Research

Overview

            1. Security protocols for the Internet of Things
            2. Security and privacy for cyber-physical systems (smart grids and
                transportation systems)
            3. Protocols and analytic models for computer networks

1. Models for IEEE 802.11 (WiFi): Over the last 20 years, IEEE 802.11 or WiFi
has become one of the most popular network access technologies. Consequently,
understanding the factors affecting its performance and developing mechanisms to
address them is an important research area. I have a large body of work on various
performance modeling aspects related to IEEE 802.11. The most important among
these are queueing models to evaluate the delays experienced by packets in IEEE
802.11 networks. The developed models provide expressions for the probability
generating function for the queue lengths and the delays.The distinguishing features
of these models include the ability to accommodate finite buffers, unsaturated traffic
conditions, and arbitrary traffic arrival processes and number of users. In a sequence
of papers (among the seminal work in this area), we have developed queueing models
for both distributed (DCF) and centralized (PCF) modes of operation of IEEE 802.11.
The models can be used to evaluate probabilistic service guarantees in terms of both
the delays and packet loss probabilities and used for purposes like call admission
control and providing statistical delay bounds.

In another line of work, my group has worked on developing models that evaluate
the spatial reuse (i.e., the simultaneous use of the same spectrum in geographically
separated locations) in IEEE 802.11 networks and then use insights from these
models to develop protocol enhancements. In many real world scenarios where the
node density is high, the interference between the nodes becomes the dominant
factor affecting the spatial reuse and the overall network performance. Our work
has developed stochastic models to evaluate the spatial reuse of the network and
its scaling, and the models may be used to characterize the variability and rate of
successful transmissions at each node. Our model is general and applicable for
the class of unbiased MAC protocols. In the specific case of IEEE 802.11 in
multi-hop scenarios, our work has shown how its aggressive behavior can throttle
the spatial reuse and reduce bandwidth efficiency. We have designed adaptive
and distributed coordination schemes for IEEE 802.11 using explicit MAC-layer
feedback to pace the transmissions on adjacent nodes, thereby assisting the
protocol to operate around its saturation state while minimizing resource contention.

The third line of my work on IEEE 802.11 is on the development of models that
characterize the impact of the protocol on the resulting traffic characteristics of the
network. We have developed analytic models for characterizing the interarrival time
distribution in IEEE 802.11 networks. Our results show that the interarrival time has
a multimodal distribution and demonstrates evidence of pacing induced by the MAC
protocol. While the interarrival time distribution gives important insights into the
traffic characteristics, the scaling, burstiness and long-range dependence associated
self-similar and multifractal has a significant impact nature of traffic can lead to a
number of undesirable effects like high buffer overflow rates, large delays and
persistent periods of congestion. We have also investigates the impact of IEEE
802.11 MAC protocol on the second order scaling of the traffic. Our results show that
while individual TCP sources in wireless networks show evidence of self-similarity,
the aggregate traffic is no longer self-similar and shows multifractal properties.

My research on wireless networks has been funded through National Science Foundation,
USA, Intel Corporation, WiMAX Forum, and Ministry of Education, Singapore.

2. Models for TCP: An overwhelming majority of the data transferred over the Internet
uses TCP as the transport layer protocol. Consequently, the performance of TCP has a
significant impact on the overall network performance as seen by the users. My most
significant contribution in this area has been the development of models for the steady-state
throughput and latency of TCP flows. Papers prior to my work had looked at steady-state
throughput, a scenario that is easier to analyze but has limited use in practical scenarios.
Our work provides results for both steady-state throughput of long flows as well as the
transfer latency of short flows (the majority of the flows in the Internet). Also, our work
comprehensively considered all the version of TCP (Reno, Tahoe, and SACK) that were
prevalent at the time of publication. A key insight obtained from the model showed that
with droptail queues implemented by most routers in the Internet, contrary to prevailing
wisdom, TCP SACK failed to provide adequate protection against timeouts. The model
also showed the importance of loss models in TCP by proving that TCP SACK performs
better than TCP Tahoe and TCP Reno under independent losses, and as losses become
correlated, TCP Tahoe can outperform both TCP Reno and TCP SACK.

The second line of my work in this area was to develop models for TCP traffic that
explained short time-scale scaling behavior in network traffic. My work was the first to
explain the contribution of TCP’s slow-start and congestion avoidance mechanism to the
self-similarity in network traffic. Using a mathematical formulation, we showed that TCP's
retransmission and congestion control mechanism results in packet dynamics of a TCP
flow being analogous to a number of ON-OFF sources, with OFF periods taken from a
heavy tailed distribution, which in turn contribute to the self-similar nature of TCP traffic.
Our work has also developed mechanisms to reduce the degree of second-order scaling
in network traffic. Our method works by reducing two related causes in TCP: (1) timeouts
and exponential backoffs (2) burstiness and ACK compression. We have proposed a
simple modification to the RED algorithm that leads to significant reductions in both multi
and mono fractal properties of TCP traffic. Although our techniques are aimed at small
time-scale TCP related causes of scaling, they are also effective in reducing the degree
of self-similarity in traffic even when application and user level causes are also present,
as long as TCP is used as the underlying transport protocol.

My work in this area was funded by DARPA, USA and National Science Foundation, USA.

1. Attack detection in power grids: Smart grids take advantage of information and
communication technologies to achieve energy efficiency, automation, and reliability.
Smart grids rely on measurement and monitoring systems to track the flow of electricity
and other parameters that allow the estimation of system state and consequently develop
necessary control actions. Traditionally, Supervisory Control and Data Acquisition
(SCADA) systems and more recently, Phasor Measurement Units (PMUs) form the core
of this monitoring system. Given the critical role they play in the operation and control of
the grid, the integrity of the data they generate is extremely important. My work has
focused on developing mechanism to detect false data injection and data manipulation
attacks in these scenarios. attacks on the integrity of the data generated by these devices.
My work has pioneered the use of physical laws (e.g., transmission line parameters) to
detect such attacks. For the scenario where the adversary has compromised one or more
PMUs and is maliciously injecting false data, we have developed an attack detection
mechanism that is also capable of isolating the parameter being modified as well as the
location of the attack. The proposed method is based on continuously monitoring the
equivalent impedances of transmission lines, and allows data integrity to be tested in a
distributed and non-iterative manner, thereby requiring less memory and processing,
and facilitating early detection using legacy systems possible. My work has also
considered the problem of detecting grayhole attacks on PMU data where compromised
routers in the network maliciously drop packets with measurement data. The main
challenge in detecting gray-hole attacks is to distinguish between packets maliciously
dropped by an attacker and those dropped naturally due to network congestion. Our
detection mechanism exploits the inherent characteristics of PMU data to detect
malicious packet drops and avoids the need for network support. The detection
mechanism tracks the one-way,end-to-end packet delays using the GPS timestamps
in PMU data, and uses the temporal trend in delays to classify the cause of the packet
loss. Finally, my work has also considered attack detection for SCADA networks. In
particular, we have considered the scenario where PMUs are used in the grid to
complement the SCADA measurements. The proposed method exploits the correlation
between SCADA and PMU data, and the classification of tampered and real data is
done through a difference measure that we have developed. Our more recent work
has developed techniques to detect attacks targeting the injection of malicious
commands in the SCADA network.

2. Privacy issues in smart grids: Smart girds rely on the use of smart meters for
pricing and feedback purposes, and the data generated by these meters is useful for
load forecasting, demand-response management, and dynamic pricing. However, the
recording and transmission of power consumption profiles from homes and businesses
leads to serious privacy issues. My work was the first to address the problem of
ensuring consumer privacy while facilitating dynamic billing and demand-response
management. To achieve such private data aggregation, we first developed an
efficient authentication and key establishment scheme. Then, a masking based data
aggregation scheme for a group of consumers (e.g. from a region/locality) was
developed that ensures the privacy of individual customers while providing a higher
degree of efficiency by avoiding the need to perform any asymmetric cryptographic
operations. We have also developed spatial data aggregation schemes that
provide up-to-date and accurate aggregated consumption information to the power
grid about any group of consumers. Our solutions only utilize lightweight cryptographic
primitives like one-way hash functions and exclusive-or operations. My comprehensive
body of work in this area addresses smart meter privacy issues in a wide range of
scenarios: dynamic billing, spatial data aggregation, hop-by-hop data aggregation,
and vehicle-to-grid communication.

3. Security for IoT using hardware security primitives: Existing security protocols
and techniques for traditional computing devices cannot be adopted for many IoT
systems due to two reasons. First, IoT devices are limited in terms of their computing,
memory, and energy resources. Second, many IoT devices are physically un-protected
and installed in locations that are easily accessible to adversaries, thus making them
susceptible to physical and side-channel attacks. My work has developed security
protocols that address these challenges for IoT devices by using hardware security
primitives in the form of physically unclonable functions (PUFs). My work was the
first to develop practical authentication protocols using PUFs that did not impose
unrealistic memory requirements in servers. Our protocol requires the authentication
server to store only one challenge-response pair for any IoT devices at a given point
in time. In addition, our protocols allow system operators to trade-off energy and
security, by developing a PUF based systems that allows the dynamic selection of
key lengths. The authentication protocols also provide a mechanism to establish a
secure session key without any extra computational or communication overhead.
Our work addresses the authentication of an IoT device with a server, as well as the
mutual authentication of two IoT devices.

In another body of work, my group has demonstrated how two-factor authentication
as well as data provenance in IoT devices may be accomplished by using PUFs.
My group has developed lightweight and privacy-preserving two-factor authentication
schemes for IoT devices where PUFs are used as one of the authentication factors.
A device using our scheme remain secures even if an adversary has physical access
to the IoT device. Our solution for data provenance ensures that the data is
generated by a specific IoT device at the specific location where it is supposed to be
deployed. A technique based on RSSI measurements of the wireless channel
between two entities is used to establish data provenance in terms of the location of
data, while PUF based authentication protocols are used to establish the data
provenance in terms of the source of data.

Finally, my group has developed remote firmware attestation protocols for IoT
devices. A key feature of the our attestation technique is its ability to ensure high
availability of the IoT devices while the attestation is being carried out. Our
techniques uses minimal secure hardware combined with a randomized approach
to detect roving malware in IoT devices. In addition, PUFs provide protection to the
secrets of the IoT device against physical attacks. This technique is currently
undergoing evaluation for implementation by Singtel in its products.

This research has been funded by National Research Foundation (Singapore), Ministry of
Education (Singapore), National Science Foundation (USA), and New York Power Authority
(USA).

Current Research Projects: