EE/CpE 345. Modeling and Simulation. Fall Class 4 September 23, 2002

Transcription

1 EE/CpE 345 Modeling and Simulation Class 4 September 23, 2002

2 This week s topics Simulation software Continuation of wireless simulation example

3 Simulation Software General purpose programming languages FORTRAN C/C++ Simulation programming languages GPSS/H SIMIAN V Simulation environments OpNet Matlab Simulink

4 Evolution of simulation software FORTRAN Simscript GASP SLAM II SIMIAN ALGOL PASCAL C Simula C++ Matlab GPSS GPSS/H CSIM Simulink Arena AutoMod OpNet General purpose Languages Simulation languagess Simulation environments

5 Some Guidelines on Simulation Software Selection Consider multiple broad issues: Ease of use, accuracy, support, training, applicability to problem Execution speed is important: slow speed during debugging can impede development Caveat Emptor vendors sometimes lie Try before you buy run on sample problems. Evaluation downloads are extremely important (e.g., MathWorks offers complete 30 day evaluation copy) Beware of simple feature lists Extending package with your own C (et. al.) routines is extremely important Graphical interface is easiest for describing process flow. Programming language-like routines are best for describing procedural steps. You ll need both. Sophisticated graphical display of complex output is essential (e.g., 3-D graphs)

6 Model-building features Feature Model world view Input data analysis Graphical model building Conditional routing Simulation programming Syntax Input flexibility Conciseness Randomness Specialized components User-built custom objects Interface with GPL Function Process interaction, events Estimate distributions from raw data Block diagrams to illustrate process flow, functional blocks, network Route entities thru simulation based on state of system Interface to allow programmatic control of simulation, preferably in standard HLL Easily understood, consistent, unambiguous, similar to a language user already understands Accepts data from external sources, e.g., sensors, files, user Powerful actions, blocks, or nodes Random process generators for common distributions. Ability to randomize or repeat same sequence Off-the-shelf components for your particular problem Reusable objects, templates, and sub-models. Ideally, they should be indistinguishable from built-in features E.g., model is C-callable, Model can include C routines

7 Runtime Environment Feature Execution speed Model size Interactive debugger Model status and statistics Runtime license Function Many short runs are needed during debugging. Long runs are needed during experimentation. Fast startup as well as efficient execution are required. Ability to compile/optimize/profile model needed There should be no hard limits, other than those imposed by platform you are running on (e.g., available RAM, hard disk space) Breakpoints, trap event, run until condition, single step, error recovery Display at any time during simulation Ability to design a model and then run it with same structure but different parameters on multiple platforms simultaneously. At $30k per license for some packages, you don t want to have to buy 10 licenses to run 10 instances of same simulation

8 Simulation example Example 4.1: Grocery Checkout Single server queue Run until 1000 customers have been served Customer arrival times: exponentially distributed with 4.5 minute average interarrival time Service time: Normally distributed with mean 3.2 minutes, standard deviation.6 minutes Simulate with C++, GPSS/H, CSIM

9 Simulation example in C++ // declarations of types and variables class Event { friend bool operator<(const Event& e1, const Event& e2); friend bool operation==(const Event& e1, const Event& e2); public: Event(); Enum EvtType { arrival, departure; Event(EvtType type, double etime) : _type(type), _etime(etime){ EvtType get_type() { return _type; double get_time() { return _etime; protected: EvtType _type; double _etime; priority_queue<event> FutureEventList; queue<event> Customers; // declaration of global variables // main routine

10 Simulation example in C++ // declarations of types and variables main() { // start simulation // get input parameters Initialization(); while(numberofdepartures < TotalCustomers) { Event evt = FutureEventList.top(); // get imminent event FutureEventList.pop(); // remove from queue Clock = evt.get_time(); // advance clock if( evt.get_type() == Event::arrival ) ProcessArrival(evt); else ProcessDeparture(evt); ReportGeneration();

11 Simulation example in C++ void Initialization() { // initialize global variables Clock = 0; //... // create first arrival event Event evt(event::arrival, exponential(meaninterarrivaltime)); FutureEventList.push(evt);

12 Simulation example in C++ void ProcessArrival(Event evt) { Customers.push(evt); // put on Customers queue QueueLength++; // increment number waiting // if idle, fetch event, update stats, put in service if (NumberInService == 0) ScheduleDeparture(); else TotalBusy += (Clock + LastEventTime); // adjust max queue length statistics if(maxqueuelength < QueueLength) MaxQueuelength = QueueLength; // schedule next arrival Event next_arrival(event::arrival, Clock+exponential(MeanIntervalArrivalTime)); FutureEventList.push(next_arrival); LastEventTime = Clock;

13 Simulation example in C++ double exponential(double mean) { return -mean*log( unif() ); double SaveNormal; int NumNormals = 0; #define PI double normal(double mean, double sigma) { //normal variates are generated in pairs. You return one, then the other double ReturnNormal; if(numnormals == 0) { double r1 = unif(); double r2 = unif(); ReturnNormal = sqrt(-2*log(r1))*cos(2*pi*r2); SaveNormal = sqrt(-2*log(r1))*sin(2*pi*r2); NumNormals=1; else { NumNormals=0; ReturnNormal = SaveNormal; return ReturnNormal*sigma + mean;

14 Simulation in GPSS SIMULATE * Define variables INTEGER &LIMIT REAL &IAT, &MEAN, &STDEV, &COUNT LET &IAT = * Copy input data to file, remember GPSS is FORTRAN based, so this follows the FORMAT command in FORTRAN PUTPIC FILE=OUT, LINES=5, (&IAT, &MEAN, &STDEV, &LIMIT) Mean interarrival time **.** minutes Mean service time **.** minutes... * GPSS/H Block Section GENERATE RVEXPO(1,$IAT) Exponential arrivals QUEUE SYSTIME QUEUE LINE Customer joins line SIEZE CHECKOUT begin checkout DEPART LINE ADVANCE RVNORM(1,&MEAN, &STDEV) service time RELEASE CHECKOUT End response time DEPART SYSTIME TEST GE M1, 4, TER is response time >= 4 minutes? BLET &COUNT=&COUNT-1 if so, add 1 to customer TER TERMINATE 1 START &LIMIT Run until LIMIT reached * Write output to file END

15 Simulation in CSIM event done( done ); //define event named done facility( f ); //define facility f // define variables long TotalCustomers; long NumberOfCustomers=0;... void customer(); //forward definition extern C void sim() //main process { create( sim ); // procedure is a process MeanServiceTime = 3.2; //initialize variables for(long i=1; i<=totalcustomers;i++) {// hold for an interarrival period hold(exponential(meaninterarrivaltime)); customer(): // generate a new customer process done.wait(); //wait for last customer ReportGeneration(); //generate a report

16 Simulation in CSIM void customer() { create( customer ); //update MaxQueueLength long length = f.qlength() + f.num_busy(); if( length && MaxQueueLength < length) MaxQueueLength = length; // sample service time, update LongService double service; while( (service = normal(meanservicetime, SIGMA) ) < 0); f.use(service); //queue up for service if( service > 4.0 ) LongService++; if(++numberofcustomers == TotalCustomers) //done? done.set(); // yes, signal main

17 Some Conclusions to Draw C and C++ weren t designed for simulation, but if you take care of the details, they can serve well GPSS is a language written specifically for simulation, but it is based on old concepts (e.g., not a procedural oriented language like Algol, C, etc.) Starting with C++, simple extensions can be added with objects and operator overloading to create an efficient simulation language, like CSIM

18 Common characteristics: GUI based, Animated graphics Automatic data collection Simulation Packages Arena modeling of business processes built on SIMIAN AutoMod focus on material handling and manufacturing processes generated AVI movies of 3D animations, with pan/zoom Deneb/QUEST robotic simulation OpNet Modeler/IT Guru graphical modeling of complex networks Matlab/SIMULINK block diagram focus focus on scientific/technical applications rich set of Blocksets/Toolboxes MathCAD equation-based worksheets includes symbolic programming (e.g., simplification/expansion of equations)

19 Trends in Simulation Packages High-fidelity simulation High-accuracy simulation of complex systems Data exchange standards Simulation input/output can be interfaced to other packages Distributed (client/server) computing support Large organization/wide-area collaboration (e.g., across LAN, Internet) General purpose simulations vs. specialized simulations Do it once, make it reusable Richer object libraries/reusable blocksets Multiple computer simulations to accelerate simulations

20 A Simulink Model User and Node Positions

21 Implementation Directions Top Down Define high level structure first, fill in details Nothing is working until the details are done Bottom Up Define the details first, stitch them together Interfaces will change as more details are defined Straight through Start at system input, progress through to final output (or vice versa) Outside In Front and back interfaces are defined first, interior details later, meet in middle Pieces may not join at the center properly Inside Out Inner connections are completed, outer pieces are added There is something to test from the beginning

22 Homework #3 Due 9/30/02 1. Translate the C++ simulation discussed today (complete version in Chapter 4), into your favorite language (C, Pascal, C++, Matlab, Java, etc. should be straightforward). 2. Modify simulation to simulate a three server queue, exponential arrivals, MeanArrivalTime=2, Normal service times, MeanServiceTime =(1,2,3), Sigma=(.5,1,1.5), 1000 customers. Output report to include server 1,2,3 utilization, maximum response time, histogram (plot) of customer time in system (minimum of 20 bins).