Environment for Hybrid Simulation of Information Security Solutions

for Grid and Cloud-systems

Valeriy Vasenin, Vladimir Roganov and Andrey Zenzinov

Institute of Mechanics, Lomonosov Moscow State University, Michurinsky pr. 1, Moscow, Russian Federation

Keywords:

Grid Computing, Distributed Computer Systems, Information Security, Virtual Modeling, Simulation, Hybrid

Modeling.

Abstract:

Efﬁcient and safe use of distributed information systems based on Grid and Cloud-computing technologies

requires an assessment and optimization of their security level. For this purpose it seems appropriate to model

the studied systems and their behavior under different working conditions. The paper presents an approach

to modeling distributed systems in full-scale, simulation, virtual and analytical sense, and also describes the

possibility of their simultaneous use in hybrid mode simulation. Such approach allows to investigate the

behavior of distributed systems from different perspectives, taking into account features of the architecture,

software, purposes of these systems and operation conditions. The paper also describes the mechanisms of the

developed modeling software.

1 INTRODUCTION

In recent times, increasing attention is paid to soft-

ware tools for security of large-scale distributed in-

formation systems. Despite the fact that there are a

signiﬁcant number of multi-level tools for Grid and

Cloud-systems with such appointment, developing of

security systems is very relevant and important re-

search area. In order to provide the effective pro-

tection of such systems it is necessary to assess and

adjust their security at different levels of their archi-

tectural implementation. This fact led to the problem:

design and implement software tools for hybrid simu-

lation of the Grid and Cloud-systems which provides

uniﬁed model to present and simultaneously test sev-

eral levels of the distributed system.

By the hybrid modeling approach we shall mean

the conjunction of several different ways of model-

ing. This approach has already proven itself and is

widely used in various ﬁelds of research and engi-

neering which are closely related to complex systems,

particularly in the space industry. It allows to study

simultaneously a large number of different modes of

behavior of a complex system without a real launch of

a spacecraft. For Grid and Cloud-systems hybrid sim-

ulation we had chose Natural, Imitating, Virtual and

Analytical modeling approaches. To stress the hybrid

nature and wide range of simulation mechanisms the

abbreviation NIVA (by ﬁrst letters, means also Field

in Russian) was used as name for consolidated mod-

eling environment.

NIVA simulation environment does not require

high qualiﬁcation from users. The environment con-

tains GUI-based editors for visual deﬁnition of con-

ﬁguration of designed distributed systems and, more

important, their usage scenarios. It supports auto-

mated deployment of the conﬁgurations, interactive

simulation mode combined with graphical visualiza-

tion and batch control via command ﬁles, which was

written on a standard interpreted language mixed with

arbitrary decision logic. Usage scenarios of dis-

tributed systems are based on the template library of

typical behavior patterns. With use of these patterns

it is very easy to describe both normal and abnormal

work ﬂow of a distributed system, under conditions of

the various attacks.

Thus, the environment allows to perform hybrid

simulation of developed Grid/Cloud-systems in order

to obtain their characteristics in terms of performance

and information security, as well as support their op-

timization using multivariate non-interactive simula-

tion with relatively small resource consumption.

2 EXISTING APPROACHES

The main idea to simulate distributed computing sys-

tems is not new and already has a long history. Start-

255

Vasenin V., Roganov V. and Zenzinov A..

Environment for Hybrid Simulation of Information Security Solutions for Grid and Cloud-systems .

DOI: 10.5220/0005105502550260

In Proceedings of the 4th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2014),

pages 255-260

ISBN: 978-989-758-038-3

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

ing from epoch of grid concept a lot of simulation

tools have been created, mainly based on discrete-

event simulation, which, apparently, is the most versa-

tile approach and has worked very well in the area of

data communication networks. At present, the most

famous and widely used specialized tools are Sim-

Grid (Casanova et al., 2008), GridSim (Buyya and

Murshed, 2002), CloudSim (Rodrigo N. Calheiros

and Buyya, 2011), which are widely used for solv-

ing various tasks, such as optimization algorithms of

key system components (Yaser Mansouri and Buyya,

2013). As an example of such kind of software devel-

oped in Russia the system developed in the Russian

Institute for System Programming RAS (Grushin D.,

2008) should be noted. In these environments, dis-

tributed system typically represented by a collection

of objects belonging to its upper layer, that is, compu-

tational nodes, inter-node connections, etc.

For our purposes we chose SimGrid tool as base

simulator. Its role would be discussed in the sec-

tion 4. GridSim and CloudSim are written in Java,

and SimGrid code is C-based, so it seems more conve-

nient to perform large series of non-interactive exper-

iments. SimGrid tool supports both Grid and Cloud-

simulation, while GridSim and CloudSim are closely

oriented on simulation of Grid and Cloud-systems, re-

spectively, and we should use both of these two tools.

We should also note SimGrid graphical conﬁguration

editor, implemented as an Eclipse plugin.

Yet another approach, based on virtual simulation

is related to widespread virtualization support in hard-

ware. In this case, the model simulates the instru-

mental level using the set of virtual machines inter-

connected by a virtual networks. Nowadays there are

some papers about nested Cloud systems (Josef Spill-

ner, 2012) but completely modeling of Grid/Cloud-

technologies using virtual machines is poorly studied.

Of course, there are other approaches, such as an-

alytical and full-scale modeling. If the simulation aim

is to reach a large number of nodes (for example, to

simulate large subset of the Internet), and is focused

on the macro-level, then somebody can use the ana-

lytical methods such as Petri nets, etc. In contrast,

the full-scale modeling can help to study such subtle

phenomena in software as race condition vulnerabili-

ties. Each of above approaches is good at its level, but

none of them can fully and effectively cover the en-

tire spectrum of possible interactions that take place

in large distributed systems. Taking into account the

above considerations, it seems appropriate to combine

all of these approaches.

In addition, most of Grid and Cloud modeling

software requires a programmer qualiﬁcation to de-

ﬁne and analyze the activities of interest. For this

reason, the main priorities in the development of the

NIVA environment were hybrid simulation support

and operator comfort.

3 FUNCTIONALITY

Simulation environment NIVA for Grid / Cloud-

systems supports mixed live, discrete-event, vir-

tual and analytical modeling of distributed comput-

ing systems in both interactive and non-interactive

modes, with an ability to visualize process and post-

processing of results. The combination of different

models allows a comprehensive study of large dis-

tributed systems, as well as to immerse them in a re-

alistic context of the global network, namely:

• live (full-scale) simulation allows to evaluate the

real computational efﬁciency of computational

nodes and interconnection used on real hardware;

• virtual simulation helps to conduct the necessary

experiments with the distribution, assess the com-

patibility of the software, and perform effective

testing for known vulnerabilities;

• discrete-event simulation model is able to sim-

ulate the behavior of a signiﬁcant number of

Grid/Cloud nodes, as well as the consequences of

possible attacks and the propagation dynamics of

mortgage elements such as viruses;

• analytical modeling makes it easy to simulate very

large Grid systems, such as signiﬁcant part of the

Internet, by taking into account the probabilities

and intensity of corresponding dynamic system.

Operator’s work consists of selecting/editing

Grid/Cloud conﬁguration, selecting/editing the

usage scenario in the provided graphic editors,

boot conﬁguration before running the scenario and

run the simulation with or without visualization,

possibly followed by a detailed analysis of obtained

results. Interactive modeling process can be stopped

at any time moment by the operator (for example,

for a detailed study of the state of the simulated

system), run up the steps (in terms of virtual time)

or prematurely terminated. There is also possibility

to program an automatic pause/stopping trigger by

certain conditions.

The result of the series of launches in batch mode

is a report. Report helps to decide if the proposed

tools and protection strategies are suitable, as well as

make recommendations for further improvements.

Of course, any simulation environment have lim-

itations. For instance, restrictions on the size of the

SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

256

simulated systems dictated by the amount of avail-

able RAM and by the hardware platform perfor-

mance. However, the advantages of the hybrid ap-

proach sometimes let get round this limitation by pro-

viding an opportunity for analytical modeling of indi-

vidual large subsystems using empirical or statistical

data. As a result, the hybrid approach allows to cover

almost all cases considered Grid/Cloud-systems at the

cost of accuracy.

4 HYBRID SIMULATION

Common part for various simulators in our case is a

discrete-event simulation model as the most general.

This approach eliminates the need to answer difﬁcult

question of the direct interaction of many diverse sim-

ulators. It requires minimum effort to create a “simu-

lation representatives” for those entities that must in-

teract directly, but are located in different simulators.

As mentioned above we chose modern, actively de-

veloping SimGrid open-source framework as the ba-

sic simulator. It was extended with tools designed for

interactive visualization, graphical scenario editor and

scenario template library with deﬁnitions of typical

activities in the Grid/Cloud-systems.

Hybrid modeling allows to use multiple simula-

tors. Interaction of simulators can be organized in

different ways. In our system we decide to do it on

the basis of the discrete-event simulation model.

From the perspective of architecture it looks as

follows.

1. Master discrete-event based simulator (for this

case SimGrid) is the principal, and all entities (ex-

cept those which are modeled analytically) have a

representation as the nodes in the master simula-

tion model. But some also have representation in

the form of real processes inside virtual machines

or real nodes in the case of virtual or full-scale

modelling, respectively.

2. If a node is marked as running on an external

(slave) simulator, then SimGrid only reassigns

tasks to an external simulator and vice versa in-

stead of the typical logic trigger actions. The

external simulator communicates with SimGrid

through a simple communication protocol.

3. External simulators receive complete system con-

ﬁguration with assigned actions. Therefore, they

either interact with other nodes using “full-scale”

way (i.e. conduct real DDoS attacks or send real

malicious code), or just forward the message to

SimGrid, if the destination node has only single

imitation on the master side simulator.

4. Since the actions executed naturally instead of the

simulation in the case of virtual or full-scale sim-

ulation, implementation support for these actions

requires coupling logic. That is, when you re-

ceive a message with malicious code from a sim-

ulation model, full-scale simulator should trans-

form it into a real malicious code. Therefore, in

many cases it may be easier to perform indepen-

dent modeling of these models and then compare

the results to improve the level of conﬁdence.

5 ESSENTIAL SOFTWARE

COMPONENTS

Structurally software environment NIVA is a bunch of

key components related to the following tasks:

• deﬁnition of conﬁgurations and usage scenarios of

Grid/Cloud-systems;

• conversion of the conﬁguration and scenario into

low-level C program code to run it under main hy-

brid simulator;

• run the resulting program in one or more condi-

tions;

• visualization and logging of signiﬁcant events;

• post-processing of results and execution of subse-

quent operator commands (for interactive mode)

or the control command ﬁle (for batch mode).

5.1 Graphical Conﬁguration Editor

Eclipse/SimGrid

Grid/Cloud-systems conﬁguration editor is already

present as addon for the SimGrid basic package and

allows you to set and modify the sample conﬁgura-

tion of Grid/Cloud-systems with use of IDE Eclipse.

Eclipse framework is traditionally used as a wrapper

in many modeling tools, and also includes tools for

teamwork development (support for running with a

network repository). Figures 1 and 2 show examples

of the creation of system conﬁgurations complying

“star” and “tree” topology, respectively. The user can

perform most of actions by “drag and drop”.

5.2 Graphical Scenario Editor

Scenario editor (determining activities in Grid/Cloud

environment) has a simple graphical interface and al-

lows you to deﬁne a usage scenario of a distributed

system as a collection of parallel distributed activi-

ties. Activity hereinafter is a set of interacting pro-

cesses in a distributed system that solve some dedi-

EnvironmentforHybridSimulationofInformationSecuritySolutionsforGridandCloud-systems

257

Figure 1: Eclipse/SimGrid conﬁguration editor. Star topol-

ogy.

Figure 2: Eclipse/SimGrid conﬁguration editor. Tree topol-

ogy.

cated task together. As examples of activities you can

consider user processes interacting during the execu-

tion of computational tasks, and also a set of attacker

malicious actions during a planned attack.

Activity consists of actions. Some actions are

preparatory, i.e. they performed before starting the

simulation. This is a very important feature, which

signiﬁcantly simpliﬁes the operator’s work, because it

is enough to modify (or, say, deactivate) some activity,

slightly modifying the conﬁguration of a distributed

system before starting instead of frequent conﬁgura-

tion changes.

On the Figure 3 you can see an example of “Virus”

activity, which consists of “Virus” and “Bind” action

templates. Each template can be adjusted in accor-

dance to the requirements for activity and scenario.

Graphical scenario editor is implemented as a Java

standalone application using a set of standard tech-

nologies — Swing, JavaBeans, and well-proven li-

braries. To simplify operator’s actions many editing

operations are performed by just dragging the objects

using the mouse.

5.3 Visualization Server Scene

We have developed a server “Scene”, that is a uni-

versal tool for displaying different states of the sim-

Figure 3: Activity parameters in scenario editor.

ulated system. SVG vector format, which is used in

the Scene, is scalable and allows you to display com-

plex graphic elements with the possibility of selective

zoom (Fig. 4).

Figure 4: “Scene” visualization server.

“Scene” server supports the concept of “visual

variables”. For example, they can represent pa-

rameters which vary during the simulation and dis-

plays by the Scene directly with their changes

(load/availability of nodes, node damages by the

virus, etc.). Another feature is the ability to run Scene

on a separate monitor (projector) as a server process

with the ability to manage with multiple interacting

parallel simulations.

“Scene” visualization server is implemented in

Java and allows process group to display complex

state of the modeled system. At the same time one of

these processes declares itself as the master and sets

the initial image (the map) passing URI of the corre-

sponding SVG ﬁle. This SVG ﬁle may also include

the logic for rendering changes, written in JavaScript.

Each process involved in the simulation, whether the

process simulation or a real process in Grid/Cloud

environment can send requests to draw individual

graphic signs (color change, text labels and so on),

SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

258

as well as to receive information about changes in the

“Scene”, that may also be demanded in the more ad-

vanced interactive visualization.

5.4 Virtual Simulation System

Another distinctive part of the developed environment

is a virtual simulator (Zenzinov, 2013). Virtual simu-

lation is based on working with virtual machines and

virtual networks. The management of these objects is

based on the well proven cross-platform library lib-

virt. Deployment system is implemented as a set of

programs written in Python. It performs the follow-

ing functions:

• deployment of virtual machines on a given conﬁg-

uration;

• runs virtual machines with network service for

performing tasks;

• when script execution is completed deployment

system stops the virtual machine and does some

cleanup (deletes temporary ﬁles).

Input data to generate a set of virtual machines

consists of a common conﬁguration for virtual ma-

chines, virtual network conﬁguration and set of tem-

plates of disk images, which contains pre-installed

operating system and the necessary software. Con-

ﬁguration ﬁles in this case are written in the JSON

format and created automatically by the main conﬁgu-

ration editor. Special network service installed on the

virtual machine allows you to perform various tasks

in the case of virtual or hybrid simulation. This ap-

proach provides the opportunity to use the real soft-

ware in the process of modeling to carry out more de-

tailed scenarios of a distributed system. For the use

of any additional software you need to install it to the

original templates of disk images.

5.5 Expanding the Environment

Functionality

Since technologies used in the creation of distributed

computing systems are continually evolving along

with the tools and methods of their protection, the

modeling environment for Grid/Cloud-systems shall

be extensible. API of NIVA environment allows you

to expand its functionality with relatively low re-

source and time consumption:

• the creation of new action patterns;

• extending the visualization server “Scene”;

• in the case of hybrid simulation — implement-

ing coupling simulation model with real programs

launched during the simulation on the virtual or

on real computers.

6 EXPERIMENTS

To illustrate the environment functions let us consider

next three kinds of experiments.

6.1 Hybrid Simulation of Normal

Grid-system Operation

In this experiment Grid-system consists of user node,

master-node and slave-nodes. A typical scenario is

the following: the user sends a request to master-

node, which then distributes the work between slave-

nodes. In this experiment slave-nodes were virtually

modelled and the others were modelled in ordinary

simulation mode.

As a typical computational task we’ve chosen the

numerical calculation of

dx using Simpson’s

rule. On master-node this computational task repre-

sents as a small Python-script. Master-node sends it

to virtual nodes via network service, and then it runs.

Using virt-manager open source tool for virtual ma-

chine management you can follow task processing: its

reception, launch and result.

6.2 Virus Attack Against the the IaaS

Cloud-system

Second experiment is a simulation of virus at-

tack against IaaS (Infrastructure-as-a-Service) Cloud-

system from external infected network. Internal net-

work of information system contains special intru-

sion detection system (IDS). Information system also

contains node with IaaS infrastructure and simulated

models of virtual machines, which are assumed to

running on this node.

The experiment was conducted in two stages: with

IDS turned on, with IDS turned off. In the ﬁrst case

IDS successfully blocked the virus propagation on the

information system. In the other case we observed

Cloud-system infection in accordance with the speci-

ﬁed parameters of infection probability. Some of in-

fected nodes have dead. Probability of failure was

also speciﬁed by the parameter.

It should be noted that if IaaS-node fails, then its

virtual machines also become unavailable and also

terminate their activity. This situation corresponds to

the real behavior of Cloud-systems.

EnvironmentforHybridSimulationofInformationSecuritySolutionsforGridandCloud-systems

259

Table 1: Summary results of virus attack

Conﬁguration

IaaS PaaS SaaS Ring Star Line

Number of completed jobs

23 33 51 15 47 5

Number of failed jobs

10 5 0 13 1 18

Rate of successful jobs, %

69.70 86.84 100 53.57 97.92 21.74

Number of dead nodes

2 1 2 1 4 1

6.3 Non-interactive Launch a Series of

Experiments

In this series of experiments we have investi-

gated virus propagation and DDoS-attack against

Grid/Cloud-systems with six different conﬁgura-

tions. Three Cloud-systems: IaaS (Infrastructure-

as-a-Service), PaaS (Platform-as-a-Service), SaaS

(Software-as-a-Service), which differ in their archi-

tecture and mechanisms of job distribution. There

are also Grid-systems with three different network

topologies: ring, star and line. As in the previous

paragraph there were two cases: IDS-on and IDS-off.

This gives a total 24 different conﬁgurations.

When the execution of experiments is completed,

the user receives a report which represents informa-

tion about executed actions and a summary of results:

the number of successful job queries, the number of

unsuccessful queries, the number of dead nodes, and

the overall success rate.

On the Table 1 you can see an example of this re-

sults in case of virus attack without IDS-protection.

These numbers are completely dependent on input pa-

rameters and probabilities.

7 CONCLUSION

As a result of this work we have created a prototype

of software environment for modeling of Grid/Cloud-

systems which supports several simulation modes.

In addition to widely used simulation approaches

Virtual-based modelling allows us to assess the ef-

fectiveness of security systems, taking into account

the features of real programs that can contain serious

bugs. Combined with other simulation modes it gives

us new unique possibilities with hybrid mode.

Our future work will be focused on improvement

of our virtual simulator and its interaction with Sim-

Grid. In general, hybrid modeling of distributed sys-

tems gives us a wide area for further research. There

are a lot of tasks and questions related to the concept

of hybrid simulation like improvement of the simula-

tors interaction, search and investigation the speciﬁc

effects resulting from this interaction etc. In addition,

there are many problems in which modeling environ-

ment serves as a tool: estimation of security software

for information systems, testing of special software

for distributed computer systems, search for the opti-

mal conﬁgurations to protect systems against speciﬁc

attacks.

REFERENCES

Buyya, R. and Murshed, M. (2002). Gridsim: A toolkit for

the modeling and simulation of distributed resource

management and scheduling for grid computing. In

Concurrency and Computation: Practice and Expe-

rience (CCPE), volume 14, pages 1175–1220. Wiley

Press, USA.

Casanova, H., Legrand, A., and Quinson, M. (2008). Sim-

grid: a generic framework for large-scale distributed

experiments. In Proceedings of the Tenth Interna-

tional Conference on Computer Modeling and Simu-

lation, UKSIM ’08, pages 126–131, Washington, DC,

USA. IEEE Computer Society.

Grushin D., Kuzjurin N., P. A. S. A. (2008). Modeling grid

behavior using workload data. In The 3-rd Interna-

tional Conference “Distributed Computing and Grid-

technologies in Science and Education”.

Josef Spillner, Andrey Brito, F. B. A. S. (2012). Analy-

sis of overhead and proﬁtability in nested cloud en-

vironments. In Cloud Computing and Communi-

cations (LATINCLOUD), pages 13–18, Washington,

DC, USA. IEEE Computer Society.

Rodrigo N. Calheiros, Rajiv Ranjan, A. B. C. A. F. D. R.

and Buyya, R. (2011). Cloudsim: A toolkit for model-

ing and simulation of cloud computing environments

and evaluation of resource provisioning algorithms. In

Software: Practice and Experience, volume 41, pages

0038–0644. Wiley Press, USA.

Yaser Mansouri, A. N. T. and Buyya, R. (2013). Broker-

ing algorithms for optimizing the availability and cost

of cloud storage services. In Proceedings of the 5th

IEEE International Conference on Cloud Computing

Technology and Science.

Zenzinov, A. (2013). Automated deployment of

virtualization-based research models of distributed

computer systems. In Proceedings of the 7th

Spring/Summer Young Researchers Colloquium on

Software Engineering (SYRCoSE 2013), pages 128–

132. National Research Technical University Kazan,

Russia: Kazan.

SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

260