Adaptive Bandwidth Reservation Scheme for Multimedia Traffic
Using
Mobile Agents
S.
S.
Mami* and P. Venkataram**
Protocol Engineering and Technology (PET) Unit, ECE Department,
Indian Institute
of
Science, Bangalore
-
5600
12,
INDIA,
*E
-
mail
:
smanvi@ieee
.
org,
http
:
//pe t. ece .iisc
.
ernet .in/sunil
**E
-
mail:
pallapa@ece.iisc.ernet.in,
http://pet.ece.iisc.ernet.in/pallapa
,
Abstract
The bandwidth reservation for multimedia traffic poses technical challenges due to bursty and delay sensitive nature of
applications. The objectives of bandwidth reservation schemes are: optimize network utilization, minimize the packet
losses and delays. The growth of multimedia services on Internet and the possible discovery of programmable
networks has made
us
to investigate new techniques for resolving bandwidth issues in multimedia communication.
Mobile agent technology seems to be the promising solution for network management and
QoS
control.
In
this paper,
we consider joint problems of adaptive bandwidth reservation and link rearrangement (rerouting) for multimedia traffic
under the event of
congestion/failures of link, and propose a mobile agent based approach to achieve these objectives.
The scheme is simulated using a multimedia
traffic model. Simulation results show that the use of agents increase the
network utilization, acceptance ratio of applications, flexibility and efficiency
of
bandwidth reservation. The flexibility
in using agent technology is that the policies can be changed and implemented easily by encoding in the agents.
1
Introduction
Multimedia applications such as video
-
conferencing, tele
-
conferencing, on
-
demand services, etc., are less data loss
sensitive, more delay sensitive and
bursty in nature, and
consume significant amount of network bandwidth. They
require real time delivery and acceptable
QoS (Quality of
service). Therefore efficient bandwidth reservation
schemes are needed to ensure better
QoS.
Static
bandwidth
reservation schemes reserve only once after
admitting the application, such as peak rate reservation,
but under
-
utilizes network bandwidth due
to
variable bit
rate
(VJ3R)
services
[I].
Adaptive bandwidth reservation
schemes reserve bandwidth online according to the
implicit
or
explicit feedback about the network state from
intermediate nodes and receivers
[2,3],
pricing policies,
application priorities, queuing based reservation, etc.
Since conventional
IP
routing uses shortest path
technique for packet routing, network load will be
unevenly distributed, paths get congested and degrade
performance of the network. Hence, it is necessary to
consider the joint problems of rerouting (rearranging
flows) and adaptive bandwidth reservation under the
congestion/link failure situations to optimize network
utilization. Self healing and backup virtual path type
schemes are used for restoration of network during
failures and reserve bandwidth on the alternate routes
[4,5].
Rerouting strategies that can be used are source
routing, local rerouting, local destination routing. The
work given in
[6]
considers a new self healing scheme
that re
-
reserves bandwidth for differentiated bandwidth
requirements of applications during rerouting under
linWnode failure situations. Some of the works done on
adaptive bandwidth reservation and rerouting under
traffic load variations are based on bumping packets,
least loaded paths, minimum delay routes, partition
policy with reservation on a link, markov decision
process and path rearrangement schemes
[7].
Mobile
agent based bandwidth adjustments at source for point to
multipoint connections subject
to
network conditions
and bandwidth reservation at network nodes over a given
path are suggested in
[8,
91.
Mobile agents are also used
to learn the load on the network and update routing
tables
so
as to balance the load in network
[
101.
In this paper, we will describe a mobile agent based
adaptive bandwidth reservation and rerouting based on
congestion occurrence or failure of a link. The scheme
assumes availability
of
mobile agent platform and
bandwidth reservation agency. The agency comprises
of
mobile agents, static agents and resource handling
information at each node in the network. We also
assume that disjoint alternate routes to patch up a link
are available in the node. These alternate paths are
supplied
to
all the nodes by a central topology manager.
Mobile agent platform comprises of agents, agent server,
interpreter and transport mechanisms for agents (TCP
or
E
-
mail). Agent server is responsible for receiving mobile
agents and sending it for execution by interpreter.
0-7803-7600-5/02/$17.00@2002
IEEE
.
3
70
A mobile agent at the node is triggered depending on
traffic thresholds or failures monitored on the links. In the
event of congestion occurrence on link (monitored traffic
>
threshold) orlink break down, mobile agent of each
attached node of link picks an disjoint alternate route for
that link from alternate routing table to patch up that link,
and travels to capture and reserve the required bandwidth
for rerouting the applications. The scheme intuitively
does load balancing and tries to evenly distribute the load
in network there
-
by increasing network utilization.
Following sections explain the mobile agents, proposed
scheme, multimedia traffic model and results.
2
Mobile Agents
Agents are the autonomous programs situated within an
environment, which senses it and acts upon it using its
knowledge base, and learns
so
as to act in future. They
have certain special properties (mandatory and
orthogonal) which make them different from the standard
programs. Mandatory properties of the agents are:
autonomy, reactive, proactive and temporally continuous.
The orthogonal properties are:
communicative, mobile,
learning and believable
[ll].
Mobile agents are the
multiagent systems which posses the mandatory
properties and some or all of the orthogonal properties
specified for an agent. It is
an
itinerant agent dispatched
from source host which contains program, data and
execution state information, migrates from one host to
another host in the heterogeneous network and executes
at remote host until they accomplish their task.
The mobile code should be platform independent,
so
that,
it can execute at any remote host in the heterogeneous
network environment. They communicate and cooperate
with other agents
to
achieve its goals. Agent can update
its information base while interacting with other agents
during its travel. Inter
-
agent communication can be
achieved by message passing,
WC
or common
knowledge base (black board). Agents can be written in
Tcl, Perl, Java and
C++
languages
[12].
In general, there
are several good reasons for using mobile agents: they
reduce network load; overcome latency; encapsulate
protocols; execute asynchronously and autonomously;
adapt dynamically.
3
Mobile agent based bandwidth
reservation scheme
The scheme assumes the availability of mobile agent
platform, bandwidth reservation agency and resource
handling information at the nodes in the network. Mobile
agent platform is responsible for receiving agents, agent
execution, transporting agents and inter
-
agent
communication. The details
of
bandwidth agency and
proposed scheme are given in the following subsections.
3.1
Bandwidth reservation agency
Bandwidth reservation agency comprises of several
agents and resource handling information. Figure
1
depicts the agents and their interactions. It consists of
manager agent, link monitoring agents, scheduler agents,
bandwidth
(BW)
capture mobile agents and profiles.
Figure
1.
Bandwidth reservation Agency.
The functions
of
the agents and agency database are as
follows:
Link
monitoring agent:
it is a static agent
responsible for monitoring the traffic and link
failures, measurement is done using time
window mechanisms, stores the monitored data
(incoming and outgoing traffic) in link
information profile and exchanges the
information about buffer occupancy in attached
nodes of link. It also computes the link status
by comparing sum
of
reservedallocated
(monitored) bandwidth, buffer occupancy of
attached nodes of link with traffic threshold and
stores status as congested, normal, fail in link
information profile.
Manager agent:
it is a static agent which
triggers monitoring agents, scheduler agents,
mobile agents and interacts with topology
manager to get new alternate routes during
topological changes. This agent continuously
monitors status of each link, if
congestedfailed,
calculates bandwidth requirements
of
excess
traffic, triggers mobile agent to capture and
reserve the bandwidth.
Scheduler agent:
it is triggered by manager
agent, after mobile agent captures bandwidth. It
schedules bandwidth for the applications to be
rerouted on alternate route.
BW
capture mobile agent:
it is triggered by
manager agent during
congestion/failures. This
is responsible for reserving bandwidth on the
alternate route by interacting with the nodes in
371
the path through spare capacity and reservation
tables. During onward journey, tentative
reservation is done and confirmation is given
during return journey.
Agency database:
it comprises of spare capacity
and reservation tables, link information,
alternate paths, and reservation for applications
on links. Spare capacity
is
the residual capacity
of link after reserving bandwidth for
applications on that link. Alternate path database
consists of patch up disjoint routes
for
all the
links
of
a node. Link information database
comprises of monitored data (traffic load,
congestion status, failure
status,
buffer
occupancy of node and adjacent node of the
link) of all the links of a node. Reservation
database contains information about the
applications running on the links and their
bandwidth scheduled. The reservation database
will be modified by scheduler agent during
rerouting.
3.2
Bandwidth reservation by mobile agents
The working of the scheme is illustrated by considering a
partial network topology as shown in Fig
2.
Let
us
say
that, at certain instant of time, the link
1
-
4
may get
congested
or
fail (congestion occurs if bandwidth
reserved
+
buffer occupancy for link
>
threshold; failed,
if
no
traffic on link). Manager agents in node
1
and
4
bigger their mobile agents to capture and reserve
bandwidth
on
the alternate paths for rerouting. These
alternate paths are supplied by the centralized
topology
manager
that also sends updated alternate paths for link
patch up in case of additioddeletion of nodes. The
changes in topology are fed
from the network
nodesladministrators.
Sl
I
I
Figure
2.
Bandwidth reservation and routing by Mobile
agents.
The mobile agent on node 1 (assume that the node with
lower
numbered address of the affected link will choose
first path) chooses the first alternate path (1
-
2
-
4) and
mobile agent on node 4 will choose second alternate path
(4
-
6
-
1) for their travel. Then, mobile agents inform to
the sender nodes about reserved bandwidth.
Mobile agents initially reserve spare bandwidth available
on links of alternate routes (minimum of required and
spare bandwidth).
On
reaching destination (4
or
1)
they
will pick the minimum of bandwidth reserved, confirms
bandwidth reservation while returning to sender node.
Agent releases the excess bandwidth reserved. After
reservation at node, mobile agents update routing tables
of the nodes in alternate path. Once bandwidth
reservation is confirmed, nodes will schedule
applications of
failedkongested link on alternate routes.
Every node contains a database of spare capacity
available on its links and, spare capacity reserved. This
will facilitate mobile agents to capture bandwidth.
The scheduler agents at the nodes will schedule the
applications
on
the alternate paths. The scheduling can
be based on priority of classes of traffic, if bandwidth
captured is not enough to support the overloaded traffic.
During such situation, some applications may be rejected
or
delayed. The following pseudo
-
code (algorithm) gives
description about the functioning of the scheme. The
algorithm is repeated periodically at each node in the
network. The symbol
li
represents the
i*
link of a node.
Algorithm: Adaptive Bandwidth Reservation
Begin
1.
Initialize
i=I,
n=
number of links of a node;
2. Manager agent at a node checks a link
(li)
status using
link information profile;
3.
If (Congestiodfailure of link
Zi
C
TRUE) go
-
to step
7;
4. Manager agent sends mobile agent
on
alternate routes
to capture and reserve bandwidth (uses spare capacity
and reservation tables of node);
5. Mobile agent confirms reservation
on
return travel,
update routing tables and inform the manager agent;
6.
Scheduler agent schedules bandwidth for applications
on alternate route, if sufficient bandwidth
is
available,
else reject some of the applications;
7.
i=i+l;
If
(
i
0
n
)
repeat steps 2
-
5 for link
Zi
of node;
8.
STOP.
End.
4
Multimedia traffic model
We assume two traffic classes which are more
prominently used in network modeling. They are
continuous
bit
rate services (CBR) and variable bit rate
services
(VBR).
VBR sources are modeled as on
-
off
models. The network model considered is a collection of
'J'links,
{ll,12,..,lJ},
where each link has a capacity of
C
bandwidth units. The network supports two classes of
traffic. Associated with each traffic class there is an
anival rate
&
and
h,
respectively, such that
&+
L=1,
large holding time and bandwidth requirement. Suppose,
if there are
100
sources contributing to the traffic, if
h,
372
=0.4,
then
&=0.6,
thus the number of class
1
traffic calls
will be
0.4* 100=40 and class
2
will be
60.
A
route for an
arrived
applicationkall is a subset of links. An arriving
caWapplication is admitted into network with its route,
r,
such that
r
is a subset
of
L,
which is a predetermined
route randomly selected from route set
R={r,,
rz,
...,
r,,
where
ri
is a subset of
L,
such that
L={ll,
...
..,IJ/'.
These
routes are uniformly distributed. Probability
of
a link
being chosen for the route in the route set is given as,
p
=I/(number-of-links-in-route),
and the offered load on a
link
is
product of
p
and bandwidth units requested by
source. This model has been used in simulation to
generate the background traffic. The bandwidth units for
class
1
(CBR) traffic is assumed to be uniformly
distributed within some range
(a,b). Similarly bandwidth
units for class 2 (VBR) traffic is uniformly distributed in
the on period within some range
(x,y). The burstiness of
VBR traffic is given by
on/(on+off). We have considered
probability of being in on period as Bernoulli distributed
with value
Pb.
Traffic is generated for an VBR
caWapplication, if it is in on period.
5
Results
and discussion
We have considered a partially .connected network
topology of
16
nodes for simulation as shown in Figure
3
under congestion and link failure situations.
Simulations have been
carried out extensively with
different random number seeds. The results presented
here are an average of them.
Ilrk
1415
moddod
u
mnpaodl
6iledlirJk.
AltmMtarcnts:
-.scar-=lrtpl
Figure
3.
Network topology for simulation
The input parameters used for simulation are: link
capacity=
20
Mbps; VBR traffic sources peak rate are
uniformly distributed in the range
(0.2,0.4) Mbps;
Pb,
Probability of being in on period (utilization of source)
for VBR source is
0.7;
CBR traffic is uniformly
distributed in the range
(0.3,0.5)
Mbps;
&=La=0.5,
probability of class
1
(CBR) traffic arrivals for
background and foreground traffic;
hy3La=0.5,
probability of class
2
(VBR) traffic arrivals;
pmc
(probability of agent capturing bandwidth on alternate
route) values considered are 1.0,
0.8
and
0.6;
threshold
=
25.0
Mbps; background traffic sources
(N)
are
300
and
400; number of additional sources on congestedfailed
link are varied from
20
to
200
sources.
Congestion is
triggered on link 14
-
15, after background
traffic is generated by
N
sources. Figures
4
and
5
depict
-
the simulation results under congestion situations. We
observe that the network utilization increases with use of
mobile agents. After certain number of applications, it
saturates due to unavailability of bandwidth on
alternate
routes (14
-
13
-
15, 15
-
16
-
14). We also observe that
percentage of applications rejected for rerouting
increases when utilization saturates. This is quite
obvious, because,
residualhpare capacity is not
sufficient on alternate route
to
divert all the existing
applications. It is observed that network utilization
decreases and rejections increase with the decrease in
probability of mobile agent
@,==I
.O,
0.8,
0.6)
captbring
bandwidth, these detailed results are not presented due to
page limitations.
NdWdutilhrion.V%NOfk&C-
Figure
4.
Network utilization with and without mobile
agents
(u-woma).
Vs.
Number of traffic sources with
Pmc=l and
N=300
and 400.
Figure
5.
Applications rejected
(%)
.Vs. Number of
traffic sources with
N=300 and 400 and Pmc=l
.
Similar results are shown in figures
6
and
7
for link
failure. These figures indicate that the proposed scheme
increases network utilization and decreases rejections of
applications. The reserved bandwidth depends on
existing traffic and hence we observe that rejections will
increase when existing traffic is raised from
300
to
400
sources for both congestion and link failure situations. It
is also observed that
as
threshold value is increased,
utilization decreases and rejection ratio reduces.
373
We have also computed the time complexity of mobile
agents,
i.e., time required by mobile agent to dynamically
allocate bandwidth. The assumptions made for
computation time are: agent migration time is uniformly
distributed between
(0.3,
0.5)
seconds, and agent
execution time is exponentially distributed with mean
execution
time=O.
1
seconds. The simulation results show
that the computation time varies in the range of
2.2
to
2.5
seconds for increasing number of applications on
congested or failed link with constant values of
background sources.
120
110
Figure
6.
Network utilization with and without mobile
agents(u-woma). Vs. Number of traffic sources on failed
link with
Pmc=l and N=300 and 400.
Figure
7.
Applications rejected
(%)
.Vs.
Number of
traffic sources on failed link with N=300 and 400 and
Pmc=l
.
6.
Conclusions
The paper considered joint problem of adaptive
bandwidth reservation and rerouting to achieve optimal
network utilization and exploited uneven network load
distribution. Proposed Mobile agent based adaptive
bandwidth reservation scheme is a flexible and scalable
approach. The scheme triggers adaptive bandwidth
reservation based on traffic load variations and link
failure. Several other factors such as, policy changes by
administrator, delays cropping up, priority can also be
considered. Agents allow encoding of all these factors in
them for bandwidth reservation, thus providing
flexibility and scalability. Simulation results
demonstrated the optimal use of network bandwidth and
the feasibility of using mobile agents especially in
heterogeneous nodes in the networks. However, several
overheads are associated with the scheme such as: agent
transport time and bandwidth; memory space for agent
platform;
CPU
slots for agent execution; security to
hosts and agents. But, these overheads are nullified
owing to advantages of using mobile agents for adaptive
bandwidth reservation.
References
[l]
LA. Sariot Ismail,
"
Bandwidth problems in high
speed networks
"
,
IBMJ.Res.
Develop,
Vol.
44,2000.
[2] M. Mowbray,
G.
Karlsson, and
T.
Kohler,
"
Capacity
reservation for multimedia traffic
"
,
Distr.
Syst
Eng.,
Vol.
[3] Y. Thomas
Hou,
Bo Li, Shivendra
S.
Panwar,
Henry
Tzeng,
"
On
network bandwidth allocation policies
and feedback control algorithms for packet networks
"
,
Computer networks,
Vol. 34,2000, pp. 481
-
501.
[4] Komut Wipusitwarakun, Hideki Tode, Hiromasa
Ikeda,
"VPls
priority based restoring function enhanced
self healing algorithm
"
,
IEICE trans.
Comn.,
Vol.
E81-B,
[5] Ayano Yamashita,
R.
Kawamura, Hisaya Hadama,
"
Dynamic VP rearrangement in ATM networks
"
,
Proc.
IEEE
Globecom,
1995, pp. 1379
-
1383.
[6] Taishi Yahara, Ryutaro Kawamura, Satoru Ohta,
"
New self healing scheme that realizes differentiated
bandwidth requirements on ATM networks
"
,
IEICE
Trans.
Comn.,
Vol.
E83-B,
No.
3,2000,
pp.
672
-
679.
[7] Eric W.M.Wong, Andy K.M.Chan, Sammy Chan,
King Tim KO,
I'
Bandwidth allocation for virtual paths in
ATM Networks with dynamic routing
"
,
IEICE
Trans.
Comn,
Vol. E83-B, No. 3,2000, pp. 626
-
637.
[8]
S.S.
Manvi and P. Venkataram
,
"QoS
management
by mobile agents in multimedia communication
"
,
Proc.
DEXA
(Workshop
on
ABIS),
Sep.
2000,
pp. 40741
1.
[9]
S.S.
Manvi and
P.
Venkataram,
"
Mobile agent based
online bandwidth allocation for multimedia
communication
"
,
IEEE
Globecorn,
Nov. 2001, pp.
2622-
2625.
[lo]
Steffen Liperts and Birgit Kreller,
"
Mobile agents
in telecommunications
--
A
simulative approach to load
balancing
"
,
Proc.
ISAS,
1999.
[
1
1
]
http://www.cetus
-
links.org/oo
-
mobile
-
agents.html.
[
121
http://www.cs.umbc.edu/agents
5,
1998, pp. 12
-
18.
N
O
. 11,1998, pp. 2100
-
2109.
374