Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Chapter 4 TCP/IP Network Simulation
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Objectives Appreciate the role of simulation in performance evaluation of TCP/IP networks Acquire the knowledge needed to conduct steady state simulation Master basic skills for analyzing confidence level Describe types of simulation tools available Familiarize with the capabilities of popular simulation tools
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Contents Why use simulation Systematic simulation study Types of simulations Simulation validation and verification Confidence level of simulation results Simulation with self similar traffic Simulation tools
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Why Use Simulation
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Why Use Simulation Predict performance for proposed network Allow performance evaluation under a wide variety of network conditions Compare alternative architectures under identical and repeatable conditions Produce results closer to reality Validate analytical results
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Systematic Simulation Study
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Systematic Simulation Study Pre-software stage Define problem/objective Design network model and select fixed parameters Select performance metrics Select variable parameters
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Systematic Simulation Study (Cont.) Software stage Model construction Simulation configuration Simulation execution/Data collection Result presentation
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Types of Simulation
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Types of simulations Continuos vs. discrete event Terminating vs. steady state Synthetic vs. trace-driven
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Steady State If we are interested in asymptotic behavior of a network system, we cannot use terminating simulations Must continue until it reaches steady state
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Trace-Driven Simulations Actual network traces can be used as simulation input Results can be more convincing
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Validation and Verification
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Simulation Validation and Verification Validation: Make sure that the assumptions are realistic Verification: Make sure that the model implements assumptions correctly Guidelines to follow Look for “surprise” in output Employ analytical modeling Compare with real network data
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Confidence Level Analysis
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Confidence Level Relative precision formula for 95% confidence (see Eq. 4.8, pp. 84) Confidence level in terminating simulation GRepeat the entire simulation many times with different random numbers (or seeds) Gp105, Fig. 4.4
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Confidence Level (cont.) Confidence level in steady-state simulation GFixed length simulation GAdaptive length simulation
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Self Similar Traffic
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Self Similar Traffic Poisson model does not capture the burstiness of TCP/IP traffic TCP/IP traffic usually exhibits self similar property Generated by superimposing many ON/OFF sources with Pareto distribution
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Simulation Tools
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Classification of Simulation Tools GPPL: General Purpose Programming Language PSL: “Plain” Simulation Language SP: Simulation Package p110, Fig. 4.7
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain NS Simulator Developed by UC Berkeley Public domain SP Object-oriented Written in C++ and object-oriented tcl (Otcl) Network components are represented by classes
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Ns class hierarchy
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Reference network for TCP simulation with background traffic
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Sequence No. vs. time
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Average throughput vs buffer size
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Single FTP with trace-driven “star wars movie” internet traffic
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain TCP throughput vs time
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain TCP packet delay vs time
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain OPNET Developed by OPNET Technologies Inc. Commercial SP Object-oriented Totally menu-driven package Built-in model libraries contain most popular protocols and applications Simulation task made easy
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Modeling hierarchy in OPNET
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Screen shot of a network-level modeling in OPNET
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Screen shot of a node-level modeling in OPNET
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Screen shot of a process-level modeling in OPNET
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Editing pad of state executives for process modeling in OPNET
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Link editor dialog box to set parameters of a point to point link object
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Packet editor to specify or edit details of TCP header fields
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Packet Editor to specify or edit details of TCP header
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Probe Editor to collect seven different types of statistics
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Simulation Tool showing two different simulation sequences
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Main menu of Analysis Tool
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Network model for RED gateway
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Evolution of buffer occupancy under FIFO and RED
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Network model for fairness evolution
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Average TCP/UDP throughput under FIFO
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Average TCP/UDP throughput under RED
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Average TCP/UDP throughput under WFQ
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Selecting the Right Tool Built-in libraries Credibility User-Friendliness Technical support Level of Details Resource consumption Cost
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain NS vs. OPNET Both have emerged as de facto “standard” for simulating TCP/IP networks P143, Table 4.6
Prentice HallHigh Performance TCP/IP Networking, Hassan-Jain Packet delay vs link capacity