TELCOM 2120Network Performance Fall 2011

TELCOM 2120Network Performance Fall 2011

Course Project Descriptions

The TELCOM2120project is to be completed no later than Friday.December 16.The project can be an individual or a group (3or less) effort covering anaspect of network performance in depth. The basic thrust of the project isto complete a mini-performance study utilizing one or more of theperformance techniques (i.e., measurements, queueing, or simulation) discussed inclass. A variety of projects are possible and each group should turn in aFINAL REPORT documenting the project.The FINAL Report should contain the following sections:

(a) introduction - giving background on the project;

(b) goals/problem statement - explaining the purpose of the study;

(c) results - detailing the methods used and the results; and

(d) conclusions.

Some project suggestions are given below – please feel free to suggest a project of personal interest.

I. Measurement Oriented Projects

1. Ethernet QoS Parameter Setting
The use of commercial off the shelf technology to replace point to point industrial cabling in a variety of industrial applications is being advocated. A major problem is the loss of a deterministic delay when using Ethernet. Construct a small testbed to see how parameter settings in Ethernet can provide almost deterministic service delay (< 10 msec) to periodic traffic emulating a sensor or crucial real time data. Examine the effect of bursty background traffic on the delay.

2. WiFi QoS parameter setting

As in project 1, WiFi technology has been advocated for replacement of point to point cabling in industrial settings. A major problem is the loss of a deterministic delay when using Ethernet. Construct a small testbed to see how parameter settings in Ethernet can provide almost deterministic service delay (< 10 msec) to periodic traffic emulating a sensor or crucial real time data.

2. Multimedia over WAN project
Benchmark the performance of remote webcam video traffic. Consider the effects of packet loss vs. quality, delay and delay jitter vs. quality. Measure the quality both objectivelyand subjectively (i.e., Mean opinion score) and try to determine which network performance metric results in the most adverse effect on the quality. Use the NIST NET emulator to serve as the WAN cloud.

3. Experimental Evaluation of Bandwidth Estimation tools in an Ad Hoc Wireless Network

Set up a single hop wireless link and measure the bandwidth using 2 or more estimation tools such as Pathload, PChirp, etc. Vary the distance and interference. Repeat for a multi-hop scenario.

4. MPLS FFR Benchmarking experiment

Consider the IETF draft on benchmarking fault protection schemes in MPLS (e.g., FRR with link protection),

and implement two of the tests in the lab benchmarking the actual failure recovery time..

5. SIP Overload project

Several SIP overload schemes have been proposed in the literature and standards bodies (e.g., IETF draft Implement one scheme and benchmark it’s performance with either measurements or simulation.

6. Temporal variations of a static mesh network's connectivity graph

The goal of the study is to examine the temporal behavior of the connectivity graph of a wireless mesh network. In particular, we will study the change in graph metrics, such as average node degree, diameter of the graph, degree distribution etc., with respect to time. This will help us identify "regular" patterns on the connectivity graph, and thus, it can be later used for anomaly detection (e.g., related with security threats on routing etc.).

The key point is the definition of connectivity between two nodes. A simple definition could be derived using "ping" tests. If two nodes can exchange ICMP_ECHO_REQUEST packets with a small drop rate (e.g., < 5%) the two nodes are considered to be connected. However, the MAC layer retransmissions can lead us towards erroneous results, in the sense that they underlying link between the two nodes might have a higher Packet Error Rate (PER) than the one identified with the ping test- this is a general problem with PER measurements. So a different connectivity can be determined with throughput tests. Two nodes are assumed to be connected if the saturated UDP throughput achieved is higher than a threshold X Mbps (e.g., X = 5).

In order to study the temporal characteristics of the connectivity graph we will need to have fairly regular measurements. This will enable us to examine the granularity of changes. For example 10 minutes intervals would be ideal; this would require 144 experimental sets every day. Additionally, at least 1 week of daily measurements would be required.The ability to access and run scripts on the machines remotely will help the measurement study tremendously. The measurementsthemselves require the use of "ping" (for the first - type of definition for connectivity) and the use of a throughput measurement tool, such as iperf. Some shell scripting is also required in order to automatically schedule the experiments.

II. Simulation Oriented Projects

1. TCP/IP performance over WiMAX Evaluation

Using the NIST 802.16 extension to the NS2 simulator compare three versions of IP which have been modified to cope with the characteristics of wireless link, namely: Snoop TCP, TCP with ELN, TCP with SACK. Compare the time to download various size files (e.g., 15Kb, 150 Kb, 1.5 Mb) using the schemes over a wireless link with a high bit error rate.

2. Analysis of Active Queue Management

Using the NS2 simulator compare three IP packet dropping policies, such as RED, SRED and BLUE, etc. A tutorial paper on IP packet dropping policies can be obtained from the instructor.

3.Ad-hoc network routing protocols

Using Opnet and the NIST contributed DSR model simulate an 8 node network using random waypoint movement and study the effects of the mobility model parameters on the throughput and delay.

4. Comparison of Packet Scheduling Schemes

Use OPNET or ns to investigate three basic queuing mechanisms: FIFO, CSFQ Priority Queuing, and Weighted Fair Queuing. A simulation model consists of a single server accepting traffic streams from three applications requiring different levels of services (e.g., FTP , web, and video application). Observe the effects of queuing mechanisms and its parameters (e.g., weight) on queuing delay of each application.

5. Evaluation of Equivalent Telephony Erlangs

The concept of mapping data traffic sources (e.g. web browsing, fax, video) into a equivalent telephony erlang (ETE) load so that the standard Erlang formulas can be adopted for trunk/bandwidth sizing and grade of service estimation has been advocated by the ITU. Using an Opnet simulation evaluate the accuracy of this approximation for various traffic sources.

III. Queuing Projects

1Ad-hoc network Performance Modeling.

Using fluid-flow models of the ad-hoc network nodes (see the instructor for details), develop a matlab program to numerically solve the fluid-flow model and predict the delay performance of a 4 node ad-hoc network with random movement.

2. Range of queueing models

Judge the effectiveness of using steady state queueing models when the load is time varying. Specifically construct a simulation model of particular queues (M/G/1) and vary the load with time – compare the results with steady state approximation approaches such as PSA and MOL.

(1/3)