Imperial College London

ISE-2 Surprise 97 Project

Approaches to Performance Evaluation

IC Crest
  • Contents
  • Introduction
  • Benchmarking
  • Discrete Event Simulation
  • Modelling using Petrinets
  • Analytical Queuing Models
  • Perturbation Analysis of Discrete Event Dynamic Systems (PADEDS)
  • Conclusions
  • Index
  • Surprise 97 Index
  • Written by Wan Ling Li

    Performance - the capabilities of a machine

    Discrete Event Simulation

    In discrete event simulation an event is an occurrence which changes the state of a system. An event happens in a set time frame. It cannot occur in-between. An example would be a shift register, shifting the data bits only on a rising edge of a clock. A diagram of a shift register is shown below:

    3 Bit Shift Register

    Discrete Event Simulation gives a fast simulation time - all the events are discrete and occur at set intervals, so simulation of events are fast. This method is also quite accurate.

    A global event queue is used to process and manage the events. The global event queue is a linked list of time nodes each node has a pointer to a linked list of events or components. All the events in a node must be executed at the same time. So in a simulation, all the time nodes are traversed sequentially the output of the time node is only passed onto the next time node when all the events within that time node are completed.

    The disadvantage of this system occurs when one wishes to traverse backwards along the queue. All the components within a time node must be reversed too, this leads to a large amount of processing time. Also in the programming of a global event queue, care must be taken not to corrupt the queue. If the queue is corrupted then every component within the simulation is affected and the simulation becomes incorrect.

    All simulations must be run for a long time to ensure that the events which have a small probability of occuring do occur in the simulation. Hence a simulation can be time consuming as well as computer intensive. Computing power is required to calculate all the arrival and service processes.

    Discrete Event queues are good at modeling deterministic systems. However in systems which are modeled heavily on probability then a Monte Carlo simulator is a better choice. Typically in a stochastic system a Monte Carlo simulator is used.

    Next Section
    Last Updated 27th May, 1997