MODELLING & SIMULATION OF A VIRTUAL CAMPUS

A Case Study Regarding the Open University of Catalonia

Angel A. Juan, Pau Fonseca, Joan M. Marquès, Xavi Vilajosana

Dep. of Computer Science, Open University of Catalonia, Rambla Poblenou, 156, 08018, Barcelona, Spain

Javier Faulin

Dep. of Statistics and OR, Public University of Navarre, Navarre, Spain

Keywords: Modelling & Simulation, Distributed Systems, Network Implementation Choices.

Abstract: In this paper we present a case study regarding the modelling and simulation of a real computer system

called Castelldefels. This system gives support to the Virtual Campus of the Open University of Catalonia

(UOC), an online university that offers e-learning services to thousands of users. After analyzing several

alternatives, the OPNET software was selected as the convenient tool for developing this network-

simulation research. The main target of the project was to provide the computer system’s managers with a

realistic simulation model of their system. This model would allow the managers: (i) to analyze the

behaviour of the current system in order to discover possible performance problems such as bottlenecks,

weak points in the structure, among others, and (ii) to perform what-if analysis regarding future changes in

the system, including the addition of new Internet-based services, variations in the number and types of

users, changes in hardware or software components, etc.

1 INTRODUCTION

In order to analyze computer systems and networks

performance, both analytical and simulation

methods can be used. Analytical methods are based

upon mathematical analysis that characterizes a

network as a set of equations. This approximation

usually implies considering several restrictive

assumptions, which tend to be not very realistic,

since networks are complex systems formed up by

hardware and software (protocols, applications,

queuing policies, etc.). Alternatively, simulation

techniques can also be used to model in detail the

dynamic nature of real computer networks (Law,

2006) (Banks et al., 2001). Simulation allows

engineers to test different network designs, even be-

fore the network physically exists, and to perform

what-if analysis with models of the already existent

networks without exposing them to failures or

inoperative periods.

The Open University of Catalonia (UOC) is an

online university located at Barcelona (Spain) with

more than 37,000 community members, including

students from Spain and Latin America, professors,

and managers among others. With this amount of

potential intranet users, performance fine-tuning of

the computer system that gives support to the UOC

Virtual Campus becomes the most important task for

system managers. For that reason, a team formed by

managers, professors and students started the so-

called Castelldefels Project. The main objective of

this project is to improve the system performance

levels and, consequently, to increase the quality of

the service offered to users of the UOC Virtual

Campus. This is carried out by selecting appropriate

values for configuration parameters such as network

topology, hardware devices, queuing and balancing

policies, protocols, etc. (Kurose and Ross, 2005)

(Peterson and Davie, 2003).

The computer system makes use of load

balancing mechanisms, which allow a convenient

load distribution (requests from distinct users)

among different available servers. Two dedicated

hardware devices perform this load balancing task.

While one of the load balancers is operating, the

other is in stand-by status. Thus, maintenance tasks

can be done without having to stop the web service

and, moreover, this service will be available even if

187

A. Juan A., Fonseca P., M. Marquès J., Vilajosana X. and Faulin J. (2008).

MODELLING & SIMULATION OF A VIRTUAL CAMPUS - A Case Study Regarding the Open University of Catalonia.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - ISAS, pages 187-190

DOI: 10.5220/0001671501870190

 SciTePress

the active load balancer fails. These load balancers

are responsible for assuring readiness, at any

moment, of web services (HTTP, HTTPS, FTP,

SMTP and other proprietary applications). Different

load assignment policies can be set up in the

balancer. Thus, the balancer device decides which

available server will pay attention to an incoming

user request based on one of the following selection

criteria: random criterion, sequential criterion, the

least-loaded-server criterion, or the smallest-

response-time-server criterion. In order to check if a

concrete server is operative, the load balancer carries

out regular tests on the servers. This monitoring

process can also be configured with different

optional parameters, such as elapsed time between

two consecutive tests.

2 SYSTEM DESCRIPTION

When a user opens a web browser and types the

University URL, the user request is balanced among

several portal servers and a portal page is loaded into

the user browser. To complete that process, the

client request has already passed through the frontier

routers, crossed the firewalls, and arrived to the load

balancers, where a new session has been settled

down with an available portal server.

Once the user has introduced her login and

password and they have been validated in the

database, she gets access to the Virtual Campus.

This means that her request has been balanced and it

has finally arrived to an available front-end server.

There are about 25 front-ends servers in the

Castelldefels system. Between the load balancers

and the front-ends there are also two more hardware

devices: a web accelerator and two application

firewalls.

3 SIMULATION WITH OPNET

There are several discrete-event simulation programs

specially designed for network simulation. One of

these programs is the open-source OMNeT++

(Varga, 2001). After some preliminary studies,

though, we decided to use the proprietary software

OPNET for three reasons: (i) it seemed to be the

most widely tested, used and documented software

in the network simulation area (Chang, 1999)

(Aboelela, 2003) (Brown and Christianson, 2004)

(Qadan and Guizani, 2005), (ii) it offers an

outstanding library of network devices, and (iii) a

free license for academic and research use was

available from the software developer.

OPNET is a complete package composed by

several modules. It can be scaled from Local Area

Networks (LANs) to Wide Area Networks (WANs)

formed up by thousands of workstations.

For advanced research, OPNET Modeler offers

advanced tools for model design, simulation, data

mining and analysis. Using this software, it is

possible to edit the source code of any hardware

device included in the OPNET library of

components, which is provided by hardware

manufacturers such as Cisco or 3Com. OPNET

Modeler is based in a three-level design hierarchy:

(1) a network model, where networks and sub-

networks are defined; (2) a node model, where

node’s (hardware devices) internal structure is

defined; and (3) a processes model, where internal

node states and functioning can be defined by using

C/C++ programming (Svensson and Popescu, 2003).

4 MODELLING THE SYSTEM

At this stage of our project, we have focused on the

partial modelling of the Castelldefels architecture,

focusing ourselves on the Campus network, which is

the infrastructure mainly used by students and

professors. Additionally, we have reduced somewhat

the model size of the Campus network, assuming

that it has fewer servers and devices than the real

system has (25 front-ends, 3 mail servers, one

backup load balancer and up to 5,000 concurrent

users requesting HTTP, FTP and email services).

We have distributed our model in three levels.

The first level represents the WAN, which is

modelled by a set of geographical nodes. All these

nodes are connected to one special node, the

Castelldefels node, using an IP (Internet) cloud. At a

second level, we found the details of each node,

including the Castelldefels one (Figure 1).

Figure 1: Second Level - The Castelldefels node.

ICEIS 2008 - International Conference on Enterprise Information Systems

188

Each of the remaining subnodes is a representation

of a Metropolitan Computer Network (MAN), which

aggregates several LANs. These MANs configure a

third level in our model (Figure 2).

Figure 2: Third Level – Modelling the MANs.

The main components of our model are:

• Applications:

Each application defines a service

that can be executed in a workstation or in a

server. It also defines the load of the system. To

emulate the behaviour of the Campus application

–Web-based front-ends with file transfer and

email support–, we defined three OPNET

applications: HTTP (Heavy Browsing), FTP

(Low Load) and Email (Medium Load)

• Profiles:

A profile is a group of applications to

be used by some type of users such as students,

professors, etc. Each profile also defines the

statistical distributions for the simulation engine.

At this stage, only one general profile –with

support for all applications– was defined.

• Servers:

There were 9 servers to execute

applications: 6 front-end servers to emulate the

basic Campus activity using HTTP and FTP, and

3 email servers. All servers were directly

connected to the load balancer.

• Load balancer:

It is a device that decides which

server will pay attention to the next user request.

Servers are assigned to balanced applications.

Each application is balanced using its own

policy: Round-Robin or sequential for HTTP and

FTP, and Number-of-connections for email. The

Round-Robin policy assigns each incoming

request to the next server in the sequence, while

the Number-of-connections policy assigns tasks

according to the number of connected users.

• Frontier router:

All former devices were

connected to a router through a firewall, which

restricts the supported applications. The router

was the single connection to the Internet.

• User networks:

several networks were set up to

emulate students’ activity. Each network may

have a different number of users, from one to

several hundreds, and it was connected to

Internet through a gateway. Services were

directly requested to the load balancer.

Even when some simplification assumptions were

made in this initial model, we expected to obtain

interesting results that guide us in the development

of more detailed models.

5 SIMULATION RESULTS

After implementing our model in the OPNET

Modeler software, we have carried out several

simulation experiments regarding parameters such as

balancing policies, traffic load, response time, CPU

utilization, or failure-recovery time, among others.

The experiments allow us to study different

scenarios (what-if analysis) depending on different

configurations of the Castelldefels computer sys-

tem, which helps its managers to get a better

understanding of the system behaviour and, which

may be even more important, to be able to predict

changes in the system performance due to changes

in the system configuration: number of servers,

traffic loads, balancing policies, maintenance

policies, etc. For instance, Figure 3 shows a graphic

that predicts the response-time of the FTP

application depending on the balancing policy

(Minimum load versus Round Robin) and on the

number of concurrent clients (500 or 1,000).

Figure 3: Response time for the FTP application.

According to Figure 3, it seems clear that

response times do not significantly depend on the

selected balancing policy. Instead, they mainly

depend on the number of concurrent users. In fact,

this happens to be the most determining factor over

the response-time parameter. Other factors, such as

the number of servers in the system, seem not to be

as decisive, since the CPUs of the available servers

are not being pushed to their full capacity (Figure 4).

Similar conclusions are reached for the HTTP and e-

mail applications.

MODELLING & SIMULATION OF A VIRTUAL CAMPUS - A Case Study Regarding the Open University of Catalonia

189

Figure 4: CPU utilization for server 2.

As the previous examples may suggest, once the

model is implemented the number of simulation

experiments that can be performed to compare

different scenarios is almost unlimited, since

managers can change the system configuration in

every possible way including: number of clients,

number of dedicated servers, balancing policy, users

profiles, applications load, individual server failures

and recoveries, etc. Furthermore, our simulation

model soon results in an invaluable tool for the

Castelldefels system managers, which have the

ultimate responsibility over the quality of the service

offered by the UOC Virtual Campus.

6 ACADEMIC APPLICATIONS

It is expensive to set up a physical networking lab

for university students. Moreover, even if one

university made that investment, such labs would

have significant limitations when dealing with

different possible scenarios (what-if analysis). In

fact, it is virtually impossible to cover, using a

physical network, the wide diversity of existing

technologies and configurations.

For that reason, use of discrete-event

simulation, as a methodology to confront network

design and fine-tuning problems, is not only

interesting in the professional arena but also in the

academic one (Theunis et al. 2003). The current

level of computer software and hardware allows the

efficient application of simulation-based methods

and algorithms to network analysis, allowing a

major comprehension of networks’ internal

functioning process. Using simulation, students are

able to analyze alternative scenarios and designs

both for LANs and WANs.

7 CONCLUSIONS

When studying networks performance, simulation

techniques offer clear advantages over analytical

ones, such as: (a) the opportunity of creating models

which faithfully reflect the real network

characteristics and behaviour, and (b) the possibility

of obtaining additional information about the system

internal functioning.

We have used OPNET to develop a model of the

computer system that gives support to the UOC

Virtual Campus. The model allows us to experiment

with some what-if scenarios and to test how the

system will perform under different configurations.

Performing what-if analysis on the simulation model

before implementing changes in the real system may

avoid unexpected problems in the system day-to-day

behaviour.

REFERENCES

Aboelela, E., 2003. Network Simulation Experiments

Manual. Morgan Kaufmann

Banks, et al., 2001. Discrete-Event System Simulation.

Prentice-Hall

Brown, K., Christianson, L., 2004. OPNET Lab Manual.

Prentice Hall.

Chang, X., 1999. Network simulations with Opnet. In

Proceedings of the 1999 Winter Simulation

Conference, pp. 307-314

Kurose, J., Ross, K., 2005. Computer Networking: A Top-

Down Approach Featuring the Internet. Addison-

Wesley

Law, A., 2006. Simulation Modeling and Analysis.

McGraw-Hill

Peterson, L., Davie, B., 2003. Computer Networks. A

Systems Approach. Morgan Kaufmann

Qadan, O., Guisan, M., 2005. OPNET Lab Manual. John

Wiley & Sons

Svensson, T., Popescu, A., 2003. Development of labo-

ratory exercises based on OPNET Modeler. Master

thesis, June 2003. Blekinge Institute of Technology,

Sweden.

Theunis, J., et al., 2003. Advanced Networking Training

for Master Students Through OPNET Projects. In

Proceedings of the 2003 OPNETWORK Conference,

Washington D.C., USA

Varga, A., 2001. The OMNeT++ Discrete Event

Simulation System. In Proceedings of the European

Simulation Multiconference (ESM'2001). June 6-9,

Prague, Czech Republic.

ICEIS 2008 - International Conference on Enterprise Information Systems

190