Page 68 - DCAP601_SIMULATION_AND_MODELING

Simulation and Modelling

Notes virtual paths and virtual circuits over the links and the traffic parameters can be defined by the
user to approach the behavior of the today’s and the future network components. Each ATM
switch is modelled as a Communicating Finite State Machine. A receipt of a CAC cell combined
with its current state activates an action routine of the CAC algorithm.
Emphasis is given in the abstract of traffic generation. Adversarial traffic leads the on-line
algorithms to their worst case competitive ratio of performance (measured against ideal off-line
algorithms that know the future). The DSS cannot simulate worst-case adversaries but can
approximate their behaviour by exploiting certain distribution of call request of high Kolmogorov
complexity. The DSS can of course use externally (pragmatic) generate call sequences.

Tasks Analyze the objectives provided by DSS tool.

4.2 Fixed Time Step vs Event-to-Event Model

In simulating any dynamic system-continuous or discrete-there must be a mechanism for the
flow of time. For we must advance time, keep track of the total elapsed time, determine the state
of the system at the new point in time, and terminate the simulation when the total elapsed time
equals or exceeds the simulation period

In a fixed time-step model a timer is simulated by the computer, this clock is updated by a fixed
time interval ?, and the system is examined to see if any event has taken place during this time
interval, all events that take place during that interval are treated as if they occurred
simultaneously at the end of this interval.
In a next-event or event-to-event model the computer advances time to the occurrence of the
next event, thus it shifts from one event to the next the system state does not change in between.
When something of interest happens to the system, the current time is kept track of.

The flowcharts for both models are shown.
To exemplify the difference between the two models, let us presuppose that we are simulating
the dynamics of the population in a fish bowl, starting with, say, 10 fish. If we used the fixed
time-step model with, say, or = 1 day, then we would scan the fish-bowl once every 24 hours, and
any births and deaths that take place are presumed to be during the last moment of this period.
Alternatively, if we use a next-event model then we will first find out when the next-event is to
take place and then advance the clock exactly to that time.

In general, the next-event model is preferred, excepting when you may be forced to use the fixed
time-step model because you do not waste any computer time in scanning those points in time
when nothing takes place.

This waste is bound to occur if we pick a reasonably small value for or. On the other hand, if or
is so large that one or more events must take place during each interval then our model becomes
unrealistic and may not yield meaningful results. Therefore in most simulations of discrete
systems the next `event model is used. The only drawback of the next-event model is that
usually its implementation turns out to be more complicated than the fixed time-step model.

62 LOVELY PROFESSIONAL UNIVERSITY