Enhancing the Performance of Network Using Agent Mobility in a Distributed Systems

mustardpruneΔίκτυα και Επικοινωνίες

23 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

53 εμφανίσεις

Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009






Abstract

As networks become pervasive, the importance of
efficient information gathering for purposes such as
mon
itoring, fault diagnosis, and performance evaluation
increases. Distributed monitoring systems based on eit
her
management protocols like S
N
M
P or distributed object
technologies such as CORBA can cope with scalability
problems only to a limited extent. They

are not well suited to
the
systems that are both very large and highly dynamic
because monitoring logic, although possibly distributed, is
statically predefined at design time. This paper presents an
active distributed monitoring system for security based

on
mobile agents. Agent
s act

as area monitors not bound to any
particular network node that could “sense” the network,
estimate better locations, and migrate in order to pursue
location optimality. Simulations demonstrate the capability of
this approach t
o cope w
ith large
scale
distributed
systems and
changing network conditions. The network monitor has to
capture the packets that flow between any two machines in a
network
and extracts

information related to the
source/destination IP addresses, packet leng
th, time of
capture,

network load, number of packet loss
, type of network,
and the network speed. This module p
erforms the activities of
Fault
management, Accounting management, Conf
iguration
and Name management, P
erformance management and
Security managem
ent.


Index Terms


Networking, Agents, Network Security,
Distributed Systems


I.
I
NTRODUCTION


As networks become pervasive, the importance of efficient
information gathering for purposes such as monitoring, fault
diagnosis and other performance evaluation

increases. The
network and its associated resources and distributed
applications become indispensable to the organization. Things
can go wrong disabling the network or portion of the network
or degrade the performance to an unacceptable level. Hence we



ne
ed to felicitate the client with our system to monitor the
network, which requires the following capabilities
:

1.
Fault Management
: The monitor should have the facility to
detect, isolate and correlate the abnormal operations of the OSI
environment.

2.
Acc
ounting Management
:

This facilitates

charges to be
established for the use of managed objects and cost to be
identified for these objects.

3.
C
onfiguration and Name Management
: This facilitates
agents

to

control over, identify, collect data from and provid
e
data to managed objects for the

purpose of assisting in the
continuous operation of interconnection service.

4.
Performance Management
: This facility evaluate
s

the
behavior of the managed objects and the effectiveness of the
communication activities.

5.
Securi
ty Management
: This facility addresses

those objects
of OSI security essential to operate OSI network management
correctly and to protect managed objects.


The existing monitoring systems were analyzed for satisfying
the future requirements imposed b
y future networked systems
based on their
topologies;

scale and increasingly complex
services and applications that rely on heterogeneous integrated
fixed and mobile networks. With the pace at which the
technology is advancing in this information era we ha
ve to keep

the networking systems ready for Neuro and Fuzzy systems
while utilizing the artificially intelligent mobile agents which not
only save manpower but also are effective in improving the
quality. The system has to be effective in its storage and h
ave a
convenient decision support system for the network monitor.

An important difference between current and future
networking is that in the former case
topology
is considered
relatively static, while this becomes an unsafe assumption for
future networks
. New models of dynamic networks are
contrasting with the situation in which the network topology
was mainly modified as a result of careful learning. Examples in
which logical topology layering can be dynamically constructed

in the real time are the follo
wing:

1.

Dynamically re
-
configurable networks

2.

Active networks

3.

Dynamic virtual private networks

4.

Mobile and survivable networks

5.

Re
-
configurable cellular networks

Enhancing the Performance of Network

Using
Agent Mobility

in a Distributed Systems

Prof.
T.
Ravi
1
*
,
S.Sasikumar
2
, M.Shamika
3

1*Professor,
travi675@yahoo.com

, 2

Assistant Professor,
ssasiin@yahoo.com
,

3
shamika.i.nair@yahoo.com


KCG College of Technology
, Chennai

Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009



Another important feature of future network is
scale
. The
tremendous success of the Internet has
made possible and
even encouraged the realization of systems characterized by
very large scale and high levels of distribution and dynamics.
Network
-
centric approaches such as Sun’s JINI architecture
envisage large numbers of comparatively simple devices a
ll
accessible through the network. Management systems in the
future will need to keep track of these devices and determine
which are present, which are functioning currently, and so on.

The third key factor is the appearance of increasingly
complex service
s and applications that rely on heterogeneous
integrated fixed and mobile networks. The variety of such
applications and services along with the possibility of
accessing them from virtually any location makes it extremely
difficult to anticipate the type a
nd distribution of network
traffic. Because of their scale and dynamics future networked
systems will be very difficult to manage and control unless
efficient means of monitoring them become available.

Conventional approach to network monitoring based on
e
ither management protocols or distributed object technologies
cannot fully satisfy the requirements of future

networked
systems explained above. Intuitively, in order to monitor large
-
scale dynamic systems we need a distributed monitoring model
that adapts

to the monitored system. The existing monitoring
models like Simple Network Management Protocol (SNMP),
Tele communications Management Networks (TMN), and
distributed object approaches such as Common Object Request
Broker Architecture (CORBA) and Java Rem
ote Method

Invocation (Java


RMI) do have a lot of limitations.


The network

monitoring techniques available to us like static
centralized monitoring, static decentralized (Hierarchical or
based on distributed objects) monitoring and dynamic
(programmable

or active) decentralized monitoring are having
advantages of their own. The advantages of these techniques
are to be utilized for a better network monitoring. These models
are shown below:




F
ig.1



Monitoring Station


Fig.2
Area Monitor


II
.
PROPOSED
SYSTEM

We propose a tool to be developed such that it would be
able to obtain the network information such as the network
load, and existing traffic scene along with the route that the
packets tak
e to travel from one host to
another. The tool
captures the
packets that flow between any two
-
host machines
and displays information that include source IP address, packet
length time and the network load. Although conventional
Sniffers are available to carry out this work, they consume
enormous amount of resources

including memory and CPU
processing. We pr
opose to use this Mobile Agent
(MA). An
MA is essentially an autonomous object containing the logic to
perform a given task and possibly migrate under its own control
from node to node in a network.

The idea of

em
ploying artificial

intelligent MA for
applications such as distributed information retrieval,
performance monitoring and remote data filtering or aggregation

has been extensively reported. It is a mechanism to realize
dynamic programmability of remote elem
ents according to the
Management by Delegation (MbD) which may be realized with
agents bound to single loop mobility


the agents move from
the managing node to remote managed nodes, and then stay
put. We enhance it to the multiple hop capability by our mo
bile
agents.

A distributed algorithm is required to compute the agent
locations, initially offline and then at run time. During the
execution of monitoring task, agents will need to sense their
environment and take actions in order to adapt to changing
con
ditions. By doing so,
we
maintain location optimality which
is concerned with minimization of network traffic incur
red

by
the agent based monitoring system and of latency in collecting
the necessary information.

Our algorithm relies on agents learning the
network topology
through node routing table information accessed through
standard management interfaces. The monitoring system is
initially deployed through a
clone

and
send

process starting at
centralized net
work wide station. The agents who

adapt to the
Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009



network changes through migration also use the same
algorithm. The main features of the algorithm are its distributed
nature and low computational complexity.

We proposed a
dynamic decentralized monitoring model with mobile agents.
The picture for which is

shown below




Fig. 3
Mobile Agents


A tool that continuously monitors server availability and
performance is to be developed. In the event of network errors,
the tool should alert the network administrator by email before
the problem is aggra
vated. This will help to reduce cost
involved in the likelihood of failure on network. The tool should
also monitor the route of the packet through the network from
one machine to the other.

The features required for this tool are

1.

It has to

run to capture
the packets that are transmitted
between the source and destination terminals in the
network.

2.

The application then displays the information related
to the packets transferred with in a certain time slot, in
the form of graphs and statistics.

3.

The display r
efreshes after every time slot and the data
related to the transmission are stored in a log file
which is later hosted on the company’s administration
web site.

4.

In case of any traffic congestion in the network that
may be detected during the analysis the n
etwork
administrator is contacted by email.

5.

Also the route the packets take are monitored and
displayed upon request.

The software must be designed to support subsequent
upgradation without altering the basic structure of the system.


III.
INTERFACE SPECIF
ICATION


Both the network administrator and associates can monitor
the access of existing network situation using the tool. The
access to the log files is restricted to the network administrator
alone.

The features required to be supported by the
tool




1.

I
t has to

run to capture the packets that are
transmitted between the source and the
destination terminals in the network.

2.

The application then displays the information
related to the packets transferred within a certain
time slot, in the form of graphs and

statistics.

3.

The display refreshes after every time slot and
data related to the transmission are stored in a log
file which is later hosted on the company’s
administration website.

4.

In case of any traffic congestion in the network
that may be detected duri
ng the analysis the
network administrator is contacted by e
-
mail.

5.

Also the route the packets take are monitored and
displayed upon request.


IV. SYSTEM DESIGN OF MOBILE AGENTS

The agent location problem consists of two phases. Initially,
we need to determi
ne the approximate number of agents for a
given monitoring problem and compute the location of each of
those agents. Subsequently upon agents development the
agent system needs to be able to self
-
regulate in order to adapt
to changing condition. This is ac
hieved by triggering agent
migration in a controlled fashion to avoid instability due to
continuous agent migration.

The problem of computing the optimal number and location
of area monitors is analogous to the optimal placement of p
serves in large networ
k, which has been studied since the early
70’s. This belongs to a class of p
-
center and p
-
medium
problems, both NP
-
complete when striving for optimality.
Approximate polynomial algorithms have been proposed, but
none of them suits the requirement of the ag
ent system.
Proposed algorithms are centralized, requiring the network
distance matrix at the main monitoring station. While this is less
of a problem in offline calculations for medium to long
-
term
optimal locations, it becomes an important problem for ac
tive
distributed solutions in which optimal location need to be
calculated by agents themselves. In this case, the monitoring
station should retain an up
-
to
-
date version of the whole
network topology, which obviously is an unrealistic
requirement for large
-
scale dynamic networked systems.


In the system proposed here in the location of area monitors
is neither fixed nor predetermined at designed time. The area
monitors are realized with MAs, simple autonomous software
entities that, having access to net
work routing information, can
adopt and roam through the network. Progressively partitioning
the network and populating each partition with monitoring
agents deploy the distributed monitoring system.


We assume the existence of an agent system support
ing
mobility and cloning. Agents are assumed to have access to
routine information obtainable from network routers through
standard network management interfaces. For simplicity, we
also assume that MA hosts are evenly distributed within the
network. This,

in other terms, means that for each router there is
always an MA host that is located relatively close to it and, for
Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009



each LAN the number of MA hosts is proportional to the
number of monitored objects that need to be in that LAN.
Under these assumptions t
he MA distribution does not differ
significantly from the routing tree router at the monitoring
station. Without loss of generality, we envisage a scenario in
which routers can act as MA hosts during MA development. In
such case, the MA distribution tree w
ould actually coincide
with the routing tree.


V.
AG
ENT DEVELOPMENT


The conventional approach to portioning the monitored
system on the basis of its topological and dynamic features is
clearly not viable in the case of very large
-
scale dynamic
systems.

The task of collecting a real time snapshot of the
topology and traffic profile of large scale rapidly changing
networking system is indeed ambitious. The basic underlying
idea used to solve the agent location problem into exploit
information that is read
ily available in the system rather than
trying to derive a new network distance matrix. This is why our
monitoring system relies solely on routing information, which is
maintained by routing protocols. The precise nature and quality
of this information wil
l depend on the routing protocol in use.
Sophisticated routing protocols such as Open Shortest Path
First (OSPF) can maintain information using multiple distance
merits. Simple routing protocols such as RIP however use only
a hop count metric. Our objectiv
e is to design a development
algorithm that will optimize agent location with respect to

whatever metric information is available. However, it is clear the
performance of our system in absolute terms will be affected by
the quality of this information.



A flow chart of the proposed development algorithm is
shown. The algorithm is illustrated through a simple example for
the network shown. For the sake of simplicity we assume that,
list of MOs consist of all the network nodes. The basic phase of

the agent

location processes is

depicted. The algorithm deploys

the area monitors during the network partitioning process
through a “clone & send” process starting at the main
monitoring station. The number and location of MAs is
computed by subsequently comparing
the monitoring task
parameters with routing information extracted from the network
routers. The monitoring task, including the list of MOs, as well
as the operations to be performed on them, is delegated to a
first MA.

VI. E
XPLOITING

A
GENTS
FOR
NETWORK

M
O
NITORING

Modular Description:

As networks become pervasive
efficient information
gathering
for purposes such as
monitoring
, fault
-
diagnosis and
performance evaluation increases. This module provides

an
active distributed monitoring system based on mobile a
gents.
Irrespective of the

topology overcoming the disadvantages of
dynamically re
-
configurable networks, active

networks,
dynamic virtual private networks, cellular networks and

network concentric

approaches such as SUN’s JINI
architecture. The mobile ag
ents are employed

for applications
such as distributed information retrieval, performance
monitoring and

remote data filtering or aggregation.



.




Fig. 4
Agent Mobility for Network monitoring






















Network

Monitoring
Path









Agent

Agent
Deploymen
t



Agent
Deployment

Algorithm

Agent

Agent Self
-
Relocation



Agent Self
Relocation

Algorithm


Packet Structure for

Agent

Packet Structure for

Ag
ent

Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009





Fig. 5

Dataflow diagram for effective location of area monitors



Fig. 6 Dataflow diagram f
or network monitoring technique


VII. RESULT


Steady state performance is measured before the failure;
then sufficient time is allowed for the system to adapt in terms
of routing table manipulation (by the relevant protocol layers)
and subsequent agen
t migration. Finally, performance is
measured again as soon as the
new steady state is reached. The
following
figures

depict

those two performance states in terms
of

traffic and response time for different system configurations.



Fig. 7
.
Performance sta
tes in terms of Network Traffic



Fig. 8
.
Performance states in terms of Response time


VIII
. FUTURE

ENHANCEMENTS

The most restrictive filter can be set on the packets needed
by the application. A restrictive filter decrease the number of
packets buffered

by the driver and copied to the capture
application. This makes space in the buffer for the needed
packets only and decreases the load on the system.

If the data of the packets is not needed (like in the most part
of the capture applications), a filter i
s set that keeps the headers
only. For this reason Argus sets a filter that tells the driver to
save only the first 68 bytes of each packet (enough for the most
part of the protocols).

In real time analysis or applications requiring statistics about
the n
etwork, the statistics mode can be used. It uses little
processor time, and it does not need any kernel buffer.
Therefore, the kernel buffer can be set to 0.



Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009



Also the entire system can be made more intelligent by using
techniques in Neural Networks to ada
pt to the situations based
on similar cases that have occurred earlier.

The system can be enhanced such that it can provide
suggestive measures to correct the existing network situation to

the network administrator


IX.

CONCLUSION

There is no saturation p
oint for any system. With the newly
developed software the users work is erased and new ideas will
crop up in his mind with more and more needs to cut short his
job. The existing manual system is for the first time put through
the machine. The users are

very much satisfied with the
performance and it a opened a new dimension of thinking
among them.


The debugged software of the developed system is tested
with the real data and the reports
generated are with 100%
accuracy

and error free and has
the
follow
ing characteristics

1.

Portable and Flexible for further
developments.

2.

Avoids errors by avoiding the manual work.

3.

User friendly screen to ente
r the data and
enquire database.

4.

Cost minimization in the case of stationery
and personnel

traffic and response time
for
different system configurations
.

REFERENCE:


1.

D.N.Legge; P.R.Baxendale
,

An Agent Based
Network management systems”
, 1999.

2.

Wooldridge M. "Agent
-
Based Software Engineering"

IEE Proceedings in Software Engineering 1997

3.

Liotta, A.


Pavlou, G.


Knight,
G.

Exploiting agent
mobility for large
-
scale network monitoring, IEEE
Explore 2002.

4.

Alan Bivens, Rashim Gupta, Ingo Mclean, Boleslaw
Szymanski, Jerome White, Contact Boleslaw
Szymanski “Scalability and Performance of an Agent
-
based Network

M
anagement Mid
dleware “,

International Journal of Network Management,2006

5.

Sameera

Abar
1
, Yukio

Iwaya
2

, Toru

Abe
3

and
Tetsuo

Kinoshita
3

,”
Exploiting Domain Ontologies
and Intelligent Agents: An Automated Network
Management Support Paradigm”, Lecture notes on
Computer Sc
ience, Springer
-
Verlog, 2006.

6.

http://Wikipedia.org











































Proceedings of the
Intern
ational Conference ,

Computational Systems and
Communication Technology”

Jan.,9
,200
9

-


by

Lord Venkateshwaraa Engineering College
,
Kanchipuram Dt.PIN
-
631 60
5
,INDIA




Copy Right @CSE/IT/ECE/MCA
-
LVEC
-
2009











































Fig. 9 Agent configuration flowchart


Agent deployment































START

Heuristic function (P
j
, T
h
); monitoring station node, u; list of
monitored objects MO

INITIALIZATION

Set current root node v


u ; clone first agent at u

Compute TOTAL cost to monitor MOs in the sub
-
tree rooted at
current node (based on local routing table costs)


Compute root’s neighbor nodes, neigh(v) and label them ‘nonvisited’ (neigh(v)

is the list of the
next_hop addresses in this local routing table)

eave all
neigh(v) been
visited?

mending
partition
empty?

pelect next unvisited neighbor


node i

START MA


Compute number of MOs reached through their associated PARTIAL monitoring

and i
cost (individual costs extracted from local routing table)

The heuristic function computes P
i

(i.e., The probability of sending an MA to node i)

Sub
-
tree rooted at i will not be
appended to the next new “pending”
partition

m
i
>T
h
?

START PARALLEL TH
READ

Clone and initialize new MA; i is the new
root node (v

i); the MOs list now
contains only the subset of MOs
belonging to the sub
-
tree rooted at i




Sub
-
tree
rooted at I
will be
appended
to the next
new
“pending”
partition

KILL MA

END

Y

N

Y

N

Y

N

Clone