![](bg2.png)
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
SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
256