03-09-2012, 11:40 AM
ANALYSIS OF PACKET FLOWS IN SIMULATED AD HOC NETWORKS USING STANDARD NETWORK TOOLS
ABSTRACT
In this thesis we introduce a new utility for the GlomoSim and Qualnet wireless
network simulators that allows them to output and log packets in a standard network format.
This allows analysis, debugging, and visualization of protocol and simulator behavior to be
performed in great detail, far beyond that of the built-in simulator statistics engine. Using
our tool, any standard network analysis utility that supports the packet capture library (pcap)
interface can read packets directly from the simulator.
As an example application, we examine the performance of several routing protocols
for mobile nodes in wireless ad hoc networks. The routing protocols are on-demand
(reactive) protocols and can be compared by their ability to deliver stable paths between a
source and destination for the life of a transmission, while maximizing throughput and minimizing
the load introduced into the network. The performance of the routing protocol can also
have drastic effects on higher level protocols like the Transmission Control Protocol (TCP).
Additionally, we examine this impact to gain high-level insight into the performance of the
routing protocol and the behavior of the simulator.
Introduction
As wireless computer networks become larger and more complex, the protocols that direct
their behavior become harder to design and implement. The size and cost of such networks
also makes testing new protocols on live systems prohibitive. To alleviate this problem, protocol
designers are turning to developing new implementations on network simulators before
attempting to deploy them in real networks.
Network simulators such as NS-2 [1], GlomoSim [2], Qualnet [3], and OpNet [4]
give developers the ability to test their protocols on nearly any network configuration. This
helps them to eliminate bugs and design flaws, as well as raise confidence that the protocol
will behave correctly and perform up to expectations when used in a live environment. To that
end, network simulators strive to emulate a real network environment with as much precision
as possible.
RelatedWork
Significant work has been done to develop modern network simulators. This work has required
the development of complex mathematical models, statistics reporting engines, protocol
implementation architectures, and methods of visualization. Using these simulators, many
routing protocols have been empirically studied with a fair amount of success, along with their
affect on the performance of TCP.
Relying on these network simulators has proven to be reasonably safe, but they
remain limited in their abilities. The primary limitation comes from the fact that even the
best simulator estimations about real network behavior can be flawed, especially when the
estimations are strongly connected to network settings chosen by the simulator user, which
may be based on flawed assumptions.
Wang and Kung found that data collected from discrete event schedule based simulators
has many drawbacks when compared to data collected from real networks [11]. They
describe an approach to create a simulator using real hardware/software mechanisms from the
operating system to allow the use of real network API’s and analysis tools.
Mobile Ad Hoc Networks
Amobile ad hoc network (MANET) is a collection of radio enabled nodes which communicate
with each other in a broadcast fashion. There is no infrastructure to facilitate communication
or organization. In a large enough area, this can cause “multi-hop” operation where transmitting
a message from a source to a destination may require the assistance of intermediate nodes
to relay the message. Nodes in this type of network may also move around, going in and out
of transmission range with their neighbors. Thus, the network topology is extremely unstable,
where all bindings between neighbors may only last for a short period of time.
Because of the dynamic nature of MANETs, finding a path from a source to a destination
can be very difficult. Some of the best performing modern routing protocols operate in
a reactive or on-demand fashion. They include the Ad hoc On-demand Distance Vector protocol
(AODV) [8], Dynamic Source Routing protocol (DSR) [10], and the Labeled Distance
Routing protocol (LDR) [9]. These protocols work by using a two phase route request and
route reply process as routes are needed. The protocols are soft state with routing information
expiring after a short lifetime.
Summary of Protocols
In this thesis, we will examine AODV, DSR, and LDR, all of whom are on-demand and soft
state. When a host has data to send to a destination but does not have a route, it broadcasts
a Route Request for that destination. Intermediate nodes forward this broadcast to their
neighbors so that the request is propagated through the network, eventually arriving at the
destination. Once the destination receives the request, it sends back a Route Reply, usually
along the reverse path that the original request took. The reply is used to inform the sender
and intermediate nodes of a potential route to the destination.
The primary difference in the protocols comes from their strategies for maintaining
route information at the nodes, their strategies for avoiding routing table loops, and their
efforts to curtail unnecessary flooding of routing control traffic.
DSR maintains routes by constructing paths and placing them in each packet header.
Path construction takes place on the transmission of a Route Reply packet. Each node along
the reverse path to the sender adds itself to the list of nodes in the header and forwards along
towards the source (source routing).
Conlcusion
Having developed our packet dumping utility for QualNet and GlomoSim, it is clear that a
similar tool could certainly be developed for other network simulators with the same benefits.
Other simulators that use real network headers at least as a basis for their packet structure
are good candidates for packet dumping. A key to such a port would be making sure that
all appropriate points of packet traversal are attached to the utility as they would be in a real
system.
More mature versions of the tool could include lower-level MAC headers, so that
wireless network infrastructure utilities like Kismet [21] could be utilized. These tools can
analyze specific characteristics in the network such as signal strength over time, traffic on
multiple channels, link breakage, security protocols, and more.
However, with the utility in its current state, we have been able to conduct a very
basic analysis of routing and TCP performance on top of LDR and AODV. The advantages
and disadvantages of each protocol were shown to be in line with expectations. We were also
able to show how the utility can be used to assist in simulator data visualization.