Lecture 36

BE (Computer Systems) Types of Simulations

Spring 2022 Semester 1. Emulation
2. Static (or Monte-Carlo) simulation
Computer Systems Modeling 3. Discrete-event simulation
(CS-438) 4. Continuous-event simulation

Emulation
Lecture #36-40
Simulation Modeling ▪ An emulator program is a simulation program
that runs on some existing system to make that
system appear to be something else.
Dr Syed Zaffar Qasim
Assistant Professor (CIS) ▪ Sacrifices performance for flexibility.

Emulation Static (Monte Carlo) Simulation

▪ Examples
o A terminal-emulator program is often executed on a
PC to make that computer appear to be a specific
type of terminal to a remote system.
o JVM is an example of a processor emulator.
❖Executes on a wide variety of different computer
systems.
❖It simulates the execution of a VP whose
instruction set is that defined in the JVM
specification.
❖JVM allows programs written in the Java to be
executed on any processor for which a JVM has
been written.
▪ No time parameter.
▪ The simulation is run until further refinement of the
state of the simulated system is no longer useful or
possible.
▪ Static simulations are often used to evaluate
o some physical phenomenon (probabilistically), or
o to numerically estimate the solution of some
mathematical expression, such as a complex
integral.

CSM: CS-438 1
 Lecture 37
Static (Monte Carlo) Simulation Example
▪ Consider the problem of numerically determining the
value of .
▪ We begin with a geometric description as shown in
Figure 1.
Fig 1
▪ Since the area of a circle with a radius of 1 is (1)2 = ,
the area of the quarter-circle within the first quadrant
is /4.
▪ The area contained within the unit square in this
quadrant is simply 1.
Static (Monte Carlo) Simulation Example
• Imagine throwing a large number of darts randomly at
figure 1 such that every dart hits within the unit
square.
• Then we count the number of times a dart hit within
the quarter-circle, ncirc, and the total number of darts
thrown, ntotal.
• Then the desired ratio of the two areas is
R = ncirc/ntotal.
▪ We can simulate this dart-throwing experiment by
generating two random numbers, u1 and u2, for each dart
thrown, such that u1 and u2 are both uniformly distributed
between 0 and 1.
CSM: CS-438
Static (Monte Carlo) Simulation Example
▪ Thus, the ratio of the area of the quarter-circle to the
area of the square, which we denote R, is R=/4.
▪ The numerical value of  then can be found from
=4R.
▪ The problem of computing the value of  has been
transformed into the equivalent geometric problem of
determining the ratio of the two areas, R.
▪ A Monte Carlo simulation can be used to find R by
modeling an equivalent physical system.
Static (Monte Carlo) Simulation Example
▪ If the distance from the origin of the point defined by (x,y)
= (u1,u2) is smaller than the radius of the circle, that is,
𝑢12 + 𝑢22 < 1, then the simulated dart has hit within the
quarter-circle.
▪ By repeating this process a large number of times, we can
theoretically compute the value of  to any level of
precision desired.
▪ The key to this type of Monte Carlo simulation is
identifying an appropriate physical model for the system
being studied.
2
 Lecture 38
Types of Simulations (cont’d)
Discrete-Event Simulations
▪ A simulation using a discrete-state model of the
system that changes as a function of time.
▪ May use discrete- or continuous-time values.
▪ Components of a discrete-event simulator:-
1. An event scheduler
2. Simulation Clock and a Time-advancing Mechanism
3. Event processing routines
4. Initialization Routines
5. Event-generation
6. Recording and summarization of data
Discrete-Event Simulations
1. An event scheduler (cont’d)
▪ Events can be manipulated in various ways as
follows:-
o Schedule event X at time T.
o Cancel a previously scheduled event X.
o Hold event X indefinitely (until it is
scheduled by another event).
CSM: CS-438
Discrete-Event Simulations
1. An event scheduler
▪ It maintains a linked list of all pending events in
their global time order.
▪ Processes the next event by removing it from the list
and dispatching it to the appropriate event-
processing routine.
▪ May be called several times during one event to
schedule other new events.
▪ The scheduler also inserts new events into the
appropriate point in the list on the basis of their
execution time.
Discrete-Event Simulations
2. Simulation Clock and a Time-advancing
Mechanism
▪ A global time variable records current simulation time.
▪ It can be updated by the scheduler using one of two
approaches:-
a) In the fixed-increment approach, the scheduler
increments the global clock by some fixed amount.
o It then checks the pending events.
o If the scheduled time for any of the pending events
matches the current time, all of these events are
dispatched for execution.
3
 Discrete-Event Simulations
2. Simulation Clock and a Time-advancing
Mechanism (cont’d)
o After all these events have been processed, the
scheduler again increments the global time
variable.
b)The alternative event-driven approach allows the
global time to jump to the value of the next event at
the head of the pending-event list.
o Here the value of the clock will change non-
uniformly.
o The event-driven approach is probably the most
common in simulations of computer systems.
Discrete-Event Simulations
3. Event processing routines
▪ Each event in the system is simulated by its own
event-processing routine.
▪ For example, routines to handle the two events of job
arrivals and job departure.
▪ These routines may update the global state and they
may generate additional events that must be inserted
into the pending-event list.
▪ For example, a memory-access event in a simulation
of a processor may result in two possible outcomes.

Discrete-Event Simulations
Discrete-Event Simulations
5. Event-generation
3. Event processing routines (cont’d)
▪ Three possible techniques
a)Cache-hit: the event may simply return the stored a) execution driven,
value. b) trace driven, and
c) distribution driven.
b)Cache-miss: the memory-access event routine may
generate a new event a) Execution driven
o that returns the corresponding data value tmiss o The simulator actually executes a benchmark
time units in the future, program.
o where tmiss is the time required to service a cache o It models the necessary details of the system being
miss. tested.
4. Initialization Routines o To simulate a program that does floating-point-
arithmetic operations or any input/output operations,
▪ These set the initial state of the system state variables ❖the simulator must provide mechanisms for actually
and initialize various random-number generation performing the necessary arithmetic and
streams. input/output operations.

CSM: CS-438 4
 Discrete-Event Simulations
o One of the main advantages of execution-driven
simulation is their accuracy i.e.
❖it provides a level of control of timing
between events, and the timing of each
individual event,
❖that is not possible with the other types of
simulation.
o Their primary disadvantage is that they can be
time consuming to develop, verify and execute
the simulator.
Discrete-Event Simulations
5. Event-generation
b) Trace driven
o Advantage: the simulator itself does not have to
provide all of the functionality needed by the
program, as is required in an execution-driven
simulation.
o Disadvantage: the trace represents only one
possible execution path which need not be
representative of the `typical' workload.
CSM: CS-438
Discrete-Event Simulations
5. Event-generation
b) Trace driven
o A trace is a time-ordered record of events on a real
system.
❖Sequence of pages
❖2, 6, 4, 11, …
o Trace-driven simulations are generally used in
analyzing or tuning resource management algorithms
❖Paging algorithms, CPU scheduling algorithms,
deadlock prevention algorithms etc.
o In this technique, trace of resource demand patterns
of key programs are obtained on a system.
❖This trace can then be used as input to the simulation
which models different algorithms
Discrete-Event Simulations
5. Event-generation
c) Distribution driven
o A distribution driven simulation is similar to a
trace-driven simulation, except that input
events are generated by the simulator itself to
follow some predefined probabilistic function.
o For example, sending messages over a
communication network could be modeled by
❖ using an exponential distribution
❖ to determine the amount of time that elapses
between each message.
5
 Discrete-Event Simulations
5. Event-generation
c) Distribution driven (cont’d)
o The simulator then produces an output that would
occur
❖if the real system were driven by an application program
❖that produced the same sequence of inputs.
o The simulation should be run many times
❖with different input sequences and outputs produced
❖then can be averaged to produce a statistics of the
performance of the system.
o This type of probabilistic (or stochastic) simulation
is often used when actual application programs are
not available to drive the simulator.
Lecture 39
Types of Simulations (cont’d)
Continuous-Event Simulations
▪ The state of the system varies continuously with
time.
▪ Used in chemical simulations where the state of
the system is described by the concentration of a
chemical substance.
CSM: CS-438
Discrete-Event Simulations
6. Recording and summarization of data
▪ The simulator maintains appropriate event counts
and time measurements.
▪ These values will be used at the end of simulation
to calculate appropriate statistics.
▪ For example, if a memory system is being
modeled, the simulator would probably count
o the total number of memory references and
o the number of those references that result in
cache misses.
▪ These values can be used to estimate the cache-
miss ratio.
The simulation algorithm
▪ Using the basic components, the overall
processing done by a discrete-event
simulator can be summarized in the
following steps:-
1. Initialize global state variables
2. Initialize the global time to 0
3. Add the first event to the pending-event list
6
 The simulation algorithm
Generation of Random Numbers
4. while ((more events in the list ) AND
(time < maximum simulation time limit))
{ ▪ Implementing a distribution-driven discrete-event
a. Advance the global time simulation and Monte-Carlo simulation requires a
b. Remove the next event from the pending-event list supply of random numbers from probability
c. Process the event distributions.
{
i. Perform event-specific operations ▪ A random number generator is an algorithm that
ii. Update global variables produces sequences of random numbers that
iii. Update simulation statistics
iv. Generate new events triggered by this event
follow a specified probability distribution.
and insert into the pending list
}
}
5. Print the simulation results

Lecture 40 Methods of Generation

Methods of Generation
▪ Three methods for generating successive
random samples (x = x1, x2, …) from a
probability distribution f(x):-
1. Inverse Transformation method.
2. Convolution method.
3. Acceptance-rejection method.
▪ The inverse method particularly suited for
analytically tractable PDFs, such as the
exponential and the uniform.
▪ The remaining two methods deal with more
complex cases, such as the normal and the
Poisson.
▪ All three methods rooted in the use of
independent and identically distributed uniform
(0, 1) random numbers.

CSM: CS-438 7
 The Inverse Transformation Method
▪ Suppose it is desired to obtain a random sample x from
the (continuous or discrete) PDF f(x).
▪ The inverse method first determines a closed-form
expression of the Cumulative Density Function, CDF,
F(x) = P{y ≤ x},
where 0 ≤ F(x) ≤ 1, for all defined values of y.
Fig 2: Sampling from a probability distribution by the
inverse method
The Inverse Transformation Method
▪ For certain continuous distributions, the inverse
transformation method can be implemented on a
computer by first solving the equation F(x) = R
analytically for x.
▪ The CDF for the exponential distribution is
F(x) = 1 – e-x, for x ≥ 0,
▪ Setting F(x) = R thereby yields
1 – e-x = R,
e-x = 1 – R,
CSM: CS-438
The Inverse Transformation Method
▪ Given that R is a random value obtained from a
uniform (0, 1) distribution, and assuming that F-1 is the
inverse of F.
▪ The steps of the method are as follows:
o Step 1. Generate a (0, 1) uniform random number, R.
o Step 2. Set F(x) = R and solve for x as x = F-1(R).
Fig 2: Sampling from a probability distribution by the
inverse method
The Inverse Transformation Method
▪ In terms of simulation, the result means that
arrivals are spaced x time units apart.
▪ For example,
o for  = 4 jobs per hour and
o R = .9,
o the time period until the next arrival occurs is
computed as
▪ The values of R used to obtain successive samples
must be selected randomly from a uniform (0, 1)
distribution.
8
 Uniform Distribution
▪ The PDF for a value x = unif(a,b) is

▪ The corresponding CDF is

• R= unif(0,1)
• x = F-1(R) = a + (b - a) R
• Suppose a = 10, b=20, R= 0.3456
• What is x?

CSM: CS-438 9