QoS
Management
By
Mobile Agents in Multimedia Communication
S.S.Manvi and
P.
Venkataram
Electrical Communication Engineering Department,
Indian Institute
of
Science, Bangalore-56001
2
INDIA,
E-mail:
{
sunil,pallapa] @ece.iisc.emet.in
Abstract
In
this paper we propose a mobile agent based
QoS
management
system
to
saris&
the jvefitncrional principles
of
QoS
architecture,
i.e.,
integration, separation, transparency, asynchronous resource
nianagenient andpetjorniance. Mobile agent paradigm is a unique
paradigm in contrast
lo
traditional client/server paradigm in a
problem inherently distributed and coniplex. such
as
QoS
manage-
ment. This paradigm saves considerable amount
of
bandwidth and
reduces network traffic.
We
present
un
analytical und siniulation
models
of
a
mobile agent
based
bandnidth negotiation and
rnan-
agenienr systeni
for
nriiltiniedia coninucnication. Response time
of
the niobile agent is conrpirted with different nuniber
of
retransnzis-
sions
and
the intermediate “hops” for both the models.
It
is ob-
served that the response tinre of the ugent increases with increase
in n~iniber of retransniissions
and
the hops. We have also showed
that the increase in
the
urrival of
niobile
agents
would
affect the
odniission
oj
new rnultiniedrri applicutions.
keywords: mobile agent,
QoS,
bandwidth, nutlriniedia com-
nucn ica tion.
1
Introduction
Distributed Multimedia applications demand real time multimedia
communication because they are
of
isochronous nature. These ap-
plications depend on
a
certain level of quality of service
(QoS).
They need a mechanism for
QoS
adaptation
in
order to deal with
temporary changes in the available
QoS
parameters.
Such
applica-
tions require
a
QoS
negotiation and management system to provide
guaranteed real-time services in multimedia communication. The
main functions of
QoS
management are:
QoS
negotiation,
QoS
renegotiation,
QoS
mapping at different levels, resource reserva-
tion,
QoS
monitoring and
QoS
adaptation.
If
negotiation ends with
an agreement on the required values, application can be launched
and managed
in
later phases. The types of agreements made could
be
best effort, stochastic
or
guaranteed.
Most of the networks
presently work on “best effort” service without
QoS
guarantees
.
Hence, there is a need to design
a
QoS
architecture which provides
guaranteed services
in
multimedia communication.
1.1
QoS
requirements in Multimedia Communica-
tion
QoS
states how valuable the services provided by the multime-
dia systems are.
QoS
parameters can be considered at the three
levels:
application, systeni(operating system) and network level
[
13.
The
application or user parameters
(up)
consists
of
media
quality descriptions for the specific media characteristics
of
each
device, such as sample size, sample rate and priority.
Sysreni pa-
ranieters
(sp)
include
CPU
power, buffers and secondary storage
capacity. The
network paranrefers (np)
are packet size, packet er-
ror
rate, end-to-end delay, packet .rate,
loss
rate(reliability), jitter
and bandwidth. The
QoS
of
a
given system is expressed
as
a set
of
(parameter-value) pairs, something called tuple: each parame-
ter is considered as a typed variable whose values can range over
a given set
for
eg., delay:<5sec, 6seo. Different applications on
same distributed systems caan have different values required.
A
QoS
mapper is required to map these parameters, i.e.,
sp=f(rip)
and
np=f(sp).
The generally considered parameters for multimedia communi-
cation are:end-to-end dclay:the elapsed time between the gcnera-
tion of
a
service and presentation
of
service; delay jitter: the varia-
tion
of
end-to-end delay; packet error rate: the percentage
of
pack-
ets discarded due to transmission failure in the path; bandwidth:
the transfer bit rate
of
a
service in the path.
1.2 Mobile agents
Agents are the autonomous programs situated within an environ-
ment, which senses it and acts upon it using its knowledge base,
and learns
so
as to act in future. Agents are classified as: local or
user interface agents, networked agents, distributed AI (Artificial
Intelligence) agents and mobile agents based on the attributes they
posses
[2].
The concept
of
mobile agents grew out
of
three earlier
technologies:
process migration, reniore evaluation and ntobile
ob-
jects
.
All these concepts are an improvement over
RPC
(remote
procedure call)
for
distributed programming. Mobile agents are
the multiagent systems which posseses all the attributes specified
for
an agent.
It
is an itinerant agent dispatched from source com-
puter which contains program, data, execution state information,
migrates from one host to another host in the heterogenous network
407
0-7695-0680-1/00
$10.00
0
2000
IEEE
and executes at remote host until they accomplish their task. The
mobile code should bc platform independent,
SO
that,
it
can execute
at any remote host in the heterogenous network environment. They
communicate and cooperate with other agents to achieve its goals.
Interagent communication can be achieved by message passing,
RPC
or
common knowledge base( black board)[3].
Mobile agent paradigm significantly reduces bandwidth con-
sumption and network traffic in contrast to clientlserver paradigm.
Mobile agents can be used in network management, telecommu-
nication services management,
QoS
management and mobile
corn-
puting. There are certain issues to be resolved in implementation of
mobile agents such as agent transfer mechanisms, addressing, ex-
porting agent state information, c0m”iCatiOn language. secrecy,
privacy, agent data transfer. authority and portability and security
[31.
1.3
do not interfere with other functions and flow transmissions
of
the
system. Transparency
is
achieved, because, users and applications
can transparently delegate tasks of negotiation and management to
the agents. Agents execute concurrently being asynchronous en-
tities, hence asynchronous resource nianagenienr principle is sat-
isfied. Since agents are solely used for management functions,
it
does not affect other network protocols, thus
it
achieves perfar-
niance principle.
The proposed architecture consists of a
QoS
agency installed
in all the networked multimedia systems. It comprises of agents
(fixed and mobile), age,1tfacilitator and a
nlultinledia
application
sewer (MAS) as depicted
in
fig.1. Flow application parameters
defines
QoS
requirements for each flow.
A
flow is referred as a
path from service generation point to service presentation point,
These are specified as a set of parameter-value pairs. MAS man-
ages these Row application parameters. Agent facilitator instanti-
ates agents when
it
receives a message containing the type of agent
to instantiate and the parameters to be passed
to
that agent. This
new agent starts executing, during execution it may communicate
with other agents
or
resources to complete its task.
Interaction
among the agents takes place through shared knowledge base. The
negotiation takes place in all the nodes of flow. Flow application
parameters can be renegotiated, if resources are not available
as
pcr
the requirements of the user.
Proposed Mobile Agent Based
QoS
Manage-
ment Technique
Mobile agent based
QoS
management system is proposed to ne-
gotiate and manage the the Rows of multimedia communication.
It consists of mobile agents and static agents which interact with
each other through shared knowledge base. Mobile agents are
the piece
of
code written in shell scripts. Every node in the net-
worked multimedia systems must have an agent execution envi-
n&#l
noJc
#I
ndr
#N
ronment consisting of an agent server and shell interpreter and the
sending agent to shell interpreter
for
execution, sending the exe-
cuted mobile code with data to next node in the network. In the
by sending a bandwidth negotiator mobile agent across the path be-
proposed scheme, the service user (multimedia client) host negoti-
ates for bandwidth, with the service provider (multimedia server)
fore launching multimedia application. The intermediate nodes in
the path between
two
end-systems (service
user
and provider) are
indirectly involved in the process of negotiation and management.
The mobile agent interacts with the local agents in the visiting node
through shared knowledge base and collects the information about
the availability of bandwidth
for
that flow during its upstream travel
from user to service provider (multimedia server).
agents. Agent sever is responsible for receiving the mobile agent,
NS
mulhmr
1
JLlnlT
‘
#
0’
\
Fig I. Mobile agent based
QoS
negotiation and management
2
QoS
Management
by
Mobile Agents
Mobile agent based
QoS
architecture offers an interface for defin-
ing desired level of
QoS,
negotiating resources
for
attending de-
sired level of
QoS,
monitoring
QoS
for adaptation and renegotiat-
ing resources when
QoS
degrades. This architecture provides
QoS
end-to-end control for each Row. All the intermediate nodes be-
tween the source and sink
are
indirectly involved in the process
of
negotiation and management.
These systems satisfy all the five functional principles of
QoS
architecture[4,5]. Integration is achieved by defining agents that
handle
QoS
at different abstraction levels (user, system, network).
The separation principle is due to autonomy of agents.
So,
they
The fixed
or
local agents used in
QoS
agency
are:
user interface agent defines an application
or
user level param-
eters; QoSmapper agent
:
maps parameters from one level to other
level;
QoS
nionitor agent: monitors the multimedia flows; resource
manager agent: manages local resources at hosting node;
local
ne-
gotiator agent: negotiates locally by interacting with the resource
manager agent;
QoS
adaptor
agent: corrects locally if there is a
minor degradation
QoS
in
the flow.
The mobile agents used in the
QoS
agency are:
pow
negotiator agent: performs global negotiation by interact-
ing with local negotiator agent at the migrated node;Jow renego-
tiator agent: acts on behalf
of
adaptor agent when some parameters
are changed for the flow;
flow
nionitor
agent:
travels periodically
to nodes in the flow, interacts with monitor agent and reports degra-
dation of
QoS
to the host node.
408
Now agent with negotiation results migrates downstream confirm-
ing or fixing negotiated values from “N” th node to hosting (source)
node
1.
pu, probability of agent migrating to upstream node from state
i
state
j
is given as:
P,=l-
Pled,
Pled=
probability of agent corrupted
or
delayed
or
lost.
qd =probability of agent migrating downstream from state i to
state
j
is given as:
qd=I-
Pled,
Transition probabilities are pu,
qd,
l-pu, I-qd,
These transition probabilities are represented as
iriJ
in the state
diagram where:
i,
j
=1,2
,......
N.
p1 ,p2,
.......
p~ are state probabilities of state space(
1
,....,
N}
and
probability of being in state
i
is given by: pi[N]
=
p( XN=i). Be-
cause of stationary probability property of markov chain, transition
probabilities are defined by one step transition probabilities:
1
-pu-qd.
nij[n,n+l]
=
p{X,+,=j
I
X,=i)
cj
irij
=
1
j=l,..N.
abilities can be obtained from steady state
e
uation
If initial transition probabilities are considered steady state prob-
41
PQ=P where: p=p,[N] and
pi=l.
QoS
negotiation Scheme
The negotiation scenario in this
QoS
management framework can
be explained in sequence of steps as follows:
1.
The user interface agent reads the user/application level flow
parameters and passes flow application parameters to
QoS
map-
per agent;
2.
The mapper agent maps userhpplication parameters
to
system level parameters and passes to local negotiator agent;
3.
The local negotiator agent tries to reserve resources locally as per
the flow requirements by interacting with resource manager agent;
4.
If local negotiator agent is successful in local negotiation, Row
negotiator mobile agent is instantiated to negotiate resources out-
side node;
5.
When arriving at remote nodes in the Row route, this
mobile agent interacts with local negotiator and reserve resources
temporarily;
6.
The flow negotiator agent returns to its hosting
node in the reverse route by fixing up the reserved resources per-
manently at each node for the Row until the flow is processed.
A mobile agent can ping the neigbouring node before migration
so
as to ensure that node is working. If a node is not working, it
will communicate to the user
or
process which has sent it,
so
that
another route can be chosen. Thus the agent can learn about bad
nodes
in
the network and communicate with other agents.
Management of negotiated flow is achieved by flow monitor and
local
QoS
monitor agents. The mobile flow monitor agent acquires
the values of the parameters of the flow exhibited at the visiting
node. If a degradation
in
QoS
is detected,
QoS
adaptor agent at the
agent hosting node is informed. The adaptor agent tries to make
minor adjustments locally to attain original level of
QoS.
If it is
not successful, flow renegotiation mobile agent is instantiated to
renegotiate the resources.
3
Analytical Model
A discrete state, discrete time markovian model is considered to
show the negotioation for one QoS parameter (bandwidth) using
mobile agents as depicted in fig.2. This model uses
“N’
number of
nodes in the Row from source
to
destination. Node
1
is considered
as agent hosting (source) node.
Fig.
2.
Analytical model for negotiation agent.
Agent migrates to upstream node starting from hosting node
1,
executes bandwidth negotiation at each nodes until
“N
th
node.
3.1
After having computed the steady state probabilities, response time
of
agent’?,” is given by
Response Time
of
Mobile Agent
N
t,=
pi
t,,,
+
2*(n-1)*taUt
+
T*R,
where:
N= no. of states, n=no. of hops, T=timeout interval(sec),
C=capacity
of
link(bps), pi=steady state probability
of
being
in
state,
x=
average agent size(kilobits),
taut=
average transmission
time of agent on link= x/c, ts,,=total service time of agent at
all
the nodes, R=number of retransmissions= (I/pl-
I),
if pl=l, then
R=O.
4
Simulation Model
The QoS parameter we have considered for implementation
is
bandwidth. This parameter is very important in multimedia com-
munication to provide real-time services. Layered functions of mo-
bile agent framework used for the implementation of mobile agent
based bandwidth negotiation comprises of agents (top layer), shell
interpreter, agent server and TCPAP (transport mechanism is the
bottom layer). Agent server is responsible for receiving the agent,
executing the agent with the aid of interpreter and sending the mo-
bile agent to outside world. The shell interpreter executes the mo-
bile shell script (shell script
for
negotiating the bandwidth). Trans-
port mechanism used for mobile agents is TCP.
A
mobile agent is
hosted from the host in need of running an application from mul-
timedia server. It consists of shell script containing the lower and
upper range of bandwidth, path of travel, statements
for
negotia-
tion and migration. Pseudocode of the mobile agent negotiation is
as
follows.
409
Algorithm
I:
Mobile agent negotiation procedure
Low-range-bwidth
=
x;
High-range-bwidth
=
y;
path=jala:protocol:pet:proptocol:jala;
check-QoS-profile(
);
If
(bandwidth-at-node
>
x
&&
bandwidth-at-node
<
y)
then collect data and set-bwidth-from-host
=
band-
wid th-at-node
else if(bandwidth-at-node
>
y)
then collect data and set-bwidth-from-host
=
y
else collect data and set-bwidth-from-host
=O
/*
0
indicate unsuccesful*/
migrate(path)
stop.
Mobile agent visits all the nodes in the path, and negotiates
with local agents at that node through common knowledge base
(QoS-profile). After reaching the service provider
or
multimedia
server (end host), the agent travels back in the same path, fixes the
negotiated bandwidth, returns to hosted node. The user
or
a pro-
cess at the hosted node decides to run the application based on the
information gathered by the agent during its travel.
A
linear system configuration is considered for simulation. Sim-
ulation has been carried out for two node and three node linear
system configuration as shown in fig.3. The machine named ‘‘jala-
system’’ is the host of agent and destination is “protocol” for two
node systems. The destination (multimedia server) is “pet-system”
for three nodc system. The specifications
of
these machines are
pentium-IOOmhz, and they are connected by linux network which
has
8
systems. We have partially simulated the mobile agent based
QoS management framework. The model simulated considers only
two agents, local negotiator and the flow negotiator mobile agent
on
it.
route and confirms bandwidth, reports about bandwidth availabil-
ity
at the nodes
in
the flow. In case, the agent is lost, agent host will
time out
for
“T”
seconds and retransmit the agent. Even then,
if
the
agent does not respond, this process is carried out
for
ten attempts.
If
it
fails in all the attempts, agent host has to try for another route.
This failure may be due to broken link
or
node failure in the path.
The pseudocode for this is as follows.
Algorithm
2:
Faultfinding
1.
send mobile agent, initialize attempt=
I.
2.
wait for results until
time-elapsed=time-our(T)
3.
If
result-received within tinieout
then display resiilts(succ
or
iinsricc)
goto
step
5
else attenipt=attenipti
I.
else
display
“link
failed
tty
another”
4.
If
attempt
<=
10
then
goto
step
I
5.
stop.
The respotise time t,sfor the agent is given as
where N=tiuniber of nodes,
tr,=
E,”=,
tc
+
2*
CElfser,
n=2*(N-l)=no of
hops
tc= communication time or transmission time
ts,r=service time
of
agent at node(waitingiexecuting)
A
partial implementation of this kind of architecture is simu-
lated using AWB and CORBA (common object request broker) for
teleconferencing test bed[5]
.
AWB is aglet work bench for imple-
mentation of mobile agents. Aglets are piece
of
code in java which
can migrate from node
to
node and execute. Interagent commu-
nication is achieved through two mechanisms:message passing of
AWB; ORB(object request broker) invocations. Several overheads
are associated with this agent based approach for QoS negotiation.
They are: comunication time; mobile agent has to be transmit-
ted
form
one node
to
another
(
for
ex., a 2mb mobile agent needs
0.2sec time for migrating on a
IO
mbps network); memory space
is
required to store the mobile agent and create mobile agent plat-
form;
CPU
time for computation of mobile agents.
Fig.
3.
Simulation model for QoS negotiation
The local negotiator agent at all the nodes generate random val-
ues for available bandwidth and stores in shared
file
(QoS-profile).
The
QoS
negotiation mobile agent is hosted from jala which mi-
grates to next node in the path, negotiates with local agent through
a shared
file
consisting
of
available bandwidth at that node. The
mobile agent after negotiation returns back to
its
host in the reverse
5
Results and Discussion
We have carried out several analytical and simulation experiments
using different types of mobile agents for different
QoS
parameters
allocation. One of results of several experiments is presented here
for discussion. One
of
the requirements in mobile agent based sys-
tems is to incur low delay in migrating agents.
Thus
we measured
the time taken for the agent
to
complete it task
of
bandwidth nego-
tiation. This time is referred to
as
response time of the agent. The
constant parameters which have been considered
for
computation
of response time of agent are C=lOmbps, x=4kilobits,
tser=l.S
sec
and 0.8sec for three node and two node system, timeout interval
T=4sec and the steady state probabilities are obtained as given in
section
3.
The response time is computed for both analytical and
simulation models using the above equations given in section
5
and
410
6.
The results are tabulated in Table
I
for
N=3 and
N=2
(N
repre-
sents the number of nodes in the network). Fig.4 shows response
time
of
the agent with increasing number of retransmissions for
three nodes (N=3) and two nodes
(N=2).
It
is clear from the graphs that response time increases with num-
ber
of
transmissions,
so,
we
should reduce the number
of
retrans-
missions
to
get better performance
for
real time applications.
It
can be observed from both Fig.4(a) and
(b),
that response time de-
creases with increasing number of retransmissions for two node
network as compared
to
three node network. This indicates that,
source hosting agent should try
to
choose the shortest route
to
the
destination.
One
of
the important aspect is to predict the agent failures
in
allocationg the required
QoS
parameter
(
bandwidth)
to
the appli-
cation. We considered a fully connected five node network with
link bandwidth= lOmbps and total network capacity
=
lolinks
X
l0mbps
=
100mbps. Maximum number of agents that can be gen-
eratcd
in
the network are assumed to be
25.
Each agent can re-
quest
for
bandwidth between
1
to
IOmbps.
It is observed in the
simulation, that, the agent rejection percentage increases with the
incrcasc in the number
of
mohile agents, thus affecting the admis-
sion of new multimedia applications. The rejection starts at
A=
0.4
(agents generated are
=
0.4
X
25
=I2
).
Thc rejection percentage
is
0.135
for
A=
0.4 and rises to 0.420
for
A=
1
.0
.
TABLE
I
(
N=3
and
N=2)
Counterproposal:
For an unsuccessful flow negotiation at any
intermediate node, a counterproposal can be made by local nego-
tiator agent at that node. This counterproposal will be cartied to
agent hosting node by flow negotiator mobile agent, which may
tune its application parameters to suit the proposal
or
reject it. For
example, it
is
required
to
negotiate for a bandwidth
of
4mbps, con-
sider that negotiation failed at node
3
in a four node network path.
The local negotiator at node
3
makes a counterproposal for 3mbps,
then the flow negotiator agent will travel with this proposal to host-
ing node
1
and communicates with local negotiator agent at node
1.
If
it
is acceptable, then counterproposal is succesful.
The
time
required to make this counterproposal is twice the time needed
for
the first proposal
(
for ex., response time for first proposal
=
1.7
sec, response time for counterproposal
=
3.1 sec).
6
Conclusions
The paper presented different levels of
QoS
parameters,
QoS
re-
quirements and
its
negotiation and management techniques in mul-
timedia communication. The partial implementation of mobile
agent based bandwidth negotiation has been shown.
It
reduces
bandwidth and network traffic required for
QoS
management. The
results showed that response time
of
agent depends on the vari-
ous
factors such as delayed execution,
lost
agent, corrupted agent,
number of nodes in the path.
So,
always agent host should try
to
choose
a
shortest path for mobile agent
to
respond immediately.
We also observed through simulation, that the increase in arrival
of mobile agents would affect the admission of new multimedia
application.
References
[I]
C.
Aurrecoechea, Andrew T. Campbell, Linda Hauw,
“
A sur-
vey of
QoS
architectures”, Multimedia
systems,
Vol
.6,
1998,PP.
138-
15
I.
[2]
D.
Wong, N. Paciorek,
D.
Moore, “Java based mobile agents”,
Comniunicufions
ofACM,
Vol42,
No.3,
March 1999.
[3]
V.Oham, A. Karmouch,
’‘
Mobile software agents: An
Overview”,
JEEE
Cornrnrtnication
mugazine, July 1998,
PP.
25-
37.
(41
A. Puliafito,
0.
Tomarchio, H. Meer,
“
Agent based framework
for
QoS
management”, 4th int.conference
on
Atialytical
and
nu-
merical modeling tech.-QoS nrodelling, singupore, sep. 1997.
[5]
L.A.Guedes, P.C.Olivcra, L.F.Paina, E.Cardozo,
“
An
agent
based approach for supporting quality
of
services in dis-
tributed multimedia systems”, Conpurer Comnlrtnications,
V01.21, 1998,
PP.
1269-1278.
Fig 4. Response time of agent Vs Retransmissions
for
N=3 and
N=2
411