Billing records generated for thousands of users of a telecommunication network are a gold mine to the network service providers, for both business and technical usages. However, how to mine the "golden nugget" from the enormous amount of data is not an easy task.

This research project deals with the billing record of a genuine wireless network. We plan to apply data mining technology on the real network data, and to develop new algorithm for clustering user groups or mobility patterns. The challenges of this task lay in the very nature of the network billing data: high dimensionality of data requires more computing power and efficient algorithm, temporal and spatial data require special treatment, while the clustered user groups require reasonable descriptions for better understanding by marketing experts.

We work with the genuine billing records collected from the Telus Mobility CDPD (Cellular Digital Packet Data) network. These network data are more typical than data obtained via simulations. The records were collected in various intervals, from several hours to of almost 20 days. Prior study of these records dealt with the discovery of network topology by identifying ``neighboring'' cells and with clustering of user groups employing AutoClass. We plan to extend the analysis of records by developing novel and viable methods for extracting useful knowledge particular for the telecommunication network, by constructing clustering algorithm specialized for network data, and by giving an understandable representation of network characteristics. Our approach could help network service providers to identify hidden user groups and to understand the mobility characteristics of the existing and potential customers in order to optimize the wireless networks and expand the business markets.

Public safety wireless networks (PSWNs) play a vital role in operations of emergency agencies such as police and fire departments. In this thesis, we describe analysis and modeling of traffic data collected from the Emergency Communications for Southwestern British Columbia (E-Comm) PSWN.

We analyze network and agency call traffic and find that lognormal distribution and exponential distribution are adequate for modeling call holding time and call inter-arrival time, respectively. We also describe a newly developed wide area radio network simulator, named WarnSim. We use WarnSim simulations to validate the proposed traffic model, evaluate the performance of the E-Comm network, and predict network performance in cases of traffic increase.

Ad hoc networks are self organizing networks and require no prior infrastructure, which makes them robust and quick to deploy. These are good attributes for use by emergency response and disaster recovery teams. These teams coordinate their work by conversing on a common voice channel. Unfortunately, multicast voice is not well supported using current protocols for ad hoc multi hop networks. In this project, we analyze the traffic requirements (using real traffic from a large public safety agency radio system) and investigate the issues associated with carrying this type of traffic on single hop and multi hop ad hoc networks.

The overall objective of the project is to identify evolving technologies, algorithms, protocols, and architectures for mobile ad hoc networks that are applicable, appropriate, and ready to be adopted by next generation land mobile radio systems and, in particular, by mobile radio systems designed for public safety communications applications.

Analysis of network traffic is important because a network behavior is determined by the characteristics of the traffic that it carries. In order to evaluate the effect of the network on various applications, it is essential to determine relevant traffic parameters. It is well known that network traffic exhibits phenomena known as self-similarity and long-range dependence. Self-similar and long-range dependent processes are classes of random processes characterized by a parameter, called Hurst parameter. There are several methods for estimation of the Hurst parameter of a given traffic trace, one of which is the wavelet-based method.

Wavelet estimator is considered to be a very robust and unbiased. However, this estimator yields non-physical results when applied to MPEG-1 and MPEG-4 encoded video sequences. It has already been shown that these traffic traces exhibit long-range dependence. The main objective of the project is to discover the cause for the unreliable performance of the estimator. We observed that, for the video traces, estimators of the Hurst parameter that work in the frequency domain (wavelet-based and periodogram) produce similar results. However, the estimates are always larger than those obtained using time-domain estimators (R/S and variance-time plot). These findings may suggest that the relationship between the Hurst parameter and the exponent alpha of the power-law shaped spectrum is different than the widely-accepted H=0.5(alpha+1).

We are investigating the performance of the Abry-Veitch wavelet-based estimator for the estimation of the Hurst parameter used in characterizing self-similar traffic. Performance results for two Ethernet traffic traces (pAug.TL, pOct.TL) indicate that the estimator can accurately capture Hurst parameter of the measured Ethernet traffic. In order to investigate the impact of long-range dependent (LRD) and short-range dependent (SRD) structures on the quality of the estimator, we apply the estimator on large sets of data generated by FARIMA(1,d,0) model. The performance analysis indicates that this estimator is not suitable for processes with strong SRD and either weak or strong LRD components. We confirm our findings by analyzing a medium-bursty Star Wars video traffic trace. Our findings imply that the estimator proves unsuitable for the medium- and high-burstiness video traffic because of the complex correlation structure of the video traces.

We also use the Abry-Veitch estimator to investigate the scaling behavior of packet loss in video transfer over UDP and TCP in a congested packet network. Using trace-driven ns-2 simulations and wavelet analysis, we show that the underlying transport protocols and time scales are essential for understanding packet loss behavior. In the case of UDP transfers, packet loss process exhibits LRD over time scales coarser than approximately 1 second. In contrast, for the TCP transfers, the loss behavior over a coarser time scale does not exhibit such behavior. We attribute this phenomenon to the feedback control mechanisms in TCP, which decrease the burstiness of packet loss. Our findings are robust and hold for various simulation scenarios.

We compared the performance of two wavelet based estimators: monofractal and multifractal, introduced by Abry and Veitch. The multifractal property of self-similar traffic implies that self-similarity still exists, but it is not uniform across all time scales. We use MPEG1, MPEG4, and H263 coded traffic video traces to compare the H parameter estimated using the wavelet based estimators, with the H parameter estimated via classical statistical methods, such as R/S. We are searching for the criteria for reliability of these two estimators. Our findings indicate that performance of both estimators is affected by the presence of SRD, but the effect of SRD is different for each estimator. It is interesting to note that graphical representations of the two wavelet based estimators are quite similar, while their numerical estimates of the H parameter tends to vary significantly.

Satellite networks have received much attention in traffic analysis due to its loss characteristics and high bandwidth-delay product. Based on traffic traces collected from a commercial hybrid satellite network, I plan to examine traffic data at the session and packet levels. I will focus on the comparison of TCP modifications in the deployed network with proposals from research literature. Furthermore, I plan to analyze and model captured traffic on the packet level segmented by applications.

Measurement and analysis of genuine network traffic is important for traffic characterization. Collection and characterization of the terrestrial Internet traffic has received considerable attention during the past decade. Numerous web-sites offer collected samples of Internet traffic traces. To the contrary, few traffic traces have been collected from satellite/wireless commercial sites. We plan to collect hybrid satellite-terrestrial traffic traces from a commercial Internet access provider. The project objective is to model and analyze these traffic traces and to characterize the underlying processes and distributions.

Discovering the properties of the Internet topology is crucial for the use, optimization, and maintenance of Internet. It is also essential for the improvement of network topology generators. Two recent approaches have dominated the research community: one based on data collected from active probing of hosts, and another based on Autonomous System (AS) topology information derived from Border Gateway Protocol (BGP) snapshots. Important rules, such as power-law distribution in AS graphs, have been discovered. The goal of our research is to qualify these two approaches and to find new valuable 1insights during the evaluation process.

We use data from CAIDA (Cooperative Association for Internet Data Analysis) and from Route Views project at the University of Oregon. We employ Normalized Laplacian Spectrum (NLS) from spectral graph theory, because NLS proved to be unique in AS graphs in spite of the exponential growth of the Internet, and distinctive in setting AS graphs apart from synthetic ones. By applying NLS to the two datasets, we expect to obtain plausible interpretations in networking terms and a hybrid model encompassing both structural and power-law properties. Our result may have impact on future protocol evaluations and designs.

Collection of user statistics and network traffic is crucial for understanding user behavior and for creating network workload models. It is also valuable for the management of commercial wireless networks. In this project, we report on the analysis of billing records collected from the Telus Mobility Cellular Digital Packet Data (CDPD) network. The longest continuous billing record that we examined covered approximately twenty one days, spanning the Christmas and New Year holiday seasons. We used various tools to graphically illustrate the billing data. We observed that network activities exhibit daily and weekly cycles. Analysis of billing record provided useful information about the usage of an operational wireless network.

The clustering analysis revealed that customers, as well as network cells, might be classified into few distinct behavioral classes. The clustering analysis using k-means algorithm (available in S-PLUS) revealed four distinct behavioral classes of customers and three classes of network cells as the best results. However, AutoClass clustering method provided thirty two distinct behavioral classes of customers and four classes of network cells. AutoClass proved to be a good tool for clustering small data sets, such as network cells database, which is composed of 60 cells and 5 attributes or the INDEX project database with 84 people and 29 attributes, but it is not suitable for clustering large data set such as customers behavior database with 2096 customers and 8 attributes.

The objective of this project was to use an automatic classification program (AutoClass) to extract useful information from the INDEX project (UC Berkeley) database in order to explore the demographic structure of Internet users. AutoClass is an unsupervised Bayesian classification system that seeks a maximum posterior probability classification. The database that we analyzed consisted of 84 cases with 23 attributes. AutoClass found a classification with three classes.

ICON: Intelligent Control of Communication Networks

The objective of this project is to model TCP mixed with active queue management (AQM) in order to understand and predict the dynamic behavior of packet networks. From the viewpoint of control theory, the network can be regarded as a complex control system. TCP adjusts its sending rate depending on whether or not it has detected a packet loss. Hence, it is natural to model the network system as a discrete model. We plan to model this process as a 'stroboscopic map' where the instant of observation is approximately one RTT. We are currently working on identifying the independent state variables, finding the mathematical relationships among them, and verifying them using ns-2 simulations.

The basic idea of RED (Random Early Detection) is to sense impending congestion before it happens, and to provide feedback to senders by either dropping or marking packets. The drop probability of RED can be seen as the control law of the network system. Its discontinuity is the main reason behind the occurrence of oscillations and chaos in the system. If the network can be modeled as a second order system, various bifurcation phenomena, such as period-doubling, border-collision (also named as C-bifurcation), saddle-node, Hopf and torus bifurcation, should be observable for various system parameters. We intend to study the nonlinear phenomena in the network by employing bifurcation and chaos theory. We plan to use bifurcation diagrams, strange attractors in phase plane, and the Largest Lyapunov Exponents (LLE) to investigate these phenomena.

We describe an application of the Legendre transform to communication networks. The Legendre transform applied to max-plus algebra linear systems corresponds to the Fourier transform applied to conventional linear systems. Hence, it is a powerful tool that can be applied to max-plus linear systems and their identification. Linear max-plus algebra has been already used to describe simple data communication networks. We first extend the Legendre transform as the slope transform to non-concave/non-convex functions. We then use it to analyze a simple communication network. We also propose an identification method for its transfer characteristic, and we confirm the results using the ns-2 network simulator.

Internet is a transport infrastructure for applications with various service requirements. However, Internet remains to be a best-effort network without widely deployed mechanisms for service differentiation and quality of service provisioning. Research efforts to provide service differentiation in the Internet have been recently directed toward non-elevated mechanisms. Majority of proposed non-elevated mechanisms rely on the idea of providing low delay service at the expense of increased loss probability. However, these proposals do not consider the influence of delay and loss differentiation on the behavior of TCP, the most widely used transport protocol in today's Internet. Service differentiation mechanisms cannot be designed without taking into account the complexity of TCP's congestion control algorithm.

Goal of this project is to design a new non-elevated service differentiation mechanism that would provide low-delay service to real-time applications and at least the same throughput to throughput sensitive applications as they would receive in a best-effort network. The new mechanism will include two building blocks: a scheduler for proportional delay differentiation and a controller that will ensure that performance of TCP applications will not be degraded by the presence of the low-delay traffic.Necessary conditions to provide desired service differentiation will be derived by considering delay differentiation algorithm and TCP's congestion avoidance algorithm as a feedback control system. We plan to use the ns-2 simulator to test the new mechanism with existing ns-2 traffic models (CBR, Pareto ON/OFF, and Exponential ON/OFF) and protocols (UDP and TCP NewReno).

We are investigating the patterns of consecutive packet losses (called loss episodes or loss bursts), and the loss behavior in video transfers over UDP and TCP. We consider the impact of loss behavior on high-quality video transfers with stringent end-to-end delay requirements (5 - 30 milliseconds). These requirements imply small maximum queuing delay for the packets in the buffers of the routers. This, in turn, implies that these buffers should be small, which increases the probability of loss and the lengths of the loss episodes.

We use ns-2 network simulator and genuine traffic traces to obtain the packet loss data and to observe packet loss patterns. We simulate both per-flow loss of a video connection and the aggregate loss at a buffer of a router. We experiment with a variety of simulation scenarios and we consider simple and complex network topologies, User Datagram (UDP) and Transmission Control (TCP) Protocols, various router buffer sizes and utilization levels, and a choice of queue management techniques (Droptail).

Our simulation results provide a quantitative measure of how the length of these loss episodes increases with the increase of the average utilization levels. They show that lengthy loss episodes contribute significantly to the overall loss patterns, and that under fixed average utilization and during the times of higher congestion, longer loss episodes and shorter loss episode distances occur.

We investigated the impact of traffic patterns on wireless data networks. By performing simulations driven by genuine traffic traces, we evaluated the performance of wireless Cellular Digital Packet Data (CDPD) networks. OPNET network simulation tool was used to simulate the CDPD network of a local commercial service provider (Telus Mobility). In our simulations, we used traffic traces collected from the Telus Mobility network. Statistical analysis of these traces revealed that they exhibit long-range dependent behavior. Our simulation results indicated that they produce longer queues and, thus require larger buffers in the deployed network's switching elements.

SIMON: Simulation of Networks

The Border Gateway Protocol, BGP, is a de facto inter-Autonomous Systems (ASs) routing protocol. The primary function of a BGP speaking system is to exchange network reachability information with other BGP systems. It was formally proposed in Request For Comments (RFC) 1771 by the network working group within the Internet Engineering Task Force (IETF). The Internet has a very dynamic nature and this has an effect on the performance of the routing protocols such as BGP. Therefore, ns-BGP was developed for the ns-2 network simulator by importing the code from BGP implementation in SSFNET and converting them to C++ and OTcl code in 2003. Later, the ns-BGP upgraded to be compatible with the latest version of ns, which was ns-2.33 in 2008. This project integrates ns-BGP for the latest stable ns release ns-2.34.

The border gateway protocol (BGP) is an inter-Autonomous System (AS) routing protocol utilized as the core routing protocol in the Internet today. It was formally proposed in Request For Comments (RFC) 1771 by the network working group within the Internet Engineering Task Force (IETF). Primarily, BGP exchanges network reachability information with other BGP systems. Since BGP performance is affected by the dynamic nature of the Internet, ns-BGP was developed for the ns-2 network simulator in 2003 to facilitate realistic, flexible BGP routing experimentation.

In parallel with the ns-BGP development, academic and research communities continued to develop ns-2. Consequently, this led to an incompatible ns-BGP module with current versions of the simulator. Therefore, in an effort to aid further BGP research efforts, this project will integrate ns-BGP with the latest stable version of the simulator thereby benefitting from core ns-2 feature enhancements and maintenance updates over the past five years.

Long propagation delays and high bit error rates in heterogeneous networks with geostationary earth orbit (GEO) satellite links have negative impact on the performance of Transmission Control Protocol (TCP). In this paper, we propose modifications to TCP by introducing adaptive delay and loss response (TCP-ADaLR) to mitigate the adverse effects of satellite link characteristics. The proposed modifications incorporate delayed acknowledgment (ACK) recommended for Internet hosts. TCP-ADaLR introduces adaptive window increase and loss recovery mechanisms to address TCP performance degradation in satellite networks. We evaluate and compare the performance of TCP-ADaLR, TCP SACK, and TCP NewReno, with delayed ACK enabled and disabled. In the absence of losses, TCP-ADaLR exhibits the shortest user-perceived latency for HTTP and FTP applications. In the presence of only congestion losses, TCP-ADaLR shows comparable performance to TCP SACK and TCP NewReno. In the presence of only error losses, TCP-ADaLR exhibits improvements up to 61% and 76% in throughput and utilization, respectively. In the presence of both congestion and error losses, TCP-ADaLR exhibits goodput and throughput improvements up to 43%. TCP-ADaLR exhibits the best performance in the absence of losses and in the presence of losses due to both congestion and errors. It also friendly to TCP NewReno, exhibits better fairness, and maintains TCP end-to-end semantics.

In this project, we propose the M-TCP+ algorithm for heterogeneous wired/wireless networks. The algorithm is a modification of M-TCP that was proposed for deployment in mobile cellular networks. It is recommended that Internet hosts enable the delayed acknowledgement (delayed ACK) option to maximize network bandwidth by reducing the number of ACKs sent to a TCP sender by a TCP receiver. The M-TCP+ algorithm performs best when the TCP delayed ACK option is enabled. The algorithm relies on feedback sent from a wireless host in anticipation of disconnections. We compare the performance of the M-TCP+ algorithm with the performance of M-TCP, TCP NewReno, and TCP SACK in both the absence and the presence of disconnections for a file transfer protocol (download) application. We also simulate network scenarios with traffic congestion. The M-TCP+ algorithm performance is evaluated in terms of file download response time, goodput, and retransmission ratio with and without the delayed ACK option. In scenarios without disconnections, the M-TCP+ algorithm does not introduce significant processing delay. Furthermore, in scenarios with disconnections, the M-TCP+ algorithm shows 2%-15% performance improvement.

We examine the behaviour of the Gnutella peer-to-peer file sharing network and propose a protocol modification intended to improve its performance. Because its overlay topology is not well matched to the underlying physical network, Gnutella exhibits sub-optimal performance in terms of message latency. In order to characterize this performance, we modified an existing Gnutella simulation framework developed for network simulator (ns-2) to gather information about query and query hit propagation. We then modified the protocol implemented in the simulation to use the Vivaldi synthetic coordinate system and to bias neighbour selection to favour nodes that are ``close" in the Euclidian sense. Simulations with the adapted Gnutella protocol showed an improvement in query and query hit propagation times.

Route flap damping (RFD) plays an important role in maintaining the stability of the Internet routing system. It functions by suppressing routes that persistently flap. Several existing algorithms address the issue of identifying and penalizing route flaps. In this project, we compare three such algorithms: original RFD, selective RFD, and RFD+. We implement these algorithms in ns-2 and evaluate their performance. We also propose possible improvements to the RFD+ algorithm.

The duration of the Minimal Route Advertisement Interval (MRAI) and the implementation of MRAI timers have a significant influence on BGP convergence time. Previous studies have reported existence of optimal MRAI values that minimize BGP convergence time for various network topologies. These optimal values depend on network topologies and traffic loads. In this project, we propose using adaptive MRAIs for each destination in BGP speakers. Furthermore, we introduce reusable MRAI timers that independently limit the number of advertisements of various destinations. The proposed modification of BGP is named BGP with adaptive MRAI. We evaluate the new algorithm by introducing a new model for the BGP processing delay. ns-2 simulation results demonstrate that BGP with adaptive MRAI results in a considerably shorter BGP convergence time, with a similar number of update messages compared to the current BGP implementation. Furthermore, BGP convergence time depends linearly on BGP processing delay. For large networks, BGP with adaptive MRAI may reduce BGP convergence time by 80% and the number of update messages by 20%.

Border Gateway Protocol (BGP) is the inter-domain routing protocol currently employed in Internet. Internet growth imposes increased requirements on BGP performance. Recent studies revealed that performance degradations in BGP are due to the highly dynamic nature of the Internet. Undesirable properties of BGP, such as poor integrity, slow convergence, and divergence, have been reported by the research community. Theoretical analysis and empirical measurements have been employed in the past, albeit with certain limitations. Simulations allow more realistic experiments with fewer simplifications than the theoretical approach and with enhanced flexibility than empirical studies permit.

In this thesis, we describe the design and implementation of a BGP-4 model (ns-BGP) in the network simulator ns-2 by porting the BGP-4 implementation from SSFNet. The ns-BGP node is based on the existing ns-2 unicast node and the SSF.OS.BGP4 model from SSFNet. In order to provide socket support and at the same time maintain the structure of SSF.OS.BGP4, we also ported to ns-2 TcpSocket, the socket layer implementation of SSFNet. In order to support the IPv4 addressing and packet forwarding, the basic address classifier was replaced with a new address classifier in ns-2 named IPv4Classifier. We also modified FullTcpAgent, the TCP agent used by TcpSocket, to support user data transmission.

We performed a suit of validation tests to ensure that the ns-BGP model complies with the BGP-4 specifications, including BGP-4 features such as: basic peer session management (keep and drop peer), route selection, reconnection, internal BGP (iBGP), and route reflection. Finally, in the scalability analysis of ns-BGP, we showed that the model scales with respect to the number of peer sessions and size of routing tables.

One of the main reasons for TCP's degraded performance in wireless networks is TCP's interpretation that packet loss is caused by congestion. However, in wireless networks, packet loss occurs mostly due to high bit error rate, packet corruption, or link failure. TCP performance in wired/wireless networks may be substantially improved if the cause of packet loss could be distinguished and appropriate rectifying measures taken dynamically. We propose a new end-to-end TCP protocol named Selective-TCP, which distinguishes between congestion and wireless link transmission losses (high bit error rate and/or packet corruption). When detecting packet loss, Selective-TCP invokes correction mechanisms. It is suited for mixed wired/wireless networks and shows increase in goodput when compared to TCP NewReno.

We propose packet control algorithms to be deployed in intermediate network routers. They improve TCP performance in wireless networks with packet delay variations and long sudden packet delays. The ns-2 simulation results show that the proposed algorithms reduce the adverse effect of spurious fast retransmits and timeouts and greatly improve the goodput compared to the performance of TCP Reno. The TCP goodput was improved by ~30% in wireless networks with 1% packet loss. TCP performance was also improved in cases of long sudden delays. These improvements highly depend on the wireless link characteristics.

GPRS is a Global System for Mobile Communications (GSM) based packet switched wireless network technology deployed around the world. The main components of a GPRS system are: Mobile Station (MS),Base Station Subsystem (BSS), Serving GPRS Support Node (SGSN), Home Location Register (HLR), and Gateway GPRS Support Node (GGSN).

In this project, we plan to enhance the current GPRS OPNET model by implementing the Medium Access Control/Radio Link Control (MAC/RLC) layer in the MS and the BSS for contention resolution and the BSS GPRS protocol (BSSGP) in the BSS and the SGSN for exchanging QoS related information. We also plan to evaluate the performance of the GPRS network using the enhanced model.

We plan to model the signaling behavior of the SGSN system and create a model using the OPNET simulation tool.

The Serving General Packet Radio Services Support Node (SGSN) signaling plane model is capable of handling various signaling traffic profiles such as attach, activation, and deactivation. The model can also generate statistics related to the system's performance. We also plan to simulate the user data procedure in order to illustrate various class of GPRS Quality of Service subscribed by Mobile Station (MS) that will lead to different end-to-end delays in the data session. The SGSN model can then be enhanced to incorporate parameters extracted from lab measurements in order to monitor the performance of the real SGSN system. The SGSN model will not only provides a flexible environment to collect a wide range of data, but will also serve as a performance predication and evaluation tool.

The objective of this project is to examine:
1. Different priorities defined in the Internet Protocol (IP) with emphasis on IPv6, and the importance of traffic engineering based on assigning priorities to packets.
2. Limitations of Multi-Protocol Label Switching (MPLS) protocol in prioritized IP traffic, and
3. Enhancement of the protocol so that Internet Serviced Providers (ISP) using MPLS can improve their network utilization when carrying prioritized IP traffic, can provide better Quality of Services (QoS), and can offer more Classes of Services (CoS).

We used ns-2 to simulate network performance.

Unlike wired networks that can provide large bandwidth, the bandwidth of wireless local area networks (WLANs) is rather limited because they rely on an inexpensive, but error prone, physical medium (air). Hence, it is important to improve their loss performance.

In this paper, we investigate several methods for improving the performance of WLANs. We survey the current research literature dealing with improving performance on various wireless network layers. We describe OPNET implementations of three approaches: tuning the physical layer related parameters, tuning the IEEE 802.11 parameters, and using an enhanced link layer (media access control) protocol. Finally, we describe several simulation scenarios and present simulation results that demonstrate the effectiveness of the three approaches.

Mobile Internet Protocol has been proposed by IETF to support portable IP addresses for mobile devices that often change their network access points to the Internet. In the basic mobile IP protocol, datagrams sent from wired or wireless hosts and destined for the mobile host that is away from home, have to be routed through the home agent. Nevertheless, datagrams sent from mobile hosts to wired hosts can be routed directly. This asymmetric routing, called ``triangle routing,'' is often far from optimal and ``route optimization'' has been proposed to address this problem. In this paper, we present the implementation of ``route optimization'' extension to mobile IP in the \ns\ simulator. We illustrate simulations of the mobile IP with route optimization with simulation scenarios, parameters, and simulations results.

We use OPNET simulation tool to model the VirtualClock scheduling mechanism. The algorithm is implemented in the output buffers of the IP router objects in OPNET. The model is then incorporated in the IP layer of the network layer hierarchy so that it can communicate with upper and lower network layer objects.

We compared the performance of the VirtualClock algorithm and several other scheduling mechanisms in packet networks, such as Weighted Fair Queuing (WFQ), Custom Queuing (CQ), and Priority Queuing (PQ). The performance was compared in terms of fairness, packet end-to-end delay, and the number of packet loss during various time periods. We also simulated the effect of these algorithms on the performance of several Internet applications, such as HTTP, FTP, IP Telephony, and Video Conferencing.

The main objective of our research is to simulate the quality of service (QoS) parameters in Internet Protocol (IP) networks with various types of traffic sources. The main QoS parameters of interest are packet loss due to buffer overflow, and packet delay due to queuing in the buffers.

We use network simulator ns-2 to perform trace driven simulations using simple network topologies. We employ MPEG-1 video traces to generate traffic that was transmitted over the User Datagram Protocol (UDP). Our simulation scenarios are used to investigate how various queuing mechanisms affect the characteristics of the QoS parameters. We simulated FIFO (First In First Out) buffer with a DropTail queue management policy, Random Early Drop (RED), Fair Queuing (FQ), Stochastic Fair Queuing (SFQ), and Deficit Round Robin (DRR) active queue management schemes.

As part of our undergraduate thesis project, we are working on enhancing the OPNET model of the Cellular Digital Packet Data (CDPD) Medium Access Control (MAC) Layer. The enhancements include handling burst uplink data transmission, competing mobile stations, and collision detection. We use the OPNET model to examine the system's queuing behavior in terms of buffer requirements and packet delays. In addition to various standard traffic source models, such as Poisson, bursty (on-off) and self-similar generators, in our simulations we also employ genuine traffic traces.

The main objectives of this project is to implement an OPNET model of the Deficit Round Robin scheduling algorithm, and to compare its performance to other scheduling mechanisms in packet networks. Deficit Round Robin enables handling of variable packet sizes, without knowing the average packet size of the flows. The OPNET simulation tool enables modeling of communication networks and distributed systems. It contains tools for design simulation, data collection, and data analysis.

Various projects

We have built an ATM testbed comprised of two ATM edge switches (Newbridge MainStreet 36150), and two Pentium III workstations connected to the ATM network via Ethernet cards. We used MBone and NetMeeting multimedia-conferencing systems to measure and evaluate performance of audio and video transmissions using both CBR and VBR services in an IP over ATM network. Using Spirent's SmartBits load generator, and in compliance with RFC 2544, we measured and analyzed throughput, packet delay, and delay jitter as main parameters for measuring forwarding performance and quality of service in multimedia applications.

The ATM Traffic Monitor script, a simple network management graphical user interface written in Tcl, Tk, and Expect scripting languages, provided an easy graphical capture of the aggregate traffic sent through Ethernet cards of the ATM switches. We also used MBone to multicast the Open Forum session at IFSA/NAFIPS 2001 conference, held in Vancouver on July 25-28, 2001. Audio and video signals were sent using MBone tools (running on Windows OS) to the MBone network using DVMRP tunneling through ADSL (Telus) line, via SFU campus network, to BCnet GigaPOP.

This project deals with the CDMA technology and the added benefits provided by the 3G CDMA2000. We describe the testing methodology within a software engineering model, and differentiate the field test from other testing activities. Field test is an important and high priority task in R&D projects and it is mandatory to include field tests in any handset development project. We elaborate on the details of field test activities, which include test planning, test tools application, test sites selection, test case design, and test results analysis. We also illustrate the typical field test analysis required for R&D field tests. In particular, our field test results confirm that the message flow complies with the IS-2000.5 Upper Layer (L3) Signaling Standard.

In this project we investigate the characteristics of Internet Protocol (IP) addressing. We first review the similarities and differences between the Variable Length Subnet Mask (VLSM) and Classless Inter-Domain Routing (CIDR). Moreover, we also consider the advantages of the classless over the classful nature of a routing protocol. We discuss the compositions of routing tables. We examine in details the Routing Information Protocol (RIP) and the Open Shortest Path First (OSPF) routing protocols.

We performed experiments involving seven Cisco routers. The three cases of interests are:
- impact of a failure Ethernet link to the OSPF convergence
- impact of a broken Frame Relay (FR) Virtual Circuit (VC) to the RIP convergence
- impact of a broken FR VC to the redistribution convergence.

The RIP's timers are changed in each of these three cases to measure performance improvements.

We are working on a hardware implementation of a crossbar packet switch for high-speed data networks. We use VHDL to describe our design, the ALTERA MAX+PLUS II tools to simulate it, and the FLEX 10KE ALTERA FPGA to implement it on a chip. The switch has 8 input and 8 output ports, with input queuing. The switch is capable of handling fixed sized packets, such as ATM cells.

The switch architecture consists of input buffers, input port controllers, destination look-up tables, a centralized scheduler, and a crossbar fabric. The packets first arrive (ingress) to the input ports of the switch. There, serial data is shifted into a serial shift register. As soon as a byte of data is received, it is loaded into a FIFO queue. Each input port has a controller that queues the header of each packet in a separate FIFO buffer, extracts address information from the header of the packet, and sends it to a programmable look-up table (LUT). The LUT returns a destination port address. Based on this address, the controller sends a request to the centralized scheduler. The scheduler receives requests from all input ports and grants them based on a simple two-dimensional ripple-carry arbiter architecture called Rectilinear Propagation Arbiter (RPA). A ``round robin'' priority scheme ensures fairness to all the input ports. Once a grant is issued, the crossbar fabric is configured to map the input port that received a grant to its output port destination. The outgoing (egress) packets are stored into another shift register and then shifted serially to the output link.