Maximising the Shared Bandwidth for Classroom Quiz …STAs connect based on RSSI(or SNR), this leads...
Transcript of Maximising the Shared Bandwidth for Classroom Quiz …STAs connect based on RSSI(or SNR), this leads...
Maximising the Shared Bandwidth for Classroom QuizConduct Scenario in Wireless Environment
M. Tech. Stage II Report
Submitted in partial fulfilment of the requirements for the degree of
Master of Technology
by
Pramendra Singh RajputRoll No : 113050025
under the guidance of
Prof. D. B. Phatak
Department of Computer Science and EngineeringIndian Institute of Technology, Bombay
June 2013
Dissertation Approval
The dissertation titled
“Maximising the Shared Bandwidth for Classroom Quiz ConductScenario in Wireless Environment”
by
Pramendra Singh Rajput
(Roll No. 113050025)
is approved for the degree of
Master of Technology in Computer Science & Engineering.
Prof. Sridhar Iyer Prof. Deepak B. Phatak(Internal Examiner) (Supervisor)
Dr. Avinash Awate Prof. Uday Gaitonde(External Examiner) (Chairperson)
Date: June 27, 2013Place: Indian Institute of Technology Bombay, Mumbai
Declaration
I declare that this written submission presents my ideas in my own words and where othersideas or words have been included, I have appropriately cited and referenced the original sources. Ideclare that I have properly and accurately acknowledged all sources used in the production of thisthesis.
I also declare that I have adhered to all principles of academic honesty and integrity and have notmisrepresented or fabricated or falsified any idea/data/fact/source in my submission. I understandthat any violation of the above will be a cause for disciplinary action by the Institute and canalso evoke penal action from the sources which have not been properly cited or from whom properpermission has not been taken when needed.
Pramendra Singh RajputDate: June 27, 2013Place: Indian Institute of Technology Bombay, Mumbai
Abstract
In 802.11 Wireless LAN performance of the network start degrading with increase in number
of clients. As the number of STAs increase beyond 40 performance start degrading significantly
because of increase in interference and collision due to this network stop supporting any more
clients. To resolve this problem we have propose a random batch-mode connectivity algorithm.
The proposed scheme can be use for classroom quiz conduct scenario. In this scheme we have used
the fact that in quiz scenario STA don’t need persistent connectivity with AP and the data to be
transferred is of small size. With our scheme large number of clients can be supported using single
AP. We have tested this scheme on real test-bed and result shows that performance is increased
many folds. Hence, in order to support large number of clients for quiz conduct, our solution can
be efficiently used.
Acknowledgements
I would like to express my sincere gratitude to my guide Prof. D.B. Phatak for his constant
encouragement and corrective guidance. He has been my primary source of motivation and advice
during my project. I would like to thank Mr. Mayank Mishra and Mr. Nagesh Karmali for constant
guidance without which this project could have been more difficult. I also thank Mr. Rajesh
Kushalkar, Mr. Raj Agrawal and my friends who have always been alongside me throughout this
project. Last I want to thanks god and my parents for their blessings.
Contents
1 INTRODUCTION 6
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Applications of Proposed Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Various Cases that can occur in class room environment . . . . . . . . . . . 7
1.3 Overview of Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 LITERATURE SURVEY 9
2.1 Load Distribution Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Based on RSSI and number of client already connected to AP . . . . . . . . 10
2.1.2 Distributed load balancing by hiding SSID . . . . . . . . . . . . . . . . . . . 11
2.1.3 Load balancing approach based on deadline miss ratio (DMR) . . . . . . . . 12
2.2 Modification of WLAN parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Physical layer parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 MAC layer parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3 PCF or DCF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Comparison of various methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 CONTRIBUTION 17
3.1 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Problem Solving Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4 SOFTWARE IMPLEMENTATION 25
4.1 Server Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.1 Quiz Create Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.1.2 Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1
4.1.3 Upload/Download Module . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.1.4 Web Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 Client module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2.1 Front End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.2 Downloader . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2.3 Quiz conduct . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.4 Uploader . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2.5 Connection Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3 Limitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5 EXPERIMENTAL RESULT 35
6 CONCLUSION 41
7 FUTURE WORK 42
Appendices 45
A 46
A.1 Quiz file Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
A.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
A.2.1 facts from simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2
List of Figures
2.1 Classification of approaches for supporting many users and improving bandwidth
utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Basic System Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 WLAN Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Determining value of α . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Comparison of PCF and DCF, source [17] . . . . . . . . . . . . . . . . . . . . . . . 16
3.1 Simulation of Random Numbers for 100 clients . . . . . . . . . . . . . . . . . . . . . 20
3.2 Simulation of Random Numbers for 200 clients . . . . . . . . . . . . . . . . . . . . . 21
3.3 Simulation of Random Numbers for 225 clients . . . . . . . . . . . . . . . . . . . . . 22
3.4 Architecture Diagram of proposed model . . . . . . . . . . . . . . . . . . . . . . . . 23
3.5 interaction Protocol between Client-Server Application . . . . . . . . . . . . . . . . 24
4.1 Software Architecture Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 server intersection Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Database Relation diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4 Web Interface for creating and managing Quizzes . . . . . . . . . . . . . . . . . . . 29
4.5 Normal quiz mode Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.6 spot quiz mode Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.7 Front End Interface of Aakash Quiz.apk . . . . . . . . . . . . . . . . . . . . . . . . 32
4.8 Aaksh quiz application Flow Diagram . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.2 Total Users Vs Number of Successful download upload . . . . . . . . . . . . . . . . 37
5.3 Total Users Vs total Successful Users . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.4 Total Users Vs time taken in download and upload . . . . . . . . . . . . . . . . . . 38
3
5.5 Time Interval Vs successful download . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.6 Time Interval Vs successful upload . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4
List of Tables
3.1 Random batch-mode connectivity algorithm . . . . . . . . . . . . . . . . . . . . . . 19
5.1 Hardware And Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 Experiment Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5
Chapter 1
INTRODUCTION
In 802.11 wireless LAN the bandwidth is limited resource. As the number of users or load increased
on the network the performance decreases rapidly. Presently 802.11g standard is being used in
most of the WLAN installation. 802.11g provide the network bandwidth of 54 Mbps but actually
this bandwidth falls to few Mbps in case of loaded network(30-40 STAs) and if load increase some
more(more then 40 STAs) then the performance of WLAN is very poor. The reason for poor
performance is the increase in the interference due to which collision get increase and performance
falls. If clients are many then usually we increase the performance by installing new APs. When we
install new APs performance of WLAN increases, now WLAN can support more users. In case of
many Wireless routers other problem comes in picture for example load distribution. As by default
STAs connect based on RSSI(or SNR), this leads to the unbalanced distribution of STA among the
APs. Hence QoS, real time services and overall bandwidth utilization of the network becomes poor.
If STAs is distributed properly such that all APs are equally loaded then bandwidth utilization
can be increase significantly compared to the general performance. In this project we are trying
to maximize the bandwidth utilization of WLAN having many STAs (tablets) ready to get service
form APs. We need to do this dynamically as the number of STAs can very time to time.
1.1 Motivation
When it comes to supporting many clients, all present approaches points out toward using multiple
AP and then use some load balancing approach that can distribute the clients evenly among the
access points. But increaseing number of APs increases the cost of installation. This method of
6
using multiple AP is not scalable as every time we need to deploy new AP with increase in number
of clients. Another problem with multiple AP is that if number of clients are less resources remain
underutilised.
No approach has been proposed that can support many STAs using single AP. We are proposing
a scheme that can support large number of STAs using single AP. In proposed approach clients
connect to AP, download their content and disconnect from AP. In this way many more clients can
be supported. We disconnect every STA in the beginning and give chance to them in batches of
≈ 25. In this way interference is minimise and we can support many clients. This scheme uses
single AP hence installation cost is very low and most important benefit is, this scheme support
many clients. In this approach me don’t need to make changes on the AP side or in the firmware
of the AP. This thing prevent us from a big overhead as most of the firmware are not open source.
1.2 Applications of Proposed Scheme
The proposed scheme can be use very efficiently and effectively for the class room quiz conduct
applications. In classroom quiz conduct application student need to download some files and upload
the answers. File size of the quiz is also small(in KiBs). Hence in class room quiz application we
can connect user only for downloading and uploading time and disconnect them for other time.
1.2.1 Various Cases that can occur in class room environment
� Overloaded N/W
Conducting quiz in the class of 200 or 250 students. In this case the load on the network
will be high and performance will be limited by the number of simultaneous connection
to the AP. Here we have to minimize the latency for establishing the connection and
downloading the quizzes for all the tablets. Also make it like real life quiz conduct
scenario.
� Under-loaded N/W
Conducting quiz in the class of 10 or 20 student. In this case our algorithm should not
become the bottleneck for the fast and smooth run of quiz as it would have been without
our algorithm.
7
1.3 Overview of Report
Chapter one gives some overview of the present WLAN service and their limitation. It give brief
idea about our proposed approach to overcome the limitation of WLAN. Chapter two is literature
survey that give information about the work done by other to handle large number of clients in
wireless environment. Most of this work revolve around using multiple APs and then do proper
load balancing. Load balancing is done in order to distribute the clients uniformly among multiple
APs so that all APs are evenly utilise. Chapter three gives the comparison and limitation of present
approaches to solve the problem of WLAN. It also present the our new propose approach that can
solve the WLAN problem of limited clients in case of class room quiz scenario. Chapter 6 present
the software implemented by us for class room quizzes on android devices particularly on Aakash
tablet. Chapter seven shows the test-bed and testing result of our approach. Chapter eight is
conclusion and future work. At the end we can see the references and appendices.
In this report we have used the STAs, tablets, clients, user and students interchangeably. All
these keywords represent one thing i.e a device having WLAN cards and want to use WLAN service.
8
Chapter 2
LITERATURE SURVEY
In order to support large number of STAs many approaches has been proposed. These Approaches
can be classified mainly in two parts, first using the load distribution algorithms and by modifying
the WLAN parameters figure 2.1.
Figure 2.1: Classification of approaches for supporting many users and improving bandwidth uti-lization
2.1 Load Distribution Algorithms
Load distribution algorithms distribute the STAs among the multiple APs whenever load get un-
balanced. In this way these methods try to maintain the load in balance condition on network and
improve the bandwidth utilization.
9
2.1.1 Based on RSSI and number of client already connected to AP
Wireless devices get connected to an AP based on RSSI. When this RSSI value falls below a
predefined value STA leave that AP and get connected to another AP. This causes the unbalanced
distribution of the clients as more clients will get connected to their nearby AP and other AP will
remain under utilized. In this approach they solve this problem by moving the STAs based on RSSI
value as well as the number of STAs already connected to that AP [19].
In this approach beacons frame and the probe response frames contain some additional informa-
tion. This information include the number of stations already associated (Ni), RSSI value of the
probe request frame (Si), mean RSSI value for the set of stations associated to the AP (Mi). When
some STA wants to connect to AP it send the probe request frame to that AP. AP reply with the
probe response frame that contain the value of Ni, Si, Mi. Based on these factors STA find the
best AP and join that AP. For finding the best AP, STA calculate a weighted function Wi as
Wi = Di ∗ Pwi ∗ Pi
Where
Di = Mi − Si
Pwi is a weight proportional function based on differences from the mean value and absolute value
of the mean value
Pwi =
1 +Mi if Di ≥ 0
1 −Mi if Di < 0
Pi is the weight proportional to Ni
Pi =Ni∑nj=0Nj
This method needs changes in the probe request and beacon frame. It does not take into account
the interference by the neighbor AP or other interfering devices and load produced by the STAs while
calculating the weighted function . Choosing the AP based on the number of already associated
10
Figure 2.2: Basic System Diagram
STAs does not ensure performance improvement.
2.1.2 Distributed load balancing by hiding SSID
This technique distribute the STA among wireless routers such that their load is shared properly
and bandwidth utilisation get maximized by hiding SSID [16] .
In this approach they have added an IDS (information distribution center) in the WLAN envi-
ronment (figure 6). All the routers send their bandwidth usage information to the IDS periodically
(period is decided manually). Then IDS find the most optimal bandwidth having router for that
time period and only that router is permitted to broadcast SSID in that period. All router dont
broadcast their SSID in that time period. These routers are hiding their SSID in their beacon
packets. Now a new coming STA can see only one router selected by IDE. This process is repeated
periodically. Hence STAs are connected to efficient router and get distributed among APs.
This scheme helps the STAs to select the optimal router automatically. Experimental results
shows the 35.6%of the bandwidth utilization improvement compared to the conventional schemes
where STAs get connected to the router arbitrary. This is implemented in two modes
1. Proposed Round Robin (RR)
2. Proposed Optimal bandwidth (OB)
In proposed RR all routers broadcast sequentially for a time interval.In proposed OB, optimal
router is selected based on examining available bandwidth at every fixed time interval and that
11
router is allowed to broadcast its SSID, remaining hide their SSID.IDS (information distribution
center)
This methods has some bottlenecks like if there is sudden increase of the STAs, in that case all
the new STAs will get connected to the present AP resulting in poor performance. Also we are
required to modify the firmware of wireless router therefore firmware should be open source s/w.
2.1.3 Load balancing approach based on deadline miss ratio (DMR)
This method improve the efficiency of WLAN and provide QoS for real time traffic. It uses a
dynamic load balancing algorithm that check if the network is unbalance. This If the network is
unbalance then a node is de-associate from an overloaded AP and joined to another under loaded
AP (in a overlapping zone). Overlapping zone is the area in which STA receive the signal from both
the APs. This approach is based on the current bandwidth Utilization of AP and the Deadline miss
ratio (DMR)[5] [13]. DMR is used for providing the real time services. DMR is the ratio of the
number of packets that arrived after their relative deadline to total the number of arrived packets.
This approach WLAN contain a central entity called controller (figure 2.3) connected to the
backbone(wired network).This load balancing algorithm run on this controller. The load balancing
algorithm is triggered whenever performance degradation occurs or with the arrival of new stations
or with the mobility of existing stations.
Figure 2.3: WLAN Architecture
12
The controller keep the information about the main parameter used in the algorithm. These
parameters are
� Ui(k) is the utilization of the ith AP obtained in kth observation window. It is calculated as
the ratio of sum of the workload of all nodes to the bandwidth of the wireless link.
� U∗i (k) is the target utilization for the ith AP.
� U∗i (k) depends on DMR and α.
� APU (Access Point’s Utilization) is the mean value of the utilization in the various APs,
calculated by dividing the sum of the Ui(k) of every AP by the number of APs.
� The α parameter is dynamic, as it varies as a function of DMR measured at time k-1. The
values of the parameter are shown in figure 2.4.
� Calculate
Urel,i(k) = Ui(K)/U∗i (k)
Figure 2.4: Determining value of α
Network is said to be balance if for all AP i
δ1 <= Ui <= δ2 (1)
Where
13
δ1 is defined as APU - APU * α
δ2 is defined as APU + APU * α
If the DMR value increases, the α value will decrease and so width of the [δ1 − δ2] interval and
the chances of finding an overloaded AP increases.Algorithm can be explain as follow
1. Calculate Ui(k)
2. Calculate α , U∗i (k)
3. Check for unbalance N/W and if it is then identify overloaded AP(using equation 1)
4. Calculate balance index Bj [13] of each overlapping zone j as
Bj = (∑i=1
Urel,i)2/(n ∗
∑i=1
U2rel,i)
5. Select zone with minimum Bj
6. Find the most heavily loaded and the least loaded APs in that overlapping zone
7. Move a station (whose B/W value is nearer to difference of least and high loaded AP) form
most to least loaded AP.
This approach improves the bandwidth utilisation and performance for real time traffic in high
as well as low load scenarios by distributing the STAs in overlapping APs in overloaded conditions.
2.2 Modification of WLAN parameters
The parameter of the WLAN can be properly tuned in order to get high performance from an
individual AP. Like adjusting changing the physical layer parameter like slot time, SIFS (short inter
frame space), contention window (CWmin, CWmax ) delay in transferring packet can be decrease and
in some cases by adjusting some Mac layer parameters (like fragmentation, RTS-CTS) throughput
can be increase[12].
14
2.2.1 Physical layer parameter
Slot time, SIFS and CWmin can be assign low value. Decreasing the CWmin value decrease the back-
off interval. SIFS will help the receiver to transmit the frame quickly. On whole we can conclude
that this will decrease the media access delay and the contention interval for frame transmission
and hence increase the transmission time [12].
Slot time and SIFS interval is limited by hardware. Slot time is generally the time the hardware
take to sense the channel and inferring, if medium is free or not. SIFS interval is the time hardware
take to deliver the packet to upper layer and reset from reception mode to transmitting mode.
Hence they can be decrease to a limit only. CWmin is limited by load on network. If there are less
users on the network then CW min can be set to low value, but if the number of user get increase
then due to the small CW min probability of choosing the same back-off interval will get increase
which will increase the number of collision resulting in the decrease in the performance.
2.2.2 MAC layer parameter
In high error prone networks lower Fragmentation threshold increase the performance as in case of
the loss or error in any packet we will need to transfer small packet again. If the network is not
more prone to errors the lower Fragmentation threshold can cause Adverse effect as each packet need
ACK. Enabling RTS-CTS increases performance in highly loaded network as it solve the hidden
terminal problem so the collision become very less( collision will be of only RTS , CTS packets) and
hence saving bandwidth.
2.2.3 PCF or DCF
Most of the WLAN are configured to use the DCF as most of the time many station does not
transmit or wants to use WLAN. But in case of DCF when number of users competing for wireless
medium increases, number of collisions also gets increase[1]. With increase in collisions the back-off
time and retransmission of packets get increase resulting in performance decrease.In case of the
PCF number of collisions does not increase with increase in load as it uses polling. Under heavy
load PCF out perform DCF as shown in figure 2.5.
15
Figure 2.5: Comparison of PCF and DCF, source [17]
2.3 Comparison of various methods
When it comes to large number of clients all approaches comes to using multiple access points.
Many approaches has been proposed for distributing the load in WLAN having multiple AP. Some
of the algorithm suffer from interference problem and some suffer from the uneven distribution of
clients. As these all approach focus on using the Multiple access point, installation cost is high. So
if we can provide the connectivity to large number of clients using single AP, that will be great.
16
Chapter 3
CONTRIBUTION
3.1 Problem Statement
Number of tablets or WiFi devices that can be connected to a WiFi access point are limited to 30-40
(approx). So if we want to support many users then we have two ways, either use the multiple APs
or use single AP and let the clients access the WLAN service as per need. In the former method cost
of using multiple access point is high, load balancing problem also occurs. When we use multiple
APs we can provide provide the persistent connectivity while in the later method STAs will be
connected to AP for some time only. After that time they will disconnect from AP and give chance
to other STAs to get the WLAN services.
Hence in this project we are trying to explore the possibility of using single AP to provide
connectivity to many tablets (STAs). Let there are N tablets and an AP, then we need to ensure
all the N tablets get desired data(mostly text files, size ≤ 1Mb) downloaded and uploaded to server
using single AP. we are going to estimate the latency involved in conducting quizzes specially the
whole set-up time that include connection of all tablets to the APs and distribution of quiz papers
and uploading the quiz answers.
3.2 Problem Solving Approach
all the approach of using multiple APs are also not very scalable as with increase in number of
STAs (tablets) we need to deploy another AP. In this Project we are going to handle the class room
scenarios where everyone don’t need to be connected to the AP for long time. We have thought of
17
a new approach for our scenario that can handle large number of STAs with a single AP. We are
calling this algorithm as random batch-mode connectivity algorithm.
Our approach can be understood well by taking the scenario of quiz in the exam hall having large
number of tablets (STAs). As we know the maximum simultaneous connection(≈ 35) are limited
to an AP in WLAN. So we will let the clients use WLAN services in sequential manner. In this
approach first any of the 25 STAs (slightly less than the maximum limit of efficient simultaneous
connection to AP) will get connected to an AP. They will download their contents(in this case
quiz paper). After Downloading their quiz paper they will release their WiFi connection. Now
another group of 25 students can get connected to AP. They will also release the connection after
downloading the quiz. This process will be repeated until everyone download the quiz paper. In
this way everyone can download the data in very short span of time using single AP only.
Here for breaking the connection with AP, we switch OFF WiFi of tablet. In this way we get
connection breakage and also no other packet transfer between AP and tablets hence minimum
interference also. For selecting the 25 people, we are using the random number generator. If there
are N users then we calculate the number of rounds(R) by dividing total users(N) by 25. All clients
will get service in R rounds and in each round ≈ 25 users will get service. Now each client generate a
random number between (0,R]. If number generated is equal to 0, user switch ON its WiFi, connect
to network, download quiz and then again switch OFF WiFi. If the number generated is not equal
to 0 then client sleep for that round. On start of next round he again generate new random number
in range (0,R-1]. Same process is repeated until he gets 0. The number of round(R) is decrease by
1 after each round because in every round (≈ 25) users get to switch ON their WiFi and download
the quiz. In this way each client will get connect to network in utmost R rounds. The round time
is kept equal to the time tablet take to switch ON WiFi, connect to network and download quiz.
Lets say this time is equal to t. So if their are N users then everyone will finish download in R ∗ t
time.
The random batch-mode connectivity algorithm is shown in Table 3.1 :
18
Table 3.1: Random batch-mode connectivity algorithm
TotalUsers(N)
NumberOfConnectionRounds(R)
if(N<35) R=1;
else {
R = N/25;
extraClients=N%25;
// increase R by 1 if the extraClients are more than 10
if(extraClient >9) R++;
}
while(R>0){
connectProbability= generate_random_number(R);
if(connectProbability ==0){
switch_ON WiFi;
connect to network;
break;
}
R--;
sleep(t);
}
generate random number(R) is a function provided by android library to generate random num-
ber [7] [11]. It returns a random integer in range(0,R]. The seed for the random number generator
is the current time in nano seconds. Therefore as long as seed is different random number sequences
will be different. The probability of users getting 0 as random number is (1/R). So if total users
are N then with probability (1/R) only 25 people will get the chance of switch ON their WiFi.
As we are using random numbers, it is not certain that in every turn only 25 users get 0 random
number. Number of users can be more or less then the 25. In order to find out the random number
distribution in our random batch-mode connectivity algorithm, we simulated the distribution of the
random number. For this simulation we ran a process that generate the random number in given
range. We repeated this process for 100 times. Figure 3.1 shows the simulation that we ran for
100 clients and therefore 4 rounds, figure 3.2 shows the simulation that we ran for 200 clients and 8
rounds , figure 3.3 shows the simulation that we ran for 225 clients and 9 rounds. Detail experiment
19
report can be seen in Appendix A.
0
5
10
15
20
25
30
35
40
1 2 3 4
No
. o
f st
ud
en
ts
Rounds
simulation1
simulation2
simulation3
simulation4
Figure 3.1: Simulation of Random Numbers for 100 clients
here are some important conclusion from this simulation
1. In each round mostly we got number of users getting 0 between 20-30.
2. Very few times we got numbers of users lesser than 18 and greater than 30.
3. The number of users that can get proper connection from AP are 35-40. We tested for 35. So
if in some case the number of users are greater then 25, it can be managed.
4. As the human speed is mostly different hence we took different time in filling the front end
information while running app. Therefore we will get the different seed for mostly each tablet.
20
0
5
10
15
20
25
30
35
40
45
1 2 3 4 5 6 7 8
No.
of s
tude
nts
Rounds
simulation1
simulation2
simulation3
simulation4
Figure 3.2: Simulation of Random Numbers for 200 clients
The architecture of the above mention approach will consists of two main components(figure 3.4).
Server application and client application.
Server application will run on the server machine connected to wired network. Server will keep
the quiz question paper and also collect the quiz answer papers. Client application will run on the
STA machine. Client application help STA in downloading quiz file and uploading quiz answers.
The communication between the client and server will follow the protocol shown in figure 3.5.
The client application switch ON its WiFi based on the random batch-mode connectivity algo-
rithm. After that it connect with the server and request the quiz file. Once it download the quiz
file, it switch OFF its WiFi. Now When it has to upload the answer it again switch ON its WiFi
using random batch-mode connectivity algorithm, upload answer and again switch OFF its WiFi.
21
0
5
10
15
20
25
30
35
40
1 2 3 4 5 6 7 8 9
No.
of s
tude
nts
Rounds
simulation1
simulation2
simulation3
simulation4
Figure 3.3: Simulation of Random Numbers for 225 clients
Let say there are N tablets in the classroom. All the tablets will try to connect to AP, but
only K tablets (≈ 25) will be able to connect to the AP, and remaining N-K will wait for WLAN
service. Now the Client application will send the download request to the Server and wait for server
response. As client get permission from the server it will download the quiz (≈ 100Kb). After the
download, client application will release the connection( by switch OFF WiFi). Time taken by a
client to download the quiz is let say t. Where
t =0.1Mb
27Mbps
Here 27 Mbps(approx) is the bandwidth of the AP that can be actually used for downloading /
uploading data (considering 802.11g Router) and K ≈ 25 , hence t comes out 37 milliseconds when
22
Figure 3.4: Architecture Diagram of proposed model
their is no load and interference.In case of load the bandwidth of AP drops to 1-2 Mbps. so
t =0.1Mb∗1Mbps
hence time t will be ≈ 10 milliseconds.
Time required in switch ON WiFi of tablet is ≈ 2.5 − 3seconds, time to establish connection with
server is ≈ 2seconds.
Hence total time require for completing one round will be
3 + 2 + 25 ∗ 0.01 = 5.25 seconds
Therefore all 200 tablets can download the quiz papers in 52.5 sec.
As for quiz paper distribution, x minute(x ≤ 5minute) time is acceptable so our proposed method
can be very helpful.
23
Figure 3.5: interaction Protocol between Client-Server Application
24
Chapter 4
SOFTWARE IMPLEMENTATION
We have implemented the whole software for the quiz conduct in the class room. This software
consist of two parts
1. Server Module
2. Client Application
Figure 4.1: Software Architecture Diagram
25
The server module is a web interface designed for examiner or instructor. It helps in creating
and storing the quizzes and quiz responses submitted by the student. Software requirement for the
server are:
1. Apache server
2. PHP
3. Mysql
The clients application is an android application(.apk) that run on the tablet and use by the
clients to give the quiz.
1. Android tablet
4.1 Server Module
Server module is designed for the instructor for creating and storing the quizzes and quiz responses
submitted by the student. It provide the web interface using which instructor can easily regulate
the quiz conduct in the classroom.
Main component of the server are :
1. Quiz Create Module
2. Database
3. Upload/Download Module
4. Web Interface
Relation between various module can be understand by using figure 4.2.
4.1.1 Quiz Create Module
This module provide the functionality to add, modify question in the database and creating quiz.
While creating quiz instructor need to select questions from the database and provide the quiz
name(unique), time duration of quiz, and a key. Quiz name is required to uniquely identify each
26
Figure 4.2: server intersection Diagram
quiz, duration of quiz is the time for which quiz is going to be displayed, and key is required in
order to synchronise the start of the quiz among the student. After downloading the quiz student
need the key for opening the quiz, without key quiz will not be displayed.
4.1.2 Database
Database is used for storing the questions so they can be use later for creating the quizzes.
Database is also use to store the quizzes and the responses submitted by the students. Tables in
database are quiz, questions, quiz questions, Quiz response (figure 4.3).
Quiz table contain the name of all the quizzes and assign an unique id to it. Questions table
contain all the question added by the instructor. This table is use to get questions for creating the
quiz. Quiz questions table contain the questions of quiz. Quiz response table contain the answers
submitted by the students.
27
Figure 4.3: Database Relation diagram
4.1.3 Upload/Download Module
This module contains the quiz file for the quiz. This quiz file is create by Create Quiz Module
and download by the students during the quiz. Quiz file is .json file. Quiz file contain quiz name,
start key, quiz time duration and questions along with options and their answers. Detail structure
of quiz file can be seen in Appendix A.
This module also facilitate the submission of the answers by the students. Submitted answers
contains the roll number of student, quiz name, response, and mac address of tablet.
4.1.4 Web Interface
This is the dashboard provided for the instructor to add questions, create quiz, view responses
(figure 4.4).
28
Figure 4.4: Web Interface for creating and managing Quizzes
4.2 Client module
This module is in the form of the android application(Aakash Quiz.apk). This application enable
the complete quiz conduct in the wireless network. This application conduct two type of quizzes,
Normal quiz and Spot quiz(idea is inspired by Prof. Sudarshan S. IIT Bombay). Normal quiz
mode figure 4.5 shows all the questions along with answers. In Spot quiz mode figure 4.6 questions
and options are shown by instructor on the projector, application is use for collecting responses only.
Client application consist of six major parts. They are
1. Front end
2. Downloader
3. Quiz conduct
4. Uploader
5. Connection Manager
6. Show Result
Relation between these parts and there role can be understood by using the figure 4.8.
29
Figure 4.5: Normal quiz mode Interface
4.2.1 Front End
It shows the GUI to the user for entering the basic parameter required for quiz figure 4.7. Basic
parameters are roll number of student, quiz name, URl, total number of student present for the
quiz, and mode of quiz.
Roll Number of the user is use to identify the user on the server side. Quiz Name is required in
order to identify the quiz file on the server. Taking quiz name from the user help in creating and
keeping many quiz file on server at any time. URl is need to identify the server hosting the quiz. It
is not hard coded in the application because now server can be hosted on any computer connected
to network. Total student is required in in order to know the number of student present in the
class. This field is used to estimate the number of Rounds(or batches) required for all clients to get
connect to AP for Downloading the quiz. Quiz mode is required to select the quiz type. Quiz mode
tells if quiz is Normal quiz or spot quiz. Mac address is collected so that we can map roll number
to mac address for double authentication check on user.
4.2.2 Downloader
This module facilitates the downloading of the quiz file from the server. After downloading the quiz
it parse the quiz file for the conduct of quiz. Whole quiz is downloaded in the form of the json file.
30
Figure 4.6: spot quiz mode Interface
It uses the android http client [20] to download the quiz. The http client library is provided by
the android , Its parameter can we change as per requirement. Once the downloader is started it
download the quiz.
4.2.3 Quiz conduct
This module is responsible for the conduct of the quiz. It shows all the objective questions and
their answers to the user. There are two ways in which quiz can end, by pressing finish button or
by time over. User can press Finish button any time. It will end that quiz at that instant and his
responses will be collected. In time over, quiz will close automatically after the given duration of
the quiz and responses will be collected.
After the end of quiz, answers are bundle up and score is generated.
4.2.4 Uploader
This module upload the responses submitted by the user. First it switch on the WiFi with the
help of connection manager and as it connect with the AP it upload the responses along with the
roll number, mac address of tablet and quiz name. After uploading the answers WiFi is switch
OFF.
31
Figure 4.7: Front End Interface of Aakash Quiz.apk
4.2.5 Connection Manager
This module manages two most important functions:
1. When to enable WiFi
2. Checking if tablet is connected to AP
As the application start, it switch off the WiFi of tablet if number of users are more than 35.As
user enter the all parameter at the front end, WiFi of tablet is switch ON as per the random
batch-mode connectivity algorithm. The seed to the random number generator is the current time
in nano second. If the time is same it will generate the same Random number sequence. But due
to human differences and different time setting and is very rare.
This module use the WiFi functionality provided by the android library to switch ON/OFF
WiFi of tablet. In Aakash tablet the time require to switch on the WiFi vary form 2 to 3 seconds.
The downloading time of the quiz depend on the quiz file size, AP bandwidth and number of clients
accessing the network. the effective AP bandwidth is approximately equal to 27 Mbps for 802.11g
router. In case of heavy load (number of clients accessing the router are high ) the bandwidth falls
to very low due to high interference and collision. Hence be are keeping the download time to 2
seconds. In this application the sleep time is set to the 5 seconds i.e 3 second for switch ON the
WiFi and 2 second to connect to network and download quiz file.
32
As the WiFi is switch ON tablet is connected to the nearest AP. Now network connection is
checked and as tablet connect to some network, Downloader or Uploader is called.
Show Result
The Quiz score of the user is shown on the screen along with the notice that his response has been
uploaded. and asked to press QUIT button. Once he press QUIT button application closes.
4.3 Limitation
Their are some limitation of the software.
1. Tablet need to connect to the AP once before the start of quiz so that tablet can know easily
the network to connect to. It will save some time of searching network when user switch on
wifi.
2. Aeroplane mode should be Switch OFF otherwise it will take more time in switch ON WiFi.
3. Number of student are taken from the user.
4. If tablet is connected to Laptop at the time of quiz then user can view the answers to the
quiz questions using eclipse. So laptop should not be use during quiz.
33
Figure 4.8: Aaksh quiz application Flow Diagram
34
Chapter 5
EXPERIMENTAL RESULT
We have conducted some experiment using Aakash Quiz application to measure the
a) The latency involved in downloading the quiz file by all tablets
b) The latency involved in uploading the quiz responses by all tablets
c) Number of tablets able to give quiz successfully
d) Increment in the quiz download/upload time with increase in the number of tablets.
CPU RAM SoftwareClient(Tablet)
Server(laptop)
Intel-i3M370
3.7GiBUbuntu 12.04Apache2PHP5
WirelessRouter(TP-LinkWA901ND)
AtherosAR7240
32 MB OpenWRT
Table 5.1: Hardware And Software
Experiment are performed by running the Aakash Quiz.apk application on the tablets and server
on the laptop. Detail configuration of server and client device(tablet) is shown in table 5.1. The
experiment arrangement is shown in the figure 5.1. Wireless router was placed in the center of
the room. We have conducted the experiment in an auditorium having capacity of 222 people.
35
Figure 5.1: Experimental Setup
Only one wireless access point was used in the all experiments. Table 5.2 shows the results of the
experiments.
Tablets QuizSize
SuccessfulDownload
DownloadTime
SuccessfullUpload
UploadTime
OverallSuccess
34 1.8KB 34 18 34 51 3473 2.8KB 73 142 71 80 71103 2.8KB 101 99 100 44 100123 2.8KB 122 50 120 86 120132 1.8KB 129 54 126 68 126
Table 5.2: Experiment Results
We have compared the result of the these experiments. Figure 5.2 shows the the comparison of
number of user who have downloaded quiz and uploaded quiz responses successfully with increase
in number of users . Figure 5.3 shows the total of number of user who have completed quiz
successfully with increase in number of users. Figure 5.4 shows the comparison of time required by
user to download quiz and upload quiz responses successfully with increase in number of users.
Here time required for download and upload is getting slightly increase with increase in number
of users. In initial cases the time is more in comparison to the number of users that is because
the people who were participating in quiz took some time in reading quiz question and submitted
answers lately.
Figure 5.5 shows the the comparison of number of successful download in interval of 10 seconds.
36
Figure 5.2: Total Users Vs Number of Successful download upload
Figure 5.6 shows the the comparison of number of successful upload in interval of 10 seconds.
37
Figure 5.3: Total Users Vs total Successful Users
Figure 5.4: Total Users Vs time taken in download and upload
38
0
5
10
15
20
25
30
35
40
10 20 30 40 50 60 70 80
No. o
f dow
nload
s
Time interval (sec)
73 tablets103 Tablets123 Tablets132 Tablets
Figure 5.5: Time Interval Vs successful download
39
0
5
10
15
20
25
30
35
40
10 20 30 40 50 60 70 80
No. o
f uplo
ads
Time interval (sec)
73 tablets103 Tablets123 Tablets132 Tablets
Figure 5.6: Time Interval Vs successful upload
40
Chapter 6
CONCLUSION
Many approaches have been proposed for supporting the multiple user using multiple APs. These
all approaches are important if STA want regular connectivity. But we don’t need regular con-
nectivity in the classroom scenario hence these all methods are not efficient for our case as using
multiple APs increase the cost and also causes scalability issues. No particular approach has been
proposed specifically for the class room environment where student don’t need persistent (regular)
connectivity, they need to download or upload data for some time only.
We have proposed a model to solve this problem. Our solution is based on three main factors.
They are :
1. In classroom quiz conduct student dont need persistent connectivity.
2. Quiz data size is small.
3. Functionality of disconnecting the WiFi driver is available on the tablets.
We have solved the synchronisation problem using the start-key. Quiz will start only when
correct key is presented. Experimental results shows that our solution is good for class room
scenario as it is providing the connectivity to many users with an AP and scalability problem is also
not issue. When number of clients get increases too much (more than 200) in that case download
time and upload time can increase as number of rounds for connection also get increases.
41
Chapter 7
FUTURE WORK
The server side of the software can be connected with clicker database or moodle database to
launch quizzes and collecting response. Synchronisation of the quiz start time can be tackle in
others way also. For starting the quiz at same time by all the STAs, we can send the quiz start
time along with the quiz file. This time will be equal to the time of distributing the quiz paper to
all the STAs plus some extra time. We have tested application upto 132 clients, in future testing
can be increase upto 250 or 300 users. Presently we are sending answers in plain text to users in
the quiz file. So we can encrypt these answers. It will be good for security purpose.
We will try to find the suitable value of the physical layer parameters like Slot time, SIFS,
contention window (CWmin, CWmax), back-off interval for tablets in terms of the number of tablets
(N). So that we can improve the performance of individual AP and hence WLAN.
42
Bibliography
[1] J.N. Al-Karaki and J.M. Chang. A simple distributed access control scheme for supporting qosin ieee 802.11 wireless lans. In Wireless Communications and Networking Conference, 2004.WCNC. 2004 IEEE, volume 1, pages 213 – 218 Vol.1, march 2004.
[2] K. Bakanoglu, S. Tomasin, and E. Erkip. Resource allocation in wireless networks with multiplerelays. In Signals, Systems and Computers, 2008 42nd Asilomar Conference on, pages 1501–1505, oct. 2008.
[3] A. Balachandran, P. Bahl, and G.M. Voelker. Hot-spot congestion relief in public-area wirelessnetworks. In Mobile Computing Systems and Applications, 2002. Proceedings Fourth IEEEWorkshop on, pages 70 – 80, 2002.
[4] Kameswari Chebrolu. Cs653 mobile computing course. http://moodle.iitb.ac.in/
moodle2011/course/view.php?id=3022, Sept 2011.
[5] M. Collotta, L.L. Bello, E. Toscano, and O. Mirabella. Dynamic load balancing techniquesfor flexible wireless industrial networks. In IECON 2010 - 36th Annual Conference on IEEEIndustrial Electronics Society, pages 1329 –1334, nov. 2010.
[6] Android Developer. Android apis. http://developer.android.com/guide/components/
index.html, April 2013.
[7] Android Developer. Random number generator api android. http://developer.android.
com/reference/java/util/Random.html, April 2013.
[8] [email protected]. handling json object. http://www.sharingsession.com/
php-script-create-json-mysql, April 2013.
[9] Marko Gargenta. Wifi apis. http://marakana.com/forums/android/examples/40.html,April 2013.
[10] grepcode.com. Android api source code. http://grepcode.com/search?query=android&n=,April 2013.
[11] grepcode.com. Random number generator api android. http://grepcode.com/file/
repository.grepcode.com/java/root/jdk/openjdk/6-b14/java/util/Random.java,April 2013.
[12] W. Hneiti and N. Ajlouni. Performance enhancement of wireless local area networks. InInformation and Communication Technologies, 2006. ICTTA ’06. 2nd, volume 2, pages 2400–2404, 0-0 2006.
43
[13] Issam Jabri, Nicolas Krommenacker, Adel Soudani, and Thierry Divoux. Load balancingapproach for wireless ieee 802.11 qos enhancement. In Proceedings of the 11th IFIP TC6international conference on Personal Wireless Communications, PWC’06, pages 88–99, Berlin,Heidelberg, 2006. Springer-Verlag.
[14] Hyosu Kim, Minsub Lee, Wookhyun Han, Kilho Lee, and Insik Shin. Aciom: Applicationcharacteristics-aware disk and network i/o management on android platform. In EmbeddedSoftware (EMSOFT), 2011 Proceedings of the International Conference on, pages 49 –58, oct.2011.
[15] Wen-Hsing Kuo and Wanjiun Liao. Utility-based resource allocation in wireless networks.Wireless Communications, IEEE Transactions on, 6(10):3600 –3606, october 2007.
[16] Hukeun Kwak, Cheongghil Kim, Younghyo Yoon, Myungwon Kim, Dongseung Kim, andKyusik Chung. A method for optimal bandwidth utilization in ieee 802.11 wlan networks.In Information Networking, 2008. ICOIN 2008. International Conference on, pages 1 –4, jan.2008.
[17] Shruti Mahajan Leena Chandran-Wadia and Sridhar Iyer. Throughput performance of thedistributed and point coordination functions of an ieee 802.11 wireless lan. Technical report,IIT Bombay, 2002.
[18] C. Mbarushimana and A. Shahrabi. Comparative study of reactive and proactive routingprotocols performance in mobile ad hoc networks. In Advanced Information Networking andApplications Workshops, 2007, AINAW ’07. 21st International Conference on, volume 2, pages679 –684, may 2007.
[19] I. Papanikos and M. Logothetis. A study on dynamic load balance for ieee 802.11 b wirelesslan. In Proc. COMCON, volume 2001, 2001.
[20] by Harshad Oak Pro Jakarta Commons. Android http client. http://www.theserverside.
com/news/1365153/HttpClient-and-FileUpload, April 2013.
[21] stack overflow. Android programming problems. http://stackoverflow.com/search?tab=
relevance&pagesize=30&q=%5bandroid%5d%20downloading, April 2013.
[22] MySql technical team. Using mysql. http://dev.mysql.com/doc/refman/5.5/en/, April2013.
[23] Wikipedia. Ieee 802.11e-2005. http://en.wikipedia.org/wiki/IEEE_802.11e-2005/, April2012.
44
Appendices
45
Appendix A
A.1 Quiz file Structure
We are using the Json file in the application. Sample file is as:
[”startKey”:”lmn”,”timeDurationOfQuiz”:”10”,”quizName”:”testing123”,
”question”:”Who is Sachin tendulkar ?”,”option1”:”cricketer”,”option2”:”football”,”option3”:”tennis”,”option4”:”handwall”,”answer”:”1”,
”question”:”What is capital of M.P.”,”option1”:”bhopal”,”option2”:”indore”,”option3”:”jabalpur”,”option4”:”gwalior”,”answer”:”1”,
”question”:”where is gateway of India ?”,”option1”:”Mumbai”,”option2”:”Channai”,”option3”:”Delhi”,”option4”:”Bhopal”,”answer”:”1”]
46
A.2 Simulation results
Random Number simulation results are shown below. Here range (R) means number of roundare R and Number of clients are R ∗ 25.
For Example:for range 431 21 23 25 0 0 0 0 0 0
It will be read asNumber of clients are R ∗ 25. So in this case total clients are 4 ∗ 25 = 100first line 31 21 23 25 0 0 0 0 0 0will be interpret as number of clients got 0 in round 1 are 31, in round 2 are 21, in round 3 are 23and so on. getting 0 means now they are allowed to Switch ON wifi.(here only first 0 of any clientis counted after that they are ignored).
for range 420 30 24 26 0 0 0 0 0 0
for range 421 18 30 31 0 0 0 0 0 0
for range 426 27 25 22 0 0 0 0 0 0
for range 425 21 29 25 0 0 0 0 0 0
for range 423 29 17 31 0 0 0 0 0 0
for range 420 27 32 21 0 0 0 0 0 0
for range 421 25 23 31 0 0 0 0 0 0
for range 425 25 24 26 0 0 0 0 0 0
for range 434 19 17 30 0 0 0 0 0 0
for range 425 24 26 25 0 0 0 0 0 0
47
for range 419 30 21 30 0 0 0 0 0 0
for range 423 20 30 27 0 0 0 0 0 0
for range 424 28 31 17 0 0 0 0 0 0
for range 423 25 26 26 0 0 0 0 0 0
for range 430 19 25 26 0 0 0 0 0 0
for range 826 21 20 31 28 21 27 26 0 0
for range 826 38 20 21 21 19 26 29 0 0
for range 827 25 17 23 20 23 37 28 0 0
for range 829 25 25 26 25 19 31 20 0 0
for range 823 21 25 33 21 20 33 24 0 0
for range 833 20 22 28 20 28 28 21 0 0
for range 821 15 35 21 20 37 20 31 0 0
for range 816 30 27 22 36 29 21 19 0 0
for range 827 25 22 25 26 34 18 23 0 0
for range 819 25 36 24 31 18 22 25 0 0
for range 826 22 28 24 27 18 27 28 0 0
48
for range 824 23 19 24 27 22 27 34 0 0
for range 825 22 27 33 22 22 31 18 0 0
for range 822 21 26 28 26 24 25 28 0 0
for range 818 28 25 26 22 28 26 27 0 0
for range 821 33 20 26 27 16 31 26 0 0
for range 828 20 27 29 18 25 25 28 0 0
for range 823 23 23 28 29 33 21 20 0 0
for range 832 26 26 20 26 26 22 22 0 0
for range 924 30 22 21 27 31 16 32 22 0
for range 924 27 29 28 21 28 26 19 23 0
for range 929 20 25 26 24 14 35 28 24 0
for range 919 23 31 20 24 22 25 27 34 0
for range 926 26 25 35 34 19 19 21 20 0
for range 924 27 22 25 18 35 23 24 27 0
for range 924 27 21 35 23 20 23 30 22 0
for range 9
49
18 18 26 26 26 31 28 21 31 0
for range 924 24 26 26 28 26 30 25 16 0
for range 920 28 18 24 30 25 32 20 28 0
for range 928 19 15 25 30 25 24 32 27 0
for range 926 29 25 27 20 21 24 27 26 0
for range 924 33 24 25 20 30 25 32 12 0
for range 931 20 33 22 27 27 24 20 21 0
for range 932 18 26 27 25 24 26 20 27 0
for range 927 33 29 25 20 23 26 24 18 0
A.2.1 facts from simulation
Average number of clients getting connected in each round for total clients 100 over 15 experimentsround 1= 22round 2=22round 3= 24round 4= 21
Average number of clients getting connected in i each round for total clients 150 over 15 experi-mentsround 1= 24round 2=23round 3= 24round 4= 24round 5= 25round 6= 23
Average number of clients getting connected in each round for total clients 200 over 15 experimentsround 1= 24round 2=25round 3= 24
50
round 4= 23round 5= 25round 6= 27round 7= 25round 8= 24
51