COMP9334讲解、辅导C/C++, Java，Matlab编程语言、讲解Analytic Model 调试Matlab程序|辅导R语言程序
- 首页 >> 其他 Fog/cloud Computing
Version 1.0
Due Date: 11:00pm Friday 26 April 2019.
This version: 20 March 2019
Updates to the project, including any corrections and clarifications, will be posted on the
subject website. Make sure that you check the course website regularly for updates.
Change log
Nothing for Version 1.0.
1 Introduction and learning objectives
This project is inspired by the research work reported in the article FogQN: An Analytic Model
for Fog/Cloud Computing [2] on modelling a computing system that makeg use of both fog and
cloud computing. The key question studied in the paper is on how to distribute work between
two computing resources, namely the fog and the cloud. In this project, you will use simulation
to try to answer a similar research question.
In this project, you will learn:
1. To use discrete event simulation to simulate a computer system
2. To use simulation to solve a design problem
3. To use statistically sound methods to analyse simulation outputs
2 Support provided
If you have problems doing this assignment, you can post your question on the message board
on the subject website. We strongly encourage you to do this as asking questions and
trying to answer them is a great way to learn. Do not be afraid that your question
may appear to be silly, the other students may very well have the same question!
3 Description of the fog/cloud computer system
3.1 Introduction to fog/cloud computing
You are probably aware that there are many more devices with Internet connectivity nowadays.
These new devices include sensors, vehicles, Internet-of-Things (IoT) etc. In general, these devices
are referred to as the edge devices. The fog and the cloud are two places to process the data coming
from these edges devices, see Figure 1.
1MOURADIAN et al.: COMPREHENSIVE SURVEY ON FOG COMPUTING: STATE-OF-THE-ART AND RESEARCH CHALLENGES 421
Fig. 4. The Fog System.
the name of Mobile Edge Computing (MEC), with a focus
on mobile networks and VM as virtualization technology.
However, its scope has been expanded in March 2017 to
encompass non-mobile network requirements (thus the
replacement of “Mobile” by “Multi-Access” in the name), as
well as virtualization technologies other than VM.
Prior to the scope expansion, the concept (envisioned as
a key technology towards 5G) [34] aimed at providing cloud
computing capabilities at the edge of mobile networks, and
within the Radio Access Network (RAN). These capabilities
are provided by mobile edge computing servers which can
be deployed at LTE macro base stations (eNodeB) sites, 3G
Radio Network Controller (RNC) sites, and at multi-Radio
Access Technology (RAT) sites. The envisioned applications
include augmented reality, intelligent video acceleration, and
connected cars. The edges of non-mobile networks and related
applications will certainly now be considered due to the
new scope.
The functional entities [41] (before the scope expansion)
include the mobile edge platform and the mobile edge host.
The mobile edge platform provides the functionality required
to provision mobile edge applications on a specific virtualization
infrastructure while the mobile edge host anchors the
platform and the virtualization infrastructure. Other functional
entities might now be added to the architecture as a result of
the scope expansion.
A fundamental goal assigned to the ETSI initiative is to
standardize the APIs between the mobile edge platform and
the applications in order to foster innovation in an open
environment. Several APIs have already been standardized
(e.g., Mobile application enablement API [42], radio network
API [43], Location API [44]). Reference [45] provides
a survey of MEC.
D. Fog Computing
Fog computing, a concept introduced by CISCO in 2012, is
an extension of cloud computing paradigm from the core to the
edge of the network. It enables computing at the edge of the
network, closer to IoT and/or the end-user devices. It also supports
virtualization. However, unlike cloudlet and MEC, fog is
tightly linked to the existence of a cloud, i.e., it cannot operate
in a standalone mode. This has driven a particular attention on
the interactions between the fog and the cloud [11]. Moreover,
fog has an n-tier architecture, offering more flexibility to the
system [13], [16].
Fig. 4 shows a fog system with a three-tier architecture.
It has three strata: The cloud stratum, the fog stratum, and
the IoT/end-users stratum. The fog stratum can be formed by
one or more fog domains, controlled by the same or different
providers. Each of these fog domains is formed by the
fog nodes that can include edge routers, switches, gateways,
access points, PCs, smartphones, set-top boxes, etc. As for the
IoT/end-users stratum, it is formed in turn by two domains,
the first including end-user devices and the second including
IoT devices. It should be noted that one of these two
domains may be absent in the stratum. It is, for instance,
the case of fog systems based – content delivery. There is
Figure 1: This figure depicts the three layers of edge devices, fog and cloud. The edge devices
are the end-user devices and IoT, which is short for Internet-of-things. Note that LAN and WAN
stand, respectively, for local area network and wide area network. This diagram shows that the
edge devices are closer to the fog computing resource than the cloud computing resource. This
figure is taken from [1].
The fog is a computing facility located closer to the edge devices. The network latency to
reach the fog computing facility is therefore low. However, the fog has a lower processing
speed.
The cloud is a computing facility farther away from the edge devices. The network latency
to get to the cloud is higher. However, the processing speed at the cloud is higher than that
in the fog.
Given that there are two facilities to process the data from the edge devices, a question is
which facility we should make use of in order to reduce the overall response time. Note that
the overall response time of using a facility is the sum of two parts, which are the network latency
to reach the facility and the response time of doing the processing at the facility. Now, let
us consider the case of doing all the processing at the fog. The network latency will be low but
the response time of doing the processing at the fog is high. On the other hand, if we do all the
processing at the cloud, then the network latency will be high but the response time of doing the
processing at the cloud is low. Neither one of these two options appears to be the best option.
In this project, you will investigate how to distribute work between the fog and the cloud in
order to minimise the overall response time. Note that this is all about fog/cloud computing that
we will tell you so that you get enough intuition to understand this project. If you want to learn
more about the fog/cloud computing paradigm, which is an area of commercial and academic
research at the moment, you can consult [1].
3.2 Fog/cloud setup for this project
Figure 2 depicts the fog/cloud system to be considered in this project. The system consists of
three components: the fog, the network and the cloud. We assume that all the requests will go
to the fog computing facility in the beginning. The fog will process these requests but imposes
2Fog Network Cloud Arriving
requests
Permanent
departure
at the fog
Requests
proceeding
to the cloud
via the
network
Permanent
departure
at the cloud
is a PS server is a server with
variable delay Notation:
Figure 2: The fog/cloud computing system in this project.
an upper limit on the processing time spent on each request. If a request requires an amount of
processing less than or equal to the upper limit, this request will be completed at the fog and leave
the system permanently upon completion. On the other hand, if a request requires an amount of
processing more than the upper limit, the request will be sent to the cloud via the network and
the cloud will complete the rest of the processing. After the cloud finishes the processing, the
request will leave the system permanently.
In this project, we model the fog as a processor sharing (PS) server. The same holds for the
cloud.
The network is modelled as a server with variable delay. We use the term network latency to
refer to this delay. Consider a request which is to be sent to the cloud. If this request leaves the
fog at time t and the network latency for this request is x, then this request will arrive at the cloud
at t + x. For a server with variable delay, the value of x for different requests can be different.
Note that there are no queues at the variable delay server modelling the network.
4 Examples
We will now present three examples to illustrate the operation of the fog/cloud system that you
will simulate in this project. In all these examples, we assume that the system is initially empty.
4.1 Example 1: Simulating a PS server
In this example, we assume there are 5 requests. The arrival times and service times required at
the fog are given in Table 1.
We further assume that the fog will give each arriving request a processing time of no more
than 5 time units. Since the service times of the five requests are all less than or equal to this
upper limit, all these requests will be completed at the fog and will not be sent to the cloud. This
effectively means that, for this example, we only need to simulate the PS server for the fog, and
3Arrival time at the fog Service time required at the fog
1 2.1
2 3.3
3 1.1
5 0.5
15 1.7
Table 1: Data for Example 1.
we can neglect the network and the cloud. The aim of this example is to show you how to simulate
a PS server.
The PS server was discussed in Lecture 4A. When there are n jobs in the server, then each job
receives 1
n
of the service.
The events in a PS server are the arrival of a job to the server and the departure of a completed
job from the server. You should convince yourselves that between two consecutive events,
the number jobs in a PS server remains the same. The discrete event simulation advances from
an event to the next one.
The key data structure that you need to maintain is the list of jobs in the server. Each job is
characterised by two attributes: the time the job arrives at the server and the remaining amount
of service the job still needs. Each time when a job arrives or departs, this data structure should
be updated. We will explain how this update can be done in a moment.
We will illustrate how the simulation of PS server works using “on-paper simulation”. Three
of the quantities that you need to keep track of are:
Next arrival time is the time of the next arrival
Next departure time assuming no more arrivals is defined as the time of the next
departure assuming that no more arrivals will come in the future. For simplicity, we will
simply use the phrase next departure time. For example, if there are three jobs in the server
at a certain time and these jobs still need 5, 6 and 10 units of service, then the next departure
time will be 15 time units later.
The list of jobs in the server. Each job is characterised by a 2-tuple. The first element of
the 2-tuple is the arrival time of the job at the server and the second element is the amount
of service it still needs.
The “on-paper simulation” is shown in Table 2. The notes in the last column explain what
updates you need to do for each event. Please note that there are more quantities that you need
to keep track of than those three that are mentioned above.
A graphical representation of the PS server status over time is given in Figure 3. This graphical
representation is the same as the one used in Lecture 4B to explain how you can calculate the
departure time of jobs in a PS server. There are three plots in the figure, showing the arrival times,
remaining amount of service for each job and the departure times. The figure is best viewed in
colour because the quantities related to each job is shown in a specific colour. Note that in between
two consecutive events, the remaining amount of service for each job is not a constant. What
is constant in between two consecutive events is the number of jobs in the server. The number
of jobs in the server determines the service rate of each job which is the slope of the remaining
service curves. You will see that the slope stays constant in between two events and changes each
time an event occurs.
4Master
clock
Event type Next
arrival
time
Next
departure
time
Job list Notes
t = 0 – 1 ∞ – We assume the server is empty at the start of the
simulation and this is indicated by departure time
of ∞.
t = 1 Arrival 2 3.1 (1,2.1) Since the event is an arrival, we need to update
(i) Next arrival time; (ii) Job list; and (iii) Next
departure time. There is one job in the list. The
notation (1,2.1) means the job arrives at t = 1
and at the time of the master clock, it still needs
2.1 units of service. If there are no more arrivals,
the next departure is expected to be at time 3.1
and this is the next departure time.
t = 2 Arrival 3 4.2 (1,1.1),
(2,3.3)
Since the event is an arrival, we need to update (i)
Next arrival time; (ii) Job list; and (iii) Next departure
time. Note that for the job list, you need
to add the new job to the list and you also need to
update the amount of service still needed by those
jobs that were in the server before the arrival of
this job. We now explain how the job (1.2.1) is
updated to (1,1.1). At the time of the last event
(which was t = 1), this job needed 2.1 units of
service. Given that the current time is t = 2,
which means 1 time unit has lapsed. Since there
was only one job in the server between t = 1 and
t = 2, the job received one unit of service hence
its remaining service time is 2.1 – 1 = 1.1. This
means that your simulation needs to remember
the time of the last event.
t = 3 Arrival 5 4.8 (1,0.6),
(2,2.8),
(3,1.1)
One unit of time has lapsed since the last event
and there were 2 jobs in the server, so the jobs in
the server received 0.5 units of service each.
t = 4.8 Departure 5 5.8 (2,2.2),
(3,0.5)
Since the event is a departure, you will need to
update the (i) Job list; and (ii) Next departure
time.
t = 5 Arrival 15 6.2 (2,2.1),
(3,0.4),
(5,0.5)
t = 6.2 Departure 15 6.4 (2,1.7)
(5,0.1)
t = 6.4 Departure 15 8 (2,1.6)
t = 8 Departure 15 ∞ –
Table 2: Table illustrating the updates in a PS server.
5Arrivals
Time
Remaining service
required
Time
Departures
Figure 3: Figure illustrating PS.
64.2 Example 2: Simulating the fog, the network and the cloud
This example illustrates the simulation of the fog/cloud system. We label the requests with indices
1, 2, 3 and so on; these indices are in the first column of Table 3. The second column of the table
shows the arrival times. The third column shows the amount of processing required and this
amount is measured by the time the request will need if all the processing is done in the fog. Since
the amount of time is measured with respect to the fog, we refer to it as the service time in the
fog time unit.
Request Arrival time at the fog Service time in the fog time unit
1 1.1 4.1
2 6.2 5.2
3 7.4 1.3
4 8.3 2.0
5 9.1 3.2
6 10.1 4.1
Table 3: Arrival times and service times in the fog time unit for Example 2.
For this example, we assume that the fog will only provide no more than 2 time units of service
to each request. This means that:
If a request requires 2 time units or less service from the fog, then the request will be
completed entirely at the fog. For the requests in Table 3, both Requests 3 and 4 will be
entirely completed at the fog.
If a request requires more than 2 time units of service from the fog, the request will get 2
time units of service from the fog. After receiving a service of 2 time units, the request will
be sent to the cloud via the network. The remaining work of the request will be completed
at the cloud. For the requests in Table 3, Requests 1, 2, 5 and 6 will each receive 2 time
units of service at the fog before they are sent to the cloud.
In general, we use the parameter fogTimeLimit to denote the maximum amount of service a
request can receive from the fog. For this example, the value of the parameter fogTimeLimit is 2.
In order to simulate those requests that are sent to the cloud, we will need their network
latency and the service time they need at the cloud. For this example, the network latency for
Requests 1, 2, 5 and 6 are given in the fourth column of Table 4.
We will now explain how the service time at the cloud is to be computed. Consider a request
whose service time in the fog time unit is x where x > fogTimeLimit, the amount of time required
by this request at the cloud is given by:
fogTimeToCloudTime ? (x ? fogTimeLimit)
where fogTimeToCloudTime is a parameter. In this example, we assume the value of the parameter
fogTimeToCloudTime is 0.6. Therefore, Request 1 will need 0.6 ? (4.1 ? 2) = 1.26 time
units of service at the cloud. Note that the quantity (x ? fogTimeLimit) represents the work that
this request still needs. This work is measured in terms of the amount of time needed at the
fog, but since this work is now to be completed at the cloud and the cloud is a faster computing
facility, so we need to convert the amount work to the time needed at the cloud. We model this
conversion by using the multiplication factor fogTimeToCloudTime. You can expect the value of
fogTimeToCloudTime is less than 1 because the cloud has a higher processing speed than the fog.
By using the setup above, Table 4 shows the actual time each request gets at the fog and the
service time needed at the cloud. Note that the actual time a request gets at the fog (third column
7in Table 4) is the smaller of the service time in the fog time unit (third column in Table 3) and
the value of the parameter fogTimeLimit. Note also that the service time needed at the cloud
(fifth column in Table 4) is deduced from the service time in the fog time unit, together with the
parameters fogTimeLimit and fogTimeToCloudTime.
Request Arrival time at
the fog
Actual service
time provided at
the fog
network latency Service time required
at the cloud
1 1.1 2 1.5 1.26
2 6.2 2 1.3 1.92
3 7.4 1.3
4 8.3 2
5 9.1 2 1.6 0.72
6 10.1 2 1.8 1.26
Table 4: This table shows arrival time at the fog, actual service time provided at the fog, network
latency and service time required at the cloud for Example 2.
In the simulation of the fog/cloud system, there are five events:
1. Arrival at the fog
2. Departure from the fog (without going to the network)
3. Departure from the fog to the network
4. Departure from the network to the cloud
5. Departure from the cloud
Table 5 shows the event times in ascending order with an explanation of the events.
8Time Request Event
1.1000 1 Arrival at the fog
3.1000 1 Departure from the fog to the network
4.6000 1 Departure from the network to the cloud. Note that
Request 1 departed from the fog at time 3.1 and the
network latency for this request is 1.5, therefore this
request departs from the network at 3.1 + 1.5 = 4.6.
5.8600 1 Departure from the cloud. Note that Request 1 arrived
at the cloud at time 4.6 and with no other jobs
arriving at the cloud when it was served, it departed
from the cloud at 4.6 + 1.26 = 5.86.
6.2000 2 Arrival at the fog
7.4000 3 Arrival at the fog
8.3000 4 Arrival at the fog
9.1000 5 Arrival at the fog
9.4333 2 Departure from the fog to the network. Note that
by this time, Request 2 has received 2 (= 1.2 + 0.9
/ 2 + 0.8 / 3 + 0.3333 / 4) units of service. Since
the value of the parameter fogTimeLimit is 2, this
request has to depart from the fog.
10.1000 6 Arrival at the fog
10.7333 2 Departure from the network to the cloud
11.2111 3 Departure from the fog
12.6533 2 Departure from the cloud
14.6611 4 Departure from the fog
15.1944 5 Departure from the fog to the network
15.5000 6 Departure from the fog to the network
16.7944 5 Departure from the network to the cloud
17.3000 6 Departure from the network to the cloud
17.7289 5 Departure from the cloud. Request 5 requires 0.72
units of service from the cloud. At time 17.7289,
the amount of service it has received is (17.3000 ?
16.7944) + (17.7289 ? 17.3000)/2 = 0.72.
18.7744 6 Departure from the cloud
Table 5: The event times for Example 2.
94.3 Example 3: Trace driven simulation
This example illustrate the data that you will be given if you are asked to do a trace-driven
simulation. You will be given:
The values of fogTimeToCloudTime and fogTimeLimit.
For each request, you will be given the arrival time at the fog, service time in the fog time
unit and network latency. Note that for a request whose service time in the fog time unit is
no more than the fogTimeLimit, we use a zero network latency to indicate this request will
not be sent to the cloud.
The data for this example are:
Arrival time at the fog Service time in the fog time unit network latency
1 3.7 1.5
2 5.1 1.4
4 1.3 0
5 2.4 0
6 4.5 1.6
fogTimeLimit = 2.5
fogTimeToCloudTime = 0.7
With the data above, the events in the fog/cloud system are shown in the table below. Note
that we have labelled the requests in the order that they arrive at the fog.
Time Request Event
1 1 Arrival at the fog
2 2 Arrival at the fog
4 3 Arrival at the fog
5 4 Arrival at the fog
5.6667 1 Departure from the fog to the network
6 5 Arrival at the fog
7.1667 1 Departure from the network to the cloud
8.0067 1 Departure from the cloud
8.7556 3 Departure from the fog
9.3556 2 Departure from the fog to the network
10.7556 2 Departure from the network to the cloud
11.8222 4 Departure from the fog
12.2 5 Departure from the fog to the network
12.5756 2 Departure from the cloud
13.8 5 Departure from the network to the cloud
15.2 5 Departure from the cloud
You can compute the system response times by using the arrival and departure times from the
system. The following table shows the arrival and departure times that can be used to compute the
system response times. The mean response time is 7.6720 without considering transient removal.
Arrival time Departure time
1 8.0067
2 12.5756
4 8.7556
5 11.8222
6 15.2
105 Project description
This project consists of two main parts. The first part is to develop a simulation program for the
fog/cloud system which is described in Section 3.2 and illustrated in Section 4. In the second part,
you will use the simulation program that you have developed to solve a design problem.
5.1 Simulation program
You must write your simulation program in one (or a combination) of the following languages or
numerical package: Python (either version 2 or 3), C, C++, Java or Matlab. All these languages
and numerical packages are available on the CSE system. We will test your program on the CSE
system so your submitted program must be able to run on a CSE computer. Note that it is
possible that due to version differences, code that runs on your own computer may not work on
the CSE system. It is your responsibility to ensure that your code works on the CSE system. In
particular, if you plan to use Matlab, you need to know that CSE runs Matlab 2013, which is
unfortunately older than expected.
We require you to write your simulation program as a function in your chosen language. Your
function must have the following seven inputs:
1. A parameter mode of string type. This parameter is to control whether your program will
run simulation using randomly generated arrival times, service times and network latencies;
or in trace driven mode. The value that the parameter mode can take is either random or
trace.
2. A parameter arrival for supplying arrival information to the program. The meaning of
arrival depends on mode. We will provide more information later on.
3. A parameter service for supplying service time in the fog time unit to the program. The
meaning of service depends on mode. We will provide more information later on.
4. A parameter network for supplying network latency to the program. The meaning of
service depends on mode. We will provide more information later on.
5. A parameter fogTimeLimit to input the value of fogTimeLimit. This parameter is a positive
floating point number.
6. A parameter fogTimeToCloudTime to input the value of fogTimeToCloudTime. This parameter
is a positive floating point number.
7. A parameter time_end which stops the simulation if the master clock exceeds this value.
This parameter is only relevant when mode is random. This parameter is a positive floating
point number.
Note that your simulation program must be a general program which allows different values
to be used. When we test your program, we will vary the parameter values. You can assume that
we will only use valid inputs for testing. You can have additional parameters if you wish.
For the simulation, you can always assume that the system is empty initially.
Note that we assume that it takes negligible time to send the requests to the fog.
5.1.1 The random mode
When your simulation is working in the random mode, it will generate the inter-arrival times,
service times in the fog time unit and network latencies in the following manner.
111. The inter-arrival probability distribution is exponentially distributed with parameter λ. This
means the mean arrival rate of the jobs is λ. You will need to supply the value of λ to your
program using the input parameter arrival.
2. The service time in the fog time unit t is generated by the probability density function g(t)
where:
Note that this probability density function has three parameters: α1, α2 and β.
3. The network latency is uniformly distributed in the open interval (ν1, ν2) where ν2 > ν1 > 0.
Note that network latency only applies to requests that are to be sent to the cloud.
5.1.2 The trace mode
When your simulation is working in the trace mode, it will read the list of arrival times, the list
of service times in the fog time unit and the list of network latencies from three separate ASCII files.
Let us use Example 3 in Section 4.3 for an illustration. The arrival times are [1, 2, 4, 5, 6],
the service times in the fog time unit are [3.7, 5.1, 1.3, 2.4, 4.5] and the network latencies are
[1.5, 1.4, 0, 0, 1.6]. Your program is required to simulate until all jobs have departed. For this
example, the last departure occurs at time 15.2 and you can stop the simulation there.
We will supply the arrival times, service times in the fog time unit and network latencies to
you using three ASCII files. The names of these three files have specific format, which we will
discuss in Section 6 on testing. For Example 3 in Section 4.3, the service times in the fog time
unit will be specified by a file whose contents are as follows:
3.700
5.100
1.300
2.400
4.500
where each service time takes up a line of the file. The same format is used for arrival times and
network latencies. You can assume that the data we provide for trace mode are consistent in
the following way: (1) The number of arrival times, the number of service times in fog unit and
the number of network latencies are equal. (2) If the service time in fog unit for a request is not
greater than fogTimeLimit, then the corresponding network latency is zero.
5.2 Determining a suitable value of fogTimeLimit
After writing your simulation program, your next step is to use your simulation program to do a
design.
For this design problem, you will assume the following parameter values: λ = 9.72, α1 = 0.01,
α2 = 0.4, β = 0.86, ν1 = 1.2, ν2 = 1.47 and fogTimeToCloudTime is 0.6.
12Your aim is to determine the value (or a range of values) of the parameter fogTimeLimit that
gives the best system response time. Recall from Section 3.1 that we will not get the best overall
response time if we do all the processing at the fog or all the processing at the cloud. Doing all
the processing at the fog corresponds to setting fogTimeLimit to infinity. Doing all the processing
at the cloud corresponds to setting fogTimeLimit to 0. This means that for some non-zero value
of fogTimeLimit, you can get a better overall response time.
In solving this design problem, you need to ensure that you use statistically sound methods
to compare systems. You will need to consider parameters such as length of simulation, number of
replications, transient removals and so on. You will need to justify in your report how you choose
the value or range of fogTimeLimit.
6 Testing simulation program
As part of the assessment for this project, you are asked to attend an interview so that we can
test your simulation program. During the interview, we will ask you to run a series of tests. Each
test is specified by a number of configuration files. For each test, we require four output files of
specific format so that we can verify the correctness of your simulation program. The aim of this
section is to specify the format of the configuration files and output files.
The number of tests that you will need to run is specified in an ASCII file with the name
num_tests.txt. If we want to run 12 tests in the interview, the file num_tests.txt contains the
string 12 only.
Each test is specified by 5 configurations files. We will index the tests from 1. If 12 tests are
used, then the indices for the tests are 1, 2, ...., 12. The names of the configuration files are:
For Test 1, the configuration files are mode_1.txt, para_1.txt, arrival_1.txt, service_1.txt
and network_1.txt. The files are similarly named for indices 2, 3, ..,9.
For Test 10, the configuration files are mode_10.txt, para_10.txt, arrival_10.txt, service_10.txt
and network_10.txt. The files are similarly named if the test index is a 2-digit number.
We will refer to these files using the generic names mode *.txt, para *.txt etc.
6.1 Wrapper file
When you submit your project, you must include a wrapper file which will loop through all the
tests. We will use this wrapper file in your interview to run your simulation. The pseudo-code for
your wrapper file is as follows.
Read the file num_tests.txt to determine the number of tests
FOR test_index from 1 to num_tests
Read in the configuration files for the test_index
Set the simulation mode and parameter values
Call your simulation function
Write the output files % This can be in the wrapper or your simulation function
The filename of this wrapper file must be called wrapper followed by the appropriate file extension
for the language that you used, e.g. wrapper.py for Python.
13Note that another purpose of the wrapper file is to show us how to run your code. We may need
to run additional tests and the wrapper file will allow us to do that. You should have comments
in the wrapper file to explain to us how to run your code.
6.2 Configuration file format
6.2.1 mode *.txt
This file is to indicate whether the simulation should run in the random or trace mode. The file
contains one string, which can either be random or trace.
6.2.2 para *.txt
If the simulation mode is trace, then this file has two lines. The first line is the value of fogTimeLimit.
The second line is the value of fogTimeToCloudTime. If the test is Example 3 in Section
4.3, then the contents of this file
Version 1.0
Due Date: 11:00pm Friday 26 April 2019.
This version: 20 March 2019
Updates to the project, including any corrections and clarifications, will be posted on the
subject website. Make sure that you check the course website regularly for updates.
Change log
Nothing for Version 1.0.
1 Introduction and learning objectives
This project is inspired by the research work reported in the article FogQN: An Analytic Model
for Fog/Cloud Computing [2] on modelling a computing system that makeg use of both fog and
cloud computing. The key question studied in the paper is on how to distribute work between
two computing resources, namely the fog and the cloud. In this project, you will use simulation
to try to answer a similar research question.
In this project, you will learn:
1. To use discrete event simulation to simulate a computer system
2. To use simulation to solve a design problem
3. To use statistically sound methods to analyse simulation outputs
2 Support provided
If you have problems doing this assignment, you can post your question on the message board
on the subject website. We strongly encourage you to do this as asking questions and
trying to answer them is a great way to learn. Do not be afraid that your question
may appear to be silly, the other students may very well have the same question!
3 Description of the fog/cloud computer system
3.1 Introduction to fog/cloud computing
You are probably aware that there are many more devices with Internet connectivity nowadays.
These new devices include sensors, vehicles, Internet-of-Things (IoT) etc. In general, these devices
are referred to as the edge devices. The fog and the cloud are two places to process the data coming
from these edges devices, see Figure 1.
1MOURADIAN et al.: COMPREHENSIVE SURVEY ON FOG COMPUTING: STATE-OF-THE-ART AND RESEARCH CHALLENGES 421
Fig. 4. The Fog System.
the name of Mobile Edge Computing (MEC), with a focus
on mobile networks and VM as virtualization technology.
However, its scope has been expanded in March 2017 to
encompass non-mobile network requirements (thus the
replacement of “Mobile” by “Multi-Access” in the name), as
well as virtualization technologies other than VM.
Prior to the scope expansion, the concept (envisioned as
a key technology towards 5G) [34] aimed at providing cloud
computing capabilities at the edge of mobile networks, and
within the Radio Access Network (RAN). These capabilities
are provided by mobile edge computing servers which can
be deployed at LTE macro base stations (eNodeB) sites, 3G
Radio Network Controller (RNC) sites, and at multi-Radio
Access Technology (RAT) sites. The envisioned applications
include augmented reality, intelligent video acceleration, and
connected cars. The edges of non-mobile networks and related
applications will certainly now be considered due to the
new scope.
The functional entities [41] (before the scope expansion)
include the mobile edge platform and the mobile edge host.
The mobile edge platform provides the functionality required
to provision mobile edge applications on a specific virtualization
infrastructure while the mobile edge host anchors the
platform and the virtualization infrastructure. Other functional
entities might now be added to the architecture as a result of
the scope expansion.
A fundamental goal assigned to the ETSI initiative is to
standardize the APIs between the mobile edge platform and
the applications in order to foster innovation in an open
environment. Several APIs have already been standardized
(e.g., Mobile application enablement API [42], radio network
API [43], Location API [44]). Reference [45] provides
a survey of MEC.
D. Fog Computing
Fog computing, a concept introduced by CISCO in 2012, is
an extension of cloud computing paradigm from the core to the
edge of the network. It enables computing at the edge of the
network, closer to IoT and/or the end-user devices. It also supports
virtualization. However, unlike cloudlet and MEC, fog is
tightly linked to the existence of a cloud, i.e., it cannot operate
in a standalone mode. This has driven a particular attention on
the interactions between the fog and the cloud [11]. Moreover,
fog has an n-tier architecture, offering more flexibility to the
system [13], [16].
Fig. 4 shows a fog system with a three-tier architecture.
It has three strata: The cloud stratum, the fog stratum, and
the IoT/end-users stratum. The fog stratum can be formed by
one or more fog domains, controlled by the same or different
providers. Each of these fog domains is formed by the
fog nodes that can include edge routers, switches, gateways,
access points, PCs, smartphones, set-top boxes, etc. As for the
IoT/end-users stratum, it is formed in turn by two domains,
the first including end-user devices and the second including
IoT devices. It should be noted that one of these two
domains may be absent in the stratum. It is, for instance,
the case of fog systems based – content delivery. There is
Figure 1: This figure depicts the three layers of edge devices, fog and cloud. The edge devices
are the end-user devices and IoT, which is short for Internet-of-things. Note that LAN and WAN
stand, respectively, for local area network and wide area network. This diagram shows that the
edge devices are closer to the fog computing resource than the cloud computing resource. This
figure is taken from [1].
The fog is a computing facility located closer to the edge devices. The network latency to
reach the fog computing facility is therefore low. However, the fog has a lower processing
speed.
The cloud is a computing facility farther away from the edge devices. The network latency
to get to the cloud is higher. However, the processing speed at the cloud is higher than that
in the fog.
Given that there are two facilities to process the data from the edge devices, a question is
which facility we should make use of in order to reduce the overall response time. Note that
the overall response time of using a facility is the sum of two parts, which are the network latency
to reach the facility and the response time of doing the processing at the facility. Now, let
us consider the case of doing all the processing at the fog. The network latency will be low but
the response time of doing the processing at the fog is high. On the other hand, if we do all the
processing at the cloud, then the network latency will be high but the response time of doing the
processing at the cloud is low. Neither one of these two options appears to be the best option.
In this project, you will investigate how to distribute work between the fog and the cloud in
order to minimise the overall response time. Note that this is all about fog/cloud computing that
we will tell you so that you get enough intuition to understand this project. If you want to learn
more about the fog/cloud computing paradigm, which is an area of commercial and academic
research at the moment, you can consult [1].
3.2 Fog/cloud setup for this project
Figure 2 depicts the fog/cloud system to be considered in this project. The system consists of
three components: the fog, the network and the cloud. We assume that all the requests will go
to the fog computing facility in the beginning. The fog will process these requests but imposes
2Fog Network Cloud Arriving
requests
Permanent
departure
at the fog
Requests
proceeding
to the cloud
via the
network
Permanent
departure
at the cloud
is a PS server is a server with
variable delay Notation:
Figure 2: The fog/cloud computing system in this project.
an upper limit on the processing time spent on each request. If a request requires an amount of
processing less than or equal to the upper limit, this request will be completed at the fog and leave
the system permanently upon completion. On the other hand, if a request requires an amount of
processing more than the upper limit, the request will be sent to the cloud via the network and
the cloud will complete the rest of the processing. After the cloud finishes the processing, the
request will leave the system permanently.
In this project, we model the fog as a processor sharing (PS) server. The same holds for the
cloud.
The network is modelled as a server with variable delay. We use the term network latency to
refer to this delay. Consider a request which is to be sent to the cloud. If this request leaves the
fog at time t and the network latency for this request is x, then this request will arrive at the cloud
at t + x. For a server with variable delay, the value of x for different requests can be different.
Note that there are no queues at the variable delay server modelling the network.
4 Examples
We will now present three examples to illustrate the operation of the fog/cloud system that you
will simulate in this project. In all these examples, we assume that the system is initially empty.
4.1 Example 1: Simulating a PS server
In this example, we assume there are 5 requests. The arrival times and service times required at
the fog are given in Table 1.
We further assume that the fog will give each arriving request a processing time of no more
than 5 time units. Since the service times of the five requests are all less than or equal to this
upper limit, all these requests will be completed at the fog and will not be sent to the cloud. This
effectively means that, for this example, we only need to simulate the PS server for the fog, and
3Arrival time at the fog Service time required at the fog
1 2.1
2 3.3
3 1.1
5 0.5
15 1.7
Table 1: Data for Example 1.
we can neglect the network and the cloud. The aim of this example is to show you how to simulate
a PS server.
The PS server was discussed in Lecture 4A. When there are n jobs in the server, then each job
receives 1
n
of the service.
The events in a PS server are the arrival of a job to the server and the departure of a completed
job from the server. You should convince yourselves that between two consecutive events,
the number jobs in a PS server remains the same. The discrete event simulation advances from
an event to the next one.
The key data structure that you need to maintain is the list of jobs in the server. Each job is
characterised by two attributes: the time the job arrives at the server and the remaining amount
of service the job still needs. Each time when a job arrives or departs, this data structure should
be updated. We will explain how this update can be done in a moment.
We will illustrate how the simulation of PS server works using “on-paper simulation”. Three
of the quantities that you need to keep track of are:
Next arrival time is the time of the next arrival
Next departure time assuming no more arrivals is defined as the time of the next
departure assuming that no more arrivals will come in the future. For simplicity, we will
simply use the phrase next departure time. For example, if there are three jobs in the server
at a certain time and these jobs still need 5, 6 and 10 units of service, then the next departure
time will be 15 time units later.
The list of jobs in the server. Each job is characterised by a 2-tuple. The first element of
the 2-tuple is the arrival time of the job at the server and the second element is the amount
of service it still needs.
The “on-paper simulation” is shown in Table 2. The notes in the last column explain what
updates you need to do for each event. Please note that there are more quantities that you need
to keep track of than those three that are mentioned above.
A graphical representation of the PS server status over time is given in Figure 3. This graphical
representation is the same as the one used in Lecture 4B to explain how you can calculate the
departure time of jobs in a PS server. There are three plots in the figure, showing the arrival times,
remaining amount of service for each job and the departure times. The figure is best viewed in
colour because the quantities related to each job is shown in a specific colour. Note that in between
two consecutive events, the remaining amount of service for each job is not a constant. What
is constant in between two consecutive events is the number of jobs in the server. The number
of jobs in the server determines the service rate of each job which is the slope of the remaining
service curves. You will see that the slope stays constant in between two events and changes each
time an event occurs.
4Master
clock
Event type Next
arrival
time
Next
departure
time
Job list Notes
t = 0 – 1 ∞ – We assume the server is empty at the start of the
simulation and this is indicated by departure time
of ∞.
t = 1 Arrival 2 3.1 (1,2.1) Since the event is an arrival, we need to update
(i) Next arrival time; (ii) Job list; and (iii) Next
departure time. There is one job in the list. The
notation (1,2.1) means the job arrives at t = 1
and at the time of the master clock, it still needs
2.1 units of service. If there are no more arrivals,
the next departure is expected to be at time 3.1
and this is the next departure time.
t = 2 Arrival 3 4.2 (1,1.1),
(2,3.3)
Since the event is an arrival, we need to update (i)
Next arrival time; (ii) Job list; and (iii) Next departure
time. Note that for the job list, you need
to add the new job to the list and you also need to
update the amount of service still needed by those
jobs that were in the server before the arrival of
this job. We now explain how the job (1.2.1) is
updated to (1,1.1). At the time of the last event
(which was t = 1), this job needed 2.1 units of
service. Given that the current time is t = 2,
which means 1 time unit has lapsed. Since there
was only one job in the server between t = 1 and
t = 2, the job received one unit of service hence
its remaining service time is 2.1 – 1 = 1.1. This
means that your simulation needs to remember
the time of the last event.
t = 3 Arrival 5 4.8 (1,0.6),
(2,2.8),
(3,1.1)
One unit of time has lapsed since the last event
and there were 2 jobs in the server, so the jobs in
the server received 0.5 units of service each.
t = 4.8 Departure 5 5.8 (2,2.2),
(3,0.5)
Since the event is a departure, you will need to
update the (i) Job list; and (ii) Next departure
time.
t = 5 Arrival 15 6.2 (2,2.1),
(3,0.4),
(5,0.5)
t = 6.2 Departure 15 6.4 (2,1.7)
(5,0.1)
t = 6.4 Departure 15 8 (2,1.6)
t = 8 Departure 15 ∞ –
Table 2: Table illustrating the updates in a PS server.
5Arrivals
Time
Remaining service
required
Time
Departures
Figure 3: Figure illustrating PS.
64.2 Example 2: Simulating the fog, the network and the cloud
This example illustrates the simulation of the fog/cloud system. We label the requests with indices
1, 2, 3 and so on; these indices are in the first column of Table 3. The second column of the table
shows the arrival times. The third column shows the amount of processing required and this
amount is measured by the time the request will need if all the processing is done in the fog. Since
the amount of time is measured with respect to the fog, we refer to it as the service time in the
fog time unit.
Request Arrival time at the fog Service time in the fog time unit
1 1.1 4.1
2 6.2 5.2
3 7.4 1.3
4 8.3 2.0
5 9.1 3.2
6 10.1 4.1
Table 3: Arrival times and service times in the fog time unit for Example 2.
For this example, we assume that the fog will only provide no more than 2 time units of service
to each request. This means that:
If a request requires 2 time units or less service from the fog, then the request will be
completed entirely at the fog. For the requests in Table 3, both Requests 3 and 4 will be
entirely completed at the fog.
If a request requires more than 2 time units of service from the fog, the request will get 2
time units of service from the fog. After receiving a service of 2 time units, the request will
be sent to the cloud via the network. The remaining work of the request will be completed
at the cloud. For the requests in Table 3, Requests 1, 2, 5 and 6 will each receive 2 time
units of service at the fog before they are sent to the cloud.
In general, we use the parameter fogTimeLimit to denote the maximum amount of service a
request can receive from the fog. For this example, the value of the parameter fogTimeLimit is 2.
In order to simulate those requests that are sent to the cloud, we will need their network
latency and the service time they need at the cloud. For this example, the network latency for
Requests 1, 2, 5 and 6 are given in the fourth column of Table 4.
We will now explain how the service time at the cloud is to be computed. Consider a request
whose service time in the fog time unit is x where x > fogTimeLimit, the amount of time required
by this request at the cloud is given by:
fogTimeToCloudTime ? (x ? fogTimeLimit)
where fogTimeToCloudTime is a parameter. In this example, we assume the value of the parameter
fogTimeToCloudTime is 0.6. Therefore, Request 1 will need 0.6 ? (4.1 ? 2) = 1.26 time
units of service at the cloud. Note that the quantity (x ? fogTimeLimit) represents the work that
this request still needs. This work is measured in terms of the amount of time needed at the
fog, but since this work is now to be completed at the cloud and the cloud is a faster computing
facility, so we need to convert the amount work to the time needed at the cloud. We model this
conversion by using the multiplication factor fogTimeToCloudTime. You can expect the value of
fogTimeToCloudTime is less than 1 because the cloud has a higher processing speed than the fog.
By using the setup above, Table 4 shows the actual time each request gets at the fog and the
service time needed at the cloud. Note that the actual time a request gets at the fog (third column
7in Table 4) is the smaller of the service time in the fog time unit (third column in Table 3) and
the value of the parameter fogTimeLimit. Note also that the service time needed at the cloud
(fifth column in Table 4) is deduced from the service time in the fog time unit, together with the
parameters fogTimeLimit and fogTimeToCloudTime.
Request Arrival time at
the fog
Actual service
time provided at
the fog
network latency Service time required
at the cloud
1 1.1 2 1.5 1.26
2 6.2 2 1.3 1.92
3 7.4 1.3
4 8.3 2
5 9.1 2 1.6 0.72
6 10.1 2 1.8 1.26
Table 4: This table shows arrival time at the fog, actual service time provided at the fog, network
latency and service time required at the cloud for Example 2.
In the simulation of the fog/cloud system, there are five events:
1. Arrival at the fog
2. Departure from the fog (without going to the network)
3. Departure from the fog to the network
4. Departure from the network to the cloud
5. Departure from the cloud
Table 5 shows the event times in ascending order with an explanation of the events.
8Time Request Event
1.1000 1 Arrival at the fog
3.1000 1 Departure from the fog to the network
4.6000 1 Departure from the network to the cloud. Note that
Request 1 departed from the fog at time 3.1 and the
network latency for this request is 1.5, therefore this
request departs from the network at 3.1 + 1.5 = 4.6.
5.8600 1 Departure from the cloud. Note that Request 1 arrived
at the cloud at time 4.6 and with no other jobs
arriving at the cloud when it was served, it departed
from the cloud at 4.6 + 1.26 = 5.86.
6.2000 2 Arrival at the fog
7.4000 3 Arrival at the fog
8.3000 4 Arrival at the fog
9.1000 5 Arrival at the fog
9.4333 2 Departure from the fog to the network. Note that
by this time, Request 2 has received 2 (= 1.2 + 0.9
/ 2 + 0.8 / 3 + 0.3333 / 4) units of service. Since
the value of the parameter fogTimeLimit is 2, this
request has to depart from the fog.
10.1000 6 Arrival at the fog
10.7333 2 Departure from the network to the cloud
11.2111 3 Departure from the fog
12.6533 2 Departure from the cloud
14.6611 4 Departure from the fog
15.1944 5 Departure from the fog to the network
15.5000 6 Departure from the fog to the network
16.7944 5 Departure from the network to the cloud
17.3000 6 Departure from the network to the cloud
17.7289 5 Departure from the cloud. Request 5 requires 0.72
units of service from the cloud. At time 17.7289,
the amount of service it has received is (17.3000 ?
16.7944) + (17.7289 ? 17.3000)/2 = 0.72.
18.7744 6 Departure from the cloud
Table 5: The event times for Example 2.
94.3 Example 3: Trace driven simulation
This example illustrate the data that you will be given if you are asked to do a trace-driven
simulation. You will be given:
The values of fogTimeToCloudTime and fogTimeLimit.
For each request, you will be given the arrival time at the fog, service time in the fog time
unit and network latency. Note that for a request whose service time in the fog time unit is
no more than the fogTimeLimit, we use a zero network latency to indicate this request will
not be sent to the cloud.
The data for this example are:
Arrival time at the fog Service time in the fog time unit network latency
1 3.7 1.5
2 5.1 1.4
4 1.3 0
5 2.4 0
6 4.5 1.6
fogTimeLimit = 2.5
fogTimeToCloudTime = 0.7
With the data above, the events in the fog/cloud system are shown in the table below. Note
that we have labelled the requests in the order that they arrive at the fog.
Time Request Event
1 1 Arrival at the fog
2 2 Arrival at the fog
4 3 Arrival at the fog
5 4 Arrival at the fog
5.6667 1 Departure from the fog to the network
6 5 Arrival at the fog
7.1667 1 Departure from the network to the cloud
8.0067 1 Departure from the cloud
8.7556 3 Departure from the fog
9.3556 2 Departure from the fog to the network
10.7556 2 Departure from the network to the cloud
11.8222 4 Departure from the fog
12.2 5 Departure from the fog to the network
12.5756 2 Departure from the cloud
13.8 5 Departure from the network to the cloud
15.2 5 Departure from the cloud
You can compute the system response times by using the arrival and departure times from the
system. The following table shows the arrival and departure times that can be used to compute the
system response times. The mean response time is 7.6720 without considering transient removal.
Arrival time Departure time
1 8.0067
2 12.5756
4 8.7556
5 11.8222
6 15.2
105 Project description
This project consists of two main parts. The first part is to develop a simulation program for the
fog/cloud system which is described in Section 3.2 and illustrated in Section 4. In the second part,
you will use the simulation program that you have developed to solve a design problem.
5.1 Simulation program
You must write your simulation program in one (or a combination) of the following languages or
numerical package: Python (either version 2 or 3), C, C++, Java or Matlab. All these languages
and numerical packages are available on the CSE system. We will test your program on the CSE
system so your submitted program must be able to run on a CSE computer. Note that it is
possible that due to version differences, code that runs on your own computer may not work on
the CSE system. It is your responsibility to ensure that your code works on the CSE system. In
particular, if you plan to use Matlab, you need to know that CSE runs Matlab 2013, which is
unfortunately older than expected.
We require you to write your simulation program as a function in your chosen language. Your
function must have the following seven inputs:
1. A parameter mode of string type. This parameter is to control whether your program will
run simulation using randomly generated arrival times, service times and network latencies;
or in trace driven mode. The value that the parameter mode can take is either random or
trace.
2. A parameter arrival for supplying arrival information to the program. The meaning of
arrival depends on mode. We will provide more information later on.
3. A parameter service for supplying service time in the fog time unit to the program. The
meaning of service depends on mode. We will provide more information later on.
4. A parameter network for supplying network latency to the program. The meaning of
service depends on mode. We will provide more information later on.
5. A parameter fogTimeLimit to input the value of fogTimeLimit. This parameter is a positive
floating point number.
6. A parameter fogTimeToCloudTime to input the value of fogTimeToCloudTime. This parameter
is a positive floating point number.
7. A parameter time_end which stops the simulation if the master clock exceeds this value.
This parameter is only relevant when mode is random. This parameter is a positive floating
point number.
Note that your simulation program must be a general program which allows different values
to be used. When we test your program, we will vary the parameter values. You can assume that
we will only use valid inputs for testing. You can have additional parameters if you wish.
For the simulation, you can always assume that the system is empty initially.
Note that we assume that it takes negligible time to send the requests to the fog.
5.1.1 The random mode
When your simulation is working in the random mode, it will generate the inter-arrival times,
service times in the fog time unit and network latencies in the following manner.
111. The inter-arrival probability distribution is exponentially distributed with parameter λ. This
means the mean arrival rate of the jobs is λ. You will need to supply the value of λ to your
program using the input parameter arrival.
2. The service time in the fog time unit t is generated by the probability density function g(t)
where:
Note that this probability density function has three parameters: α1, α2 and β.
3. The network latency is uniformly distributed in the open interval (ν1, ν2) where ν2 > ν1 > 0.
Note that network latency only applies to requests that are to be sent to the cloud.
5.1.2 The trace mode
When your simulation is working in the trace mode, it will read the list of arrival times, the list
of service times in the fog time unit and the list of network latencies from three separate ASCII files.
Let us use Example 3 in Section 4.3 for an illustration. The arrival times are [1, 2, 4, 5, 6],
the service times in the fog time unit are [3.7, 5.1, 1.3, 2.4, 4.5] and the network latencies are
[1.5, 1.4, 0, 0, 1.6]. Your program is required to simulate until all jobs have departed. For this
example, the last departure occurs at time 15.2 and you can stop the simulation there.
We will supply the arrival times, service times in the fog time unit and network latencies to
you using three ASCII files. The names of these three files have specific format, which we will
discuss in Section 6 on testing. For Example 3 in Section 4.3, the service times in the fog time
unit will be specified by a file whose contents are as follows:
3.700
5.100
1.300
2.400
4.500
where each service time takes up a line of the file. The same format is used for arrival times and
network latencies. You can assume that the data we provide for trace mode are consistent in
the following way: (1) The number of arrival times, the number of service times in fog unit and
the number of network latencies are equal. (2) If the service time in fog unit for a request is not
greater than fogTimeLimit, then the corresponding network latency is zero.
5.2 Determining a suitable value of fogTimeLimit
After writing your simulation program, your next step is to use your simulation program to do a
design.
For this design problem, you will assume the following parameter values: λ = 9.72, α1 = 0.01,
α2 = 0.4, β = 0.86, ν1 = 1.2, ν2 = 1.47 and fogTimeToCloudTime is 0.6.
12Your aim is to determine the value (or a range of values) of the parameter fogTimeLimit that
gives the best system response time. Recall from Section 3.1 that we will not get the best overall
response time if we do all the processing at the fog or all the processing at the cloud. Doing all
the processing at the fog corresponds to setting fogTimeLimit to infinity. Doing all the processing
at the cloud corresponds to setting fogTimeLimit to 0. This means that for some non-zero value
of fogTimeLimit, you can get a better overall response time.
In solving this design problem, you need to ensure that you use statistically sound methods
to compare systems. You will need to consider parameters such as length of simulation, number of
replications, transient removals and so on. You will need to justify in your report how you choose
the value or range of fogTimeLimit.
6 Testing simulation program
As part of the assessment for this project, you are asked to attend an interview so that we can
test your simulation program. During the interview, we will ask you to run a series of tests. Each
test is specified by a number of configuration files. For each test, we require four output files of
specific format so that we can verify the correctness of your simulation program. The aim of this
section is to specify the format of the configuration files and output files.
The number of tests that you will need to run is specified in an ASCII file with the name
num_tests.txt. If we want to run 12 tests in the interview, the file num_tests.txt contains the
string 12 only.
Each test is specified by 5 configurations files. We will index the tests from 1. If 12 tests are
used, then the indices for the tests are 1, 2, ...., 12. The names of the configuration files are:
For Test 1, the configuration files are mode_1.txt, para_1.txt, arrival_1.txt, service_1.txt
and network_1.txt. The files are similarly named for indices 2, 3, ..,9.
For Test 10, the configuration files are mode_10.txt, para_10.txt, arrival_10.txt, service_10.txt
and network_10.txt. The files are similarly named if the test index is a 2-digit number.
We will refer to these files using the generic names mode *.txt, para *.txt etc.
6.1 Wrapper file
When you submit your project, you must include a wrapper file which will loop through all the
tests. We will use this wrapper file in your interview to run your simulation. The pseudo-code for
your wrapper file is as follows.
Read the file num_tests.txt to determine the number of tests
FOR test_index from 1 to num_tests
Read in the configuration files for the test_index
Set the simulation mode and parameter values
Call your simulation function
Write the output files % This can be in the wrapper or your simulation function
The filename of this wrapper file must be called wrapper followed by the appropriate file extension
for the language that you used, e.g. wrapper.py for Python.
13Note that another purpose of the wrapper file is to show us how to run your code. We may need
to run additional tests and the wrapper file will allow us to do that. You should have comments
in the wrapper file to explain to us how to run your code.
6.2 Configuration file format
6.2.1 mode *.txt
This file is to indicate whether the simulation should run in the random or trace mode. The file
contains one string, which can either be random or trace.
6.2.2 para *.txt
If the simulation mode is trace, then this file has two lines. The first line is the value of fogTimeLimit.
The second line is the value of fogTimeToCloudTime. If the test is Example 3 in Section
4.3, then the contents of this file