fit a traffic stream by an MMPP has been proposed. It used two 
phases for the Markov chain, gave an estimation procedure for the parameters of the MMPP from 
count data, and solved several queuing models with exponential service times. Meier-Hellstern 
[30] provided the first estimation algorithm based on inter-arrival time measurements. It has 
proven difficult to obtain estimators for MMPPs with more than a few states because of the 
computational burden. Moreover, traffic data frequently consists of counts of events during fixed 
length intervals, such as packets per second. Hence, model fitting has to be done with this type of 
data. 
 The IPP is the simplest non-degenerate special case of an MMPP. There are two states, and 
the arrival rate is zero in one of them. In [31], they have developed a maximum likelihood 
estimation algorithm for MMPPs based on the superposition of IPPs. Their algorithm requires that 
the arrival times are given in addition to the counts. Lee et al. [32] describe the properties of a 
generalized IPP in which the times between phase transitions need not be exponential. They focus 
on the autocorrelations in the times between arrival epochs. Since only one positive Poisson rate 
is used, we do not think this model will fit the data we use. 
 Since the variance of the Poisson distribution equals the mean (?), and for large ? the 
Poisson and Normal distribution functions are close, most of  P(?) will be within g2019?g1853?g2019 with 
a=2. When the peak-to-mean ratio of the data set is large, the two rates in states one and two, g2019g2868 
and	g2019g2869, with g2019g2868 g3408g2019g2869 ,will be very different. Then g2019g2869 +2g3493g2019g2869	will be much less than g2019g2869 ?2g3493g2019g2869 
and traffic with rates between these values will not be described. Hence, there is a need to fit the 
raw data into two discrete states representing different rates of arrivals to ensure that all the values 
will be described in the model. Figures 3.11, 3.12, 3.13 and 3.14 show the raw data fitted to the 
MMPP model for varying network sizes of 20 nodes, 200 nodes and 6000 nodes respectively. 
44 
 
 
Figure 3.11: Raw data fitted to MMPP process for a 20 node network 
 
Figure 3.12: Raw data fitted to MMPP process for a 200 node network 
45 
 
 
Figure 3.13: Raw data fitted to MMPP process for a 6000 node network 
 
Figure 3.14: Raw data fitted to MMPP process for a 10000 node network 
46 
 
3.2.3 LAMBDA algorithm to determine average rates g2245g2183g2204 and states N 
 The LAMDA algorithm presented in [26] is originally intended for IP traffic at high data 
rates. The algorithm takes raw data as packet arrivals per second and estimates the average rate 
g2019g3028g3049 and the number of MMPP states that the data fits into. We utilize this algorithm shown in 
Table 3.4.1 to estimate the number of MMPP states that the data from smart metering simulation 
trace can be fit into. We expected the algorithm to generate two distinct states, according to the 
two types of packets at varying arrival rates of  g2019g2868	g1853g1866g1856	g2019g2869. As the starting point of the LAMBDA 
algorithm, we determine the initial rate to be the rate which covers the largest observations. 
Equation (3.22) is used to determine this rate at as the upper bound. Observations larger than the 
initial rate are calculated using Equation (3.23), which is the solution to the quadratic in g2019g2868. The 
lower bound of g2019g2868 is chosen as the upper bound of g2019g2869 in equation 3.24. The solution to the quadratic 
in g2019g2869 is given by equation 3.25. The stopping points are data rates which are 0.  
 
        g585g2868 +2g3493g585g2868 = g1842g1857g1853g1863	g1867g1858	g1856g1853g1872g1853	                                      (3.22) 
 
g585g2868 =	g3435?1+g1842g1857g1853g1863?1g3439g2870       (3.23) 
 
g585g2869 +2g3493g585g2869 =		g585g2868 ?2g3493g585g2868           (3.24) 
 
g585g2869 =	g3435g3493g585g2868 ?2g3439g2870       (3.25) 
 
 
 
47 
 
Table 3.1: LAMBDA algorithm 
Algorithm: LAMBDA 
0: Choose width parameter a [default=2] and ending parameter ? [default=0] 
1: Choose g2019g2869 using Equation (3.22) 
a) Set N=1 
b) Set r = g2019?g1853?g2019 
c) If r ? ?, Stop. Else go to Step 2. 
2: Set j = N+1 
a) Set g2019g3037 =	g3435g3493g2019g3037g2879g2869 ?g1853g3439g2870 
b) Set r = g2019g3037 ?g1853g3493g2019g3037 
c) Set N = j 
d) If r ? ?, Stop. Else repeat Step 2. 
 
48 
 
  
Figure 3.15: Fitted data showing 2 state MMPP process for 20 node network 
 
Figure 3.16: Fitted data showing 2 state MMPP process for 200 node network 
49 
 
 
Figure 3.17: Fitted data showing 2 state MMPP process for 6000 node network 
 
Figure 3.18: Fitted data showing 2 state MMPP process for 10000 node network 
50 
 
Figures 3.14, 3.15, 3.16 and 3.17 show the raw data fitted to the MMPP model with the 
two distinct states for data rates g585g2868	g1853g1866g1856	g585g2869 as computed by the LAMBDA algorithm. 
 
3.2.4 Fitting data to Markov Chain 
 Using the LAMBDA algorithm, we have generated the number of states N to fit the data. 
Here, we have 2 states with rates g585g2868	g1853g1866g1856	g585g2869. We construct the state transition matrix in order to 
get the steady state probabilities in those states. Let g4668g1876g3036	, g1861 =1,2g4669 be the observation states. Let g2016g3036 
be the state associated with g1876g3036. The range of g1876g3036 for which g2016g3036	is defined, can be given by equation 
(3.26) as defined by the upper and lower bounds. 
 
g585g2920 ?2g3493g585g2920 g3407	g1876g3036	g3409	g585g2920 +2g3493g585g2920  ?  g2016g3036 = g1862                              (3.26) 
 
 Let g1842 =	g3435g1868g3036g3037g3439 be the state transition matrix for the process. g1868g3036g3037. The Maximum Likelihood 
Estimate (MLE) [33] [34] of g1868g3036g3037 is given by equation (3.27). 
 
g1868g3036g3037 =	g3015g3048g3040g3029g3032g3045	g3042g3033	g3021g3045g3028g3041g3046g3036g3047g3036g3042g3041g3046	g3033g3045g3042g3040	g3036	g3047g3042	g3037g3015g3048g3040g3029g3032g3045	g3042g3033	g3047g3045g3028g3041g3046g3036g3047g3036g3042g3041g3046	g3042g3048g3047	g3042g3033	g3036                               (3.27) 
 
 
3.2.5 Mean queue length 
 Let g1849g3049 and g1849g3028 be the virtual and waiting time seen by the arrival process respectively. If 
the mean service time ? is 1, then from the results of [35] [36] we have expected waiting times as 
equations (3.28) and (3.29), where g1843g2868 and g1843g2869 are infinitesimal generator matrices for states with 
arrival rates g2019g2868 and g2019g2869 respectively as defined by equation (3.19).  
 
g1831g4666g1849g3049g4667= g4666g885g2025g1499 ?2g1854g1843g2869g1857g4667g512g46662g46661?g2025g1499g4667g4667                                     (3.28) 
 
51 
 
g1831g4666g1849g3028g4667= 1?g3029g3018g3117g3032g3096g1499 +	g1831g4666g1849g3049g4667                       (3.29) 
g1875g1860g1857g1870g1857,g1854 = g3435g46661?g2025g1499g4667g1859+g2024g1843g2869g3439	g4666g1857g2024+	g1843g2868 +	g1843g2869g4667g2879g2869 
 
Here, g1859 is the stationary probability vector of the irreducible stochastic matrix g1833, a unique solution 
to the matrix functional equation (3.30) and ? is the stationary vector of the infinitesimal generator 
function g1843g2868 +	g1843g2869. The starting point for calculating g1833 by successive iteration is     g1833g2868 =g1857g2024.  
g1833	=	g1857g3018g3116g2878g3018g3117g3008                         (3.30) 
Using Little?s formula [37], the mean virtual queue length at mean queue length seen at arrivals is 
given by equations (3.31) and (3.32). 
        g1831g4666g1843g3049g4667=	g2025g1499	g1831g4666g1849g3049g4667        (3.31) 
       g1831g4666g1843g3028g4667=	g2025g1499	g1831g4666g1849g3028g4667                                                           (3.32) 
 
3.3 Packet loss and Blocking 
 Using the Two-state MMPP fitted data and the model described above, we compute the 
mean delay and mean packet loss rate for the arrival process. The packet loss rate is given by 
equation (3.33). Figures 3.17 and 3.18 show the mean packet loss rate and mean delay respectively 
for the MMPP arrival process. Using thus computed g2019g3028g3049 and state transition probabilities for 
arrival rates of g2019g2868 and g2019g2869, we can estimate the blocking probabilities for traffic at varying network 
sizes. Figure 3.19 shows the blocking probabilities as computed using equation (3.21) for the 
arrival process at varying network sizes.  
g1838g3045g4666g1861g4667 =g4666g1868g1863g1872g1871g4666g1861g4667g1499g1868g1863g1872g820g1871g1861g1878g1857g4667?	?    (3.33) 
52 
 
Here, the loss rate at a given sample g1861,g1838g3045 is in bps and is obtained by the packets arriving at 
instant g1861 times the packet size (g1868g1863g1872g820g1871g1861g1878g1857) less the service rate ? of the collector device.  
 
 
Figure 3.19: Mean packet loss rate for two-state MMPP process 
53 
 
       
Figure 3.20: Mean delay for two-state MMPP process 
 
Figure 3.21: Blocking Probability g2172g2184with varying network size 
54 
 
3.4 Interpreting MMPP model results 
We can conclude the following from the results of fitting the raw data obtained from the 
smart utility network simulation data of varying network sizes from 20 nodes to 6000 nodes in 
terms of mean delay, mean packet loss rates and blocking probabilities:  
? With offered loads more than 60%, the packet loss rate increases exponentially for network 
sizes above 2500 nodes. 
? Mean delay increases exponentially with offered loads more than 50% for network sizes 
greater than 2500 nodes. 
? The blocking probability is 0.5 for offered loads of 50% and increases linearly thereafter. For 
network sizes greater than 4000 nodes, the blocking probability approaches 1 at loads greater 
than 80%. 
? In terms of scalability, we can draw a conclusion that deploying smart utility networks of size 
4000 nodes per collector in any traffic conditions (offered load) is detrimental to the data 
acquisition performance of the entire advanced metering infrastructure. 
3.5 Summary 
 In this chapter, we presented two models to address network scalability in smart 
utility networks. Both the models find an upper bound on the number of nodes that can be 
supported by each collector device. Approach I uses the link packet success probability and a 
Poisson connectivity model to measure scalability. We back this approach with simulated data 
from smart utility networks of up to 7500 nodes per collector. Approach II uses the two-state 
MMPP to model traffic in smart utility networks. The motivation stems from the two distinct types 
of traffic with distinct arrival rates in smart utility networks. The regular metering data and the 
random bursty outage information data are modeled well by the two-state MMPP process. We 
55 
 
have observations which are conclusive that as the network size increases, critical performance 
metrics such as delay and packet loss rate increase exponentially and from the data, we can 
estimate the upper bound on the network size at which scalability fails, using the raw data arriving 
at the collector from the smart utility network. 
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
56 
 
 
 
 
 
 
Chapter 4 
Cost of Security in Smart Utility Networks 
 
 A 2009 report stated that cyber-spies had penetrated into the North American power grid 
system and left behind malicious software, that could be triggered by them later-on during a time 
of crisis or war sounded alarm bells for utility companies whose power infrastructure has been 
proven to be vulnerable. Basic facilities such as electricity grid, water, gas, waste water 
management systems, traffic signaling systems etc. will be directly affected by a disruption caused 
to the power grids. Hence the vulnerabilities would imply that the purpose of the utilities serving 
households, commercial establishments and industry will be severely compromised [38]. At the 
heart of these vulnerabilities is the supervisory control and data acquisition (SCADA) system. 
These systems control, collect and analyze data from equipment which are employed largely in 
power generation stations, transmission and distribution sub-stations and utilities which deliver 
the power to households, commercial establishments and the industry [39]. 
 
4.1 NIST data on Cyber-Physical Security 
 The National Institute of Standards and Technology (NIST) have specified threats to 
computer related commerce in North America. Over 66% of these threats relate to SCADA 
systems, directly connected to power substations [39]. The risks involved in SCADA controllers 
are not very different from those that personal computers and commercial computer networks 
57 
 
encounter. As manual labor is being replaced by automated control, equipment are being 
networked for control from a single SCADA system. To avoid the cost of private lines which carry 
control data, the internet is an attractive option to network such devices. This provides an 
opportunity for intrusion into these systems which control critical infrastructure [40]. Security in 
these systems can be built around well-established security standards for commercial networks. 
NIST has established a Smart Grid Cyber Security Coordination Task Group (CSCTG) which 
addresses and evaluates processes leading to comprehensive cyber security policies for smart grids. 
The types of risks assessed by the CSCTG are not limited to intentional attacks from disgruntled 
employees, terrorists and industrial espionage but also include; 
? Complexity of the grid, leading to weak points, unintentional errors, openings to attackers. 
? Inter-connection of networks, leading to cascading errors. 
? Denial of service attacks resulting from complex communication networks. 
? Breach of consumer privacy. 
The group CSCTG defines cyber infrastructure and cyber security with respect to smart grids as 
Cyber Infrastructure: 
? Electronic information and communications systems and services and the information contained 
in these systems and services. 
? Information and communications systems and services are composed of all hardware and 
software that process, store, and communicate information, or any combination of all of these 
elements. Processing includes the creation, access, modification, and destruction of information. 
Storage includes paper, magnetic, electronic, and all other media types. 
? Communications include sharing and distribution of information. These include the following: 
o computer systems; control systems (SCADA); 
58 
 
o Networks, such as the Internet; and cyber services (managed security services). 
Cyber Security is defined as ?The protection required to ensure confidentiality, integrity and 
availability of the electronic information communication system.? [41] 
 
4.1.1 Risks to the Smart Grid 
? Intrusion risks: This form of risk is not new to information networks. Since the smart grid will 
be based on an information network, hacking into AMI or utility databases is a huge threat. 
Intrusion vulnerability points include SCADA systems, modems used by components and 
other networking components such as smart meters and any device participating as a routing 
device in the Advanced Metering Infrastructure (AMI). 
? Malware: Malware or malicious software can be found across the public internet in large 
numbers. The inter-connection of the smart grid network with the public internet will cause a 
potential spread of these malware into grid software, causing disruption in power services or 
false data being exchanged in communication links [42]. 
? Espionage and information warfare from state sponsored sources. 
 
4.1.2 Mitigating risks in the Smart Grid 
 A number of steps can be taken to mitigate many of these vulnerabilities. A combination 
of many of them will provide defense-in-depth solutions that will be much superior to any single 
strategy or mitigation. We propose the following mitigation strategies: 
? Restricting Smart Meter data: The AMI and the smart meters generate a significantly large 
amount of data even with reporting frequencies as low as 1 packet/4 hours. With over 50 
million smart meters currently in deployment, the AMI generates data on par with some 
59 
 
internet applications. By restricting the data frequency, we can not only improve the reliability 
of smart utility networks as evident from the results in Chapter 3, but also mitigate the risk of 
simple replay attacks and the more complex man-in-the-middle type attacks popular in IP 
networks.   
? Strong Encryption: Using state-of-the-art cryptographic standards popularly deployed in e-
commerce applications for smart metering data will mitigate simplistic attacks on intercepting 
data. Further in this chapter, we will describe two public-key cryptographic schemes and the 
trade-offs in adding security in smart utility networks. 
? Network Isolation: Although this mitigation strategy defeats the purpose of having an Internet 
Of Things (IoT) architecture, in which all smart metering devices will be accessible on the 
IPv6 world wide web [43], it is prudent to isolate the communication network of the AMI with 
the global internet until proven and effective security mechanisms are in place to protect the 
smart grid infrastructure.  
 
4.2 Public-Key Cryptographic Schemes 
 Public key cryptographic schemes have been widely deployed in the Internet for a variety 
of applications ranging from e-commerce to social networks. The two popular public-key 
cryptographic schemes of RSA (Rivest-Shamir-Adleman) [44] [45] and ECC (Elliptic Curve 
Cryptography) [46] [47] have been widely used. In this chapter we will describe these schemes in 
brief and analyze the performance of these cryptographic schemes in smart utility networks and 
low-power wireless devices in general. We executed simulation runs of a small network 
comprising of smart metering nodes using the IEEE 802.15.4 low power wireless interface. The 
simulation framework has been validated and the validation is described in chapters 2 and 3.  
60 
 
4.2.1 RSA Public-Key Cryptography 
 RSA was proposed in 1977 by Rivest, Shamir and Adleman. Its security is based on the 
presumed intractability of the integer factorization problem. The RSA function is defined as 
follows [48]: 
1: Let g1868	 g3405g1869 be large prime numbers and define g1866 = g1868g1869 and g2004 =g4666g1868?1g4667g4666g1869?1g4667 
2: Choose random encryption exponent g1857,1 g3407g1857 g3407 	g2004, so that g1859g1855g1856g4666g1857,g2004g4667 =1 
3: Use the Euclidean algorithm to find decryption exponent g1856,	1 g3407 g1856 g3407 g2004, so that                       g1857g1856	 g1568
1	g4666g1865g1867g1856	g2004g4667 
4. Define g1858 g1527g1852g3041	?	g1852g3041 by g1858g4666g1865g4667=	g1865g3032	g1865g1867g1856	g1866. g1858g4666g1865g4667is the function for encrypting message g1865. 
 Encryption of a message m is done by following above steps to choose the random 
encryption exponent g1857, and computing the ciphertext g1858g4666g1865g4667= 	g1855 by using equation (4.1).  
           g1855 =		g1865g3032	g1865g1867g1856	g1866                                                          (4.1) 
 Decryption of ciphertext g1855 is computed using the decryption exponent g1856 using equation 
(4.2). The original message  g1865	g1865g1867g1856	g1866 is obtained as the decrypted plaintext. 
             g1855g3031g1865g1867g1856	g1866 g1568		g4666g1865g3032g4667g3031	g1865g1867g1856	g1866	g1568 	g1865	g1865g1867g1856	g1866                            (4.2) 
In the RSA cryptographic scheme, g4666g1866,g1857g4667 is the public key which can be distributed to the 
communication entity, while g1856 is the private key which must not be compromised. If public key 
component g1866 can be factored, then g1856	may be computed and g1865 recovered from g1855 efficiently. 
However, factoring g1866 for correctly chosen g1868 and g1869 remains a hard problem and the premise for the 
robustness of the RSA cryptographic scheme. 
RSA has a signature scheme with message recovery, which can be described as follows. If 
two entities, Alice and Bob wish to exchange a signed message g1865, Alice?s signature on g1865 is g1871 =
g1865g3031	g1865g1867g1856	g1866	, where g1856 is Alice?s private key. On receiving the signature g1871, Bob computes   
61 
 
g1871g3032	g1865g1867g1856	g1866 =g1865 using Alice?s public key pair g4666g1866,g1857g4667. Thus, Bob will obtain a signed message g4666g1865,g1871g4667 
, ensuring the message has not been tampered with. This is an effective defense against man-in-
the middle attacks on communication links.  
 
4.2.2 ECC Public-Key Cryptography 
 The use of elliptic curves over finite fields for data encryption and decryption was first 
suggested by Neal Koblitz and Victor S. Miller. Elliptic Curve Cryptography (ECC) is an alternate 
approach to public-key cryptography vis-?-vis RSA. It is based on the algebraic structure of elliptic 
curves over finite fields. In RSA, the large integer factorization problem is the intractability used 
to provide security. Similarly, in ECC, it is assumed that finding the Discrete Logarithm of a 
random elliptic curve element with respect to a publicly known base point is infeasible. This is 
known as the Discrete Log Problem (DLP) [49].  
 In current ECC practices, the elliptic curve is a plane curve which consists of points 
satisfying equation (4.3), along with a distinguished point at infinity. The ECC principle is 
deployed in the real-world in combination with the Diffie-Hellman (ECDH) key agreement 
protocol. 
                      g1877g2870 =	g1876g2871 +g1853g1876+g1854                                                   (4.3) 
 
 In this scheme, suppose Alice wants to establish a shared key with Bob, on an un-secured 
channel. Initially, the domain parameters g4666g1868,g1858g4666g1876g4667,g1853,g1854,g1833g4667, which are the large prime, cruve 
function, constants and finite field, must be agreed upon. Also, each party must have a key pair 
suitable for elliptic curve cryptography, consisting of a private key g1856, randomly selected in the 
interval g46701,g1866?1g4671 and a public key g1843,g1843 =g1856g1833. So, Alice?s key pair is g4666g1856g3002,g1843g3002g4667 and Bob?s key pair 
62 
 
isg4666g1856g3003,g1843g3003g4667. Each of them must have the other?s pubic key using an exchange mechanism. Alice 
computes a co-ordinate point on the curve g4666g1876g3038,g1877g3038g4667 = g1856g3002g1843g3003. Bob computes, g1863 =	g1856g3003g1843g3002. The shared 
key is the x co-ordinate point of the curve, g1876g3038. Both Alice and Bob have computed the same number 
and have a shared key according to equation (4.4). 
      g1856g3002g1843g3003 =	g1856g3002g1856g3003g1833 =	g1856g3003g1856g3002g1833 =	g1856g3003g1843g3002                               (4.4) 
Since only Alice and Bob can compute the shared key, they can encrypt their messages 
with the shared key without risking decryption by an eavesdropper.  
 
4.3 Power consumption and Processing delay of RSA and ECC 
 In this section we present the results of power consumed by each of the cryptographic 
schemes RSA and ECC. It has been shown that a key size of 1024 bits in the RSA scheme offers 
equivalent security to a 160 bit key size in the ECC scheme [50]. We have implemented each of 
these schemes on the low power wireless hardware platform Tmote Sky [51]. In addition we use 
the validated simulation framework to simulate a 100 node smart utility network to analyze 
processing delays imposed by deploying these cryptographic schemes. We also apply RSA and 
ECC to encrypt routing protocol packets presented in Chapter 2 in order to quantify the processing 
delays incurred due to RSA and ECC. 
For a 9 node real network on the Tmote-Sky hardware platform, power consumption 
comparison of these two encryption schemes versus nodes with no-encryption is shown in figure 
4.1. This data shows that there is a trade-off in implementing security in terms of increased power 
consumption in highly constrained nodes in the low power and lossy smart utility networks. While 
RSA-1024 bit performs the worst consuming maximum power of the three schemes, ECC-160 bit 
performs better.  
63 
 
 
Figure 4.1: Power consumption vs. encryption scheme 
 
Figure 4.2 shows the additional processing delay imposed by implementing RSA and ECC 
over routing protocols in smart utility networks. RSA imposes 10 times more delay per node than 
ECC in this analysis. In signature generation, RSA-1024 is 10 times slower than ECC-160 per 
node. Figure 4.3 shows the power consumed by the hardware during signature generation and 
verification operations for each of the cryptographic schemes. The order of magnitude faster ECC-
160 consumes 11 times less power than the slower RSA-1024 per node during signature 
generation. Due to the larger number of computations involved in ECC signature verification it is 
approximately 5 times slower than RSA and similarly consumes approximately 5 times more 
power than RSA during signature verification operations. 
64 
 
 
Figure 4.2: Time taken by RSA and ECC for signature generation and verification 
 
Figure 4.3: Power consumed by RSA and ECC for signature operations 
65 
 
4.4 Summary 
 In this chapter, we present the urgent need for securing the communication architecture for 
the smart power grids. We detail the threats to the communication infrastructure and propose 
appropriate mitigation schemes. Further, we describe two public key cryptographic schemes of 
RSA and ECC. We compare the performance of the two cryptographic schemes in terms of power 
consumed and processing delay imposed by implementing them on highly resource constrained 
nodes in the smart utility network. We find that ECC performs better than RSA in terms of power 
consumption. Although RSA performs better in signature verification, the gains are moderate in 
comparison to the power and processing time saved by implementing ECC. We choose ECC has 
the applicable cryptographic scheme to be deployed in smart utility networks. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
66 
 
 
 
 
 
 
Chapter 5 
Conclusions and Future Work 
 
 In this chapter, we summarize the research work presented in this dissertation and follow 
it up with possible future directions to this work. 
 
5.1 Summary of Research Work 
 In this dissertation, we have introduced and studied the aspects of Reliability, Scalability 
and Security in smart utility networks in detail. We have developed a validated simulation 
framework using OMNeT++ in order to execute large scale simulations of smart utility networks. 
We have presented the wireless mesh architecture as the favored method of deploying the 
Advanced Metering Infrastructure (AMI). Further, we have presented an analysis of emerging 
wireless mesh routing protocols that are most applicable to these networks. In addition, we have 
presented two approaches to determine the upper bound on scalability at the collector in smart 
utility networks, backed by simulation results using data from large scale simulations conducted 
at the high performance computing facility at Alabama Supercomputing Authority in Huntsville, 
AL.  
In Chapter 1, we introduced the various components of the Smart Grid and emphasized on 
the importance of efficient data routing protocols to enable the future power grid. We discussed 
the disadvantage of existing communication infrastructure for the power grid and the urgent need 
67 
 
to move to a synergy of internet communication technologies and the power grid. Further, we 
presented the smart meter and smart utility network as being at the heart of a successful smart grid 
implementation. 
In Chapter 2, we presented the wireless mesh architecture of deployment as the most 
favored method. We discussed the wireless environment and channel conditions that are prevalent 
in the deployment zones of smart meters. We presented the simulation framework that we 
developed using OMNeT++ and Castalia to closely model the wireless channel conditions, low 
power radio based on IEEE 802.15.4 standards and drafts of routing protocols. Further, we have 
shown the validation method by comparing corresponding results with real world experiments. We 
then discuss the three routing protocols of RPL, LOAD and Geographical routing used to simulate 
network sizes of up to 7500 nodes in the thus developed simulation framework. We presented the 
performance comparison of these routing protocols and their applicability to smart utility 
networks. RPL is the most applicable of the three routing protocols analyzed, outperforming 
LOAD and Geographical routing in terms of end to end delay, hop count and packet delivery ratio. 
It is more reliable that geographical routing and LOAD in the order of reliability.  
In Chapter 3, we studied the scalability in smart utility networks with the aim of 
determining an upper bound on the number of metering nodes that can be supported at each 
collector device. We presented two approaches, one dependent on link packet success probabilities 
and network topology and the other based on network traffic arrival rates and their effect on 
scalability. We backed this model with large scale simulations, the results of which closely match 
the theoretical model proposed. The relationship between average link packet success probabilities 
and maximum expected successful transmissions in the network was established, allowing to 
estimate network scale. In the second approach we modeled the smart utility network traffic as a 
68 
 
two-state Markov Modulated Poisson Process. Following collection of simulation traces of packet 
arrivals at the collector for varying network sizes, we fit the raw data to the MMPP model and 
presented the results of how scale affects performance of the queue at the collector. In our findings, 
network performance degraded exponentially in terms of packet loss rates, delay and blocking 
probabilities beyond 4000 nodes per collector for more than 50% offered loads. This concluded 
the scalability analysis in smart utility networks. 
In Chapter 4, we presented the urgent need for securing communication protocols for the 
smart grid. We presented some of the security risks and their mitigation in smart grid 
communication networks. Further, we detailed two popular cryptographic schemes of RSA and 
ECC. We presented the results comparing performance of the two cryptographic protocols in terms 
of power consumption and computation delay. We have implemented both these cryptographic 
schemes on the Tmote-Sky hardware platform and collected the data from a 9-node real network, 
representing the low power wireless mesh smart utility networks. We concluded in this chapter, 
that trade-offs come with adding security to communication protocols in networks with low power 
radios and highly resource constrained nodes. Under these conditions, the ECC cryptosystem 
performs better than the RSA cryptosystem in terms of power consumption and time taken for 
various cryptographic operations. 
 
5.2 Future Work 
Several lines of future work can be pursued towards meaningful conclusion following this 
work.  
? A more efficient Simulation Framework: Simulating large scale networks poses challenging 
problems in terms of computation time. The available memory is often insufficient to load and 
69 
 
initialize the starting conditions and states for a large number of nodes. In addition, during 
runtime, maintaining states of the routing layer, MAC layer and Radio states poses severe 
limitations on the scale. One approach to solving these problems is to move to a distributed 
computing environment, in which each simulated node?s states are loaded onto different 
machines. Although this solves the memory constraints problem, inter-process and inter-
machine communication to keep data consistent and exchange simulation messages 
significantly slows down the simulation. At scale, the run-time increases dramatically. Figure 
5.1 shows the simulation time versus network deployment. The speed advantages we get from 
parallelization and the use of OpenMPI library for implementing parallelization for OMNeT++ 
and Castalia simulation frameworks will make for an interesting study.  
 
Figure 5.1: Simulated Network size versus Simulation Execution Time 
 
70 
 
? Wireless Channel Models with higher accuracy: Although we have validated the simulation 
framework with real-world results, there are spatio-temporal aspects of wireless environment 
that can be modeled more closely and accurately. The real challenge is in making the model 
scalable to large scale simulations and efficiently implemented on computing systems. 
? Rapid Prototyping: Although we have tried to cover all the aspects of the model and routing 
protocol drafts in our code implementations, there are always patches and modifications 
emerging from IETF standards bodies and other groups. A suitable framework to rapidly 
prototype such changes on both simulation frameworks and suitable hardware devices will 
greatly benefit the community of researchers in the wireless networks domain.  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71 
 
 
 
 
 
 
Bibliography 
 
[1]  I.F. Akyildiz and X. Wang, ?A Survey on Wireless Mesh Networks," IEEE Communications 
Magazine, vol. 43, Sept. 2005. 
[2]  ?NIST Framework and Roadmap for Smart Grid Interoperability Standards,? US Dept. of 
Commerce, Release 1.0. 
[3]  ?Utility Scale Smart Meter Deployments, Plans and Proposals,? Institute for Electric 
Efficiency, May 2012.  
[4]   T. Winter, P. Thubert, ?RPL: IPv6 Routing Protocol for Low Power and Lossy Networks,? 
draft-ietf-roll-rpl-19.  
[5]  T. Watteyne, K. Pister, D. Barthel, M. Dohler, I. Auge-Blum, ?Implementation of Gradient 
Routing in Wireless Sensor Networks,? IEEE GLOBECOM, Hawaii, December 2009. 
[6]  K. Kim, S. Daniel, et.al. , ?6lowpan Ad-hoc On-demand Distance Vector Routing Protocol 
(LOAD),? draft-daniel-6lowpan-load-adhoc-routing-03                                                                
[7]  T. Clausen, et.al. , ?The Lightweight On-demand Ad hoc Distance-vector Routing Protocol ? 
Next Generation (LOADng),? draft-clausen-lln-loadng-08. 
[8]  B. Lichtensteiger, B. Bjelajac, C. M?ller and C. Wietfeld, ?RF Mesh Systems for Smart 
Metering: System Architecture and Performance,? IEEE SmartGridComm, Maryland, Oct. 
2010. 
[9]  G. Iyer, P. Agrawal, E. Monnerie, R. Cardozo, ?Performance Analysis of Wireless Mesh 
Routing Protocols for Smart Utility Networks,? IEEE SmartGridComm, Brussels, Oct 2011.  
72 
 
[10] T.S. Rappaport, ?Wireless Communications: Principles and Practice,? IEEE Press, 
Piscataway, NJ, 1996. 
[11] Zigbee Alliance, ?Zigbee Specification,? October.  2010                    
http://www.zigbee.org/Specifications.aspx 
[12]  Zigbee Alliance, ?Understanding Zigbee Gateway,?, September 2010 
http://www.zigbee.org/LearnMore/WhitePapers.aspx  
[13] M. McGranaghan, D. Von Dollen, P. Myrda, E. Gunther, ?Utility experience with 
developing a smart grid roadmap,? IEEE Power and Energy Society General Meeting - 
Conversion and Delivery of Electrical Energy in the 21st Century, 2008   
[14] A. Boulis, ?Castalia Wireless Channel characteristics,? Castalia User Manual, v. 3.2, 
NICTA, March 2011. 
[15] P. Levis, et.al. , ?The Trickle Algorithm,? IETF RFC 6206.  
[16] I.C. Perkins, et.al. , ?Ad hoc On-Demand Distance Vector (AODV) Routing,? IETF        
RFC 3561. 
[17] A.F. Varga, R. Hornig, ?An overview of the OMNeT++ simulation environment,? ACM 
SIMUTOOLS 2008, no. 60.  
[18] Castalia Framework for OMNEST simulator, http://castalia.npc.nicta.com.au/   
[19] J. Jun and M.L. Sichitiu, ?The nominal capacity of wireless mesh networks,? Wireless 
Communications Magazine, Vol. 10, no. 5, pp. 8-14, Oct. 2003. 
[20] P. Gupta and P.R. Kumar, ?The capacity of wireless networks,? IEEE Trans. On 
Information Theory, vol. 46, no. 2, pp. 388-404, Mar? 2000. 
[21] J. Li, C. Blake, D.S.J. De Couto, H.I. Lee, R. Morris, ?Capacity of Ad-hoc wireless 
networks,? Proceedings of ACM MobiCom, July, 2001.  
73 
 
[22] G. Iyer, P. Agrawal, R. Salazar, ?Analytic model and simulation study for network 
scalability in smart utility networks,? IEEE ISGT Asia 2013, pp. 1-6, Bangalore, India. 
November 2013. 
[23] Y. Feng, X. Shen, Z. Gao, G. Dai, ?Queuing Based Traffic Model for Wireless Mesh 
Networks,? 15th International Conference on Parallel and Distributed Systems, 2009.  
[24] X. Wu, J. Liu, G. Chen, ?Analysis of Bottleneck Delay and Throughput in Wireless Mesh 
Networks,? IEEE Conference on Mobile Adhoc and Sensor Systems, 2006. 
[25] C. Hauser, E.D. Bakken, A. Bose, ?A Failure to Communicate,? IEEE Power and Energy 
Magazine, pp. 47-55, Mar-Apr, 2005. 
[26] W. Fischer, K. Meier-Hellstern, ?The Markov-modulated Poisson process 
(MMPP) cookbook,? Performance evaluation, Elsevier, 1993.  
[27] P. Heyman, D. Lucantoni, ?Modeling Multiple IP Traffic Streams with Rate Limits,? 
IEEE/ACM Trans. On Networking, vol. 11, no. 6, December, 2003. 
[28] S. Bali, V.S. Frost, ?An algorithm for fitting MMPP to IP traffic traces,? IEEE 
Communication Letters, vol. 11, no. 2, pp. 207-209, February 2007. 
[29] H. Heffes, ?A class of data traffic processes?Covariance function characterization and 
relating queuing results,? Bell Syst. Tech. J., vol. 59, pp. 437?488, 1980. 
[30] K. Meier-Hellstern, ?A fitting algorithm for Markov-modulated Poisson processes having 
two arrival rates,? Eur. J. Oper. Res., vol. 29, pp. 370?377, 1987. 
[31] S. Andersson and T. Ryden, ?Maximum likelihood estimation of a structured MMPP with 
applications to traffic modeling,? 13th ITC Specialist Seminar, Monterey, CA, 2000. 
[32] Y. D. Lee, A. van de Liefvoort and V. L.Wallace, ?Modeling Correlated Traffic with a 
Generalized IPP,? Perform. Eval. , vol. 40, no. 1?3, pp. 99?114, 2000. 
74 
 
[33] I. L. MacDonald and W. Zucchini, ?Hidden Markov and Other Models for Discrete-Valued 
Time Series,? London, U.K.: Chapman-Hall, 1997. 
[34] C. Blondia and T. Theimer, ?A discrete-time model for ATM traffic,? RACE, Doc. PRLB-
123-0018-CD-CC/UST-123-0022-CD-CC, 1989. 
[35] D. Lucantoni, ?The BMAP/G/1 Queue: A Tutorial,? Models and Techniques for 
Performance Evaluation of Computer and Communications Systems, L. Donatiello and R. 
Nelson, Eds. New York: Springer-Verlag, 1993, pp. 330?58. 
[36] D. Lucantoni, ?New results for the single server queue with a batch Markovian arrival 
process,? Stochastic Models, vol. 7, pp. 1?46, 1991. 
[37] J. D. C. Little, ?A proof for the formula L = ?W,? Oper. Res., vol. 14, pp. 723?27, 1961. 
[38] W. Dong, L. Yan, S. Paul, J. Mohsen,  R. Kenneth, M. Michael, ?Power Infrastructure 
Security: Fundamental Insights of Potential Cyber Attacks and Their Impacts on Power Grid,? 
Part of project ?Protecting Intelligent Distributed Power Grids against Cyber Attacks? for 
DOE. 
[39] O. Paul, O.S. Edmund, F. Deborah, ?Concerns About Intrusions into Remotely Accessible 
Substation Controllers and Scada Systems.? Schweitzer Engineering Laboratories, Inc., 2000. 
[40] W. David, ?Security and Vulnerability in Electric Power Systems,? NAPS 2003, 35th North 
American Power Symposium, University of Missouri-Rolla in Rolla Missouri, October 20-21, 
2003. pp.559-566. 
[41] National Institute of Standards and Technology; ?NIST Framework and Roadmap for 
Smart Grid Interoperability Standards Release 2.0,? U.S. Department of Commerce, September 
2009. 
75 
 
[42] G. Iyer, P. Agrawal, ?Smart Power Grids,? 42nd IEEE SSST 2010, Tyler, TX. pp. 152-155, 
March 2010. 
[43] C. Bennett, D. Highfill, ?Networking AMI Smart Meters,? IEEE Energy 2008, Energy 
2030 conference, pp. 1-8, November 2008. 
[44] R.L. Rivest, A. Shamir, L. Adleman, ?A method for obtaining digital signatures and public-
key cryptosystems,? Communications of the ACM magazine, vol. 21, no. 2, pp. 120-126, Feb. 
1978. 
[45] H. Williams, ?A modification of RSA public-key encryption procedure,? IEEE Trans. On 
Information Theory, vol. 26, no. 6, pp. 726-729, November 1980. 
[46] N. Koblitz, ?Elliptic curve cryptosystems,? Mathematics of Computations, no. 48, pp. 203-
209, 1987. 
[47] S. Kumar, M. Girimondo, A. Weimerskirch, C. Paar, ?Embedded end-to-end wireless 
security with ECDH key exchange,? 46th IEEE Midwest Symposium on Circuits and Systems, 
vol. 2, pp. 786-789, December 2003. 
[48] D.R. Hankerson, D.G. Hoffman, D.A. Leonard, C.C. Lindner, K.T. Phelps, C.A. Rodger, 
J.R. Wall, ?Coding Theory and Cryptography: The Essentials,? New York, Marcel Dekker, 
2000. 
[49] R. Balasubramanian, N. Koblitz, ?The Improbability That an Elliptic Curve Has 
Subexponential Discrete Log Problem under the Menezes?Okamoto?Vanstone Algorithm,? 
Journal of Cryptology, Springer-Verlag, vol. 11, no. 2, pp. 141-145, March 1998. 
[50] K. Piotrowski, L. Peter, S. Peter, ?How public key cryptography influences wireless sensor 
node lifetime,? SASN 2006 4th ACM workshop on Security of ad hoc and sensor networks, pp. 
169-176, 2006. 
76 
 
[51] Tmote sky IEEE 802.15.4 development kit, http://www.ti.com/tool/msp430-3p-motei-
tmotesky-dsgkt