An Adversarial Evaluation of Network Signaling and Control Mechanisms - Information Security and Cryptology-ICISC 2010 - page 258

9

x 10

500

15

Mean
99%

450

SN
HYBR
DHT

400

350

10

300

250

200

5

150

100

50

0

0

(a) Latency measurement (b) Control cost measurement

Fig. 1. The evaluation results under normal operations. Figure (a) depicts the mean notification
time and the completion time for 99% of the nodes and for different control plane mechanisms.
Figure (b) presents the total network cost in terms of bytes for the test duration of 600 seconds.

5.1 Evaluation Metrics

To evaluate the reliability and efficiency of the different control plane mechanisms, we
introduce three metrics: coverage , latency ,and control efficiency . Coverage is measured
by enumerating the number of nodes that receive the alert message when the system is
under attack. Alternatively, all alert messages that are not delivered within the duration
of the experiment instance are regarded as failures. Latency is defined to be the period
of time that it takes for alerts to reach each participant from the moment that an alert
message is dispatched. Control efficiency is the cost to utilize the control mechanisms.
This is calculated by summing the total number of bytes required for network operations
during the experiment.

5.2 Evaluation Design

To validate the network behavior under adversarial conditions, we implemented three
control mechanisms using the OverSim [3] simulation framework. The use of simula-
tion was mandated for the following reasons:

Scalability. We were interested in observing the behavior of large scale networks imple-
menting signaling systems in the presence of network-wide malicious attacks. Having
tens of thousands of number of participants, Oversim framework enabled us to quantify
the design parameters that really influence the behavior of the system.

Global Adversary. Emulating global adversary in a real-world large-scale testbed is
a costly and time-consuming task and it does not allow repetition of experiments. The
simulation framework not only helped us to instantiate this size of network but also
provided the interface to implement a more precise behavior for the global adversary.