An efficient CAC scheme for IP traffic over wireless ATM networks

Abstract

Traffic control and resource management are complex issues in wireless asynchronous transfer mode (WATM) due to broad range quality-of-services (QoS), limited bandwidth spectrum and susceptibility to error. In WATM, traffic control is essential in order to protect the network from congestion and to achieve realistic network efficiency in compliance with the QoS. Call admission control (CAC) is one of the critical mechanisms in providing QoS. This paper describes bandwidth allocation algorithms for IP traffic over WATM network based on VCT cell cluster threshold. Simulation work using network simulator (NS2) is also carried out to observe the network performance based on the proposed bandwidth allocation.

Full text

Page 1

An Efficient CAC Scheme for IP Traffic over
Wireless ATM Networks
H. Zainol Abidin’, N Fisa?, M.S.Ansari
Faculty of Electrical Engineering,
University Teknologi Malaysia,
81310 Johor Bahru, Johor Darul Ta’zim.
Email: 2sheila~suria.&e.utni,my
Abstract-
complex issues in wireless asynchronous transfer mode (WATM)
due to broad range quality-of-services (QoS), limited bandwidth
spectrum and susceptibility to error. In WATM, traffic control is
essential in order to protect the network from congestion and to
achieve realistic network efficiency in compliance with the QoS.
Call Admission Control (CAC) is one of the critical mechanisms
in providing QoS. This paper describes bandwidth allocation
algorithms for IP traffic over WATM network based on VCT cell
cluster threshold. Simulation work using Network Simulator
(NS2) is also carried out to observe the network performance
based on the proposed bandwidth allocation.
Traffc control and resource management are
Keywords- Wireless ATM; CAC; QoS; effecfive bandwidlh; IP
1.
INTRODUCTION
Asynchronous Transfer Mode (ATM) is a data transport
technology that supports a single high-speed infrastructure for
integrated broadband communications involving voice, video
and data. Bandwidth efficiencies are achieved through
statistical multiplexing of transmission bandwidth. ATM
networks are characterized by virtual channel connections
(VCCs) that cany small, fixed size packet (53 bytes) called
cells within the network irrespective of the applications being
supported. ATM is able to handle a wide range of information
bit rate together with various types of real-time and non real-
time service classes with different traffic attributes and quality
of service (QoS) guarantees at cell and call levels.
Call Admission Control (CAC) is one of the critical
mechanisms in providing QoS. In CAC, network attempts to
deliver required QoS by allocating an appropriate amount of
resources such as bandwidth. Bandwidth allocation in ATM is
usually done by obtaining the effective bandwidth of the
sources.
The virtual connection tree (VCT) concept as described in
[1,2] reduces the call setup and routing load on the network call
processor (NCP), while maintaining the QoS of the handoff
calls. The VCT is a collection of cellular base stations (BSs),
which is called the cell cluster, while the backbone network or
the wired network, which consists of switching nodes and links
as shown in Figure 1.
0-7803-8114-9/03/$17.00 02003 IEEE,
Figure 1. Svucfllre of VCT
This paper presents and analyzes the bandwidth allocation
algorithms for IP traffic over WATM. We assumed the
network operates based .on VCT cell cluster threshold. The rest
of the paper will be organized as follows. Section 2 will focus
on wireless ATM (WATM). Then, description on Call
Admission Control (CAC) including statistical multiplexing
and IP trafic will be introduced in section 3. Section 4 outlines
the proposed bandwidth allocation technique while section 5
will briefly discuss the simulation work. Finally, section 6
concludes the paper.
11.
WIRELESS ATM (WATM)
Wireless communication offer borderless freedom to
customers to using mobile devices such as phones, television
and wireless computers. According to some forecast the
number of mobile phone subscribers may reach one billion by
the year 2010 and surpass fixed phone lines [3]. As market
demands for wireless communication continue to grow rapidly,
it is apparent that a new generation of wireless network; will he
needed for two main reasons:
To enable different wireless technologies to
intenvorking seamlessly with existing wired networks;
To meet the diverse traffic demands (e.g. voice, video,
data) required by current and future customers.
556

Page 2

The convergence of ATM in the wired and wireless
domains serves as an effective platform in achieving these
objectives. This trend can be found in several projects within
research program of the European Community (EC). There, a
new generation of mobile communications technology had
been introduced as the Universal Mobile Telecommunication
System (UMTS) for the RACE project, which focus on the
design of the Mobile Broadband System (MBS), a wireless
cellular network fully integrated into 9-ISDN. This mobile
application can reach an approximate range of 2 to 155 Mhls
and RACE mobile is now studying a Mobile Broadband
System operating in 60 GHz hands 141.
Ill. CALL ADMISSION CONTROL (CAC)
Call admission control (CAC) is the first line of defense for
a network in protecting itself from excessive loads. In WATM,
the CAC function is located in the BS. Usually, the CAC
function has to he called before a handoff procedure can be
completed. In order to guarantee a certain degree of mobility
for the wireless ATM terminals, the CAC function needs to
-reserve some of the available radio bandwidth for handoff
calls. A handoff call drop results an intemption of on going
call and occurs if the bandwidth required by a connection
cannot be supplied in the new cell site location.
In virtual connection tree (VCT), the CAC function is a
mechanism that limits the incoming calls into the network in
order to protect the already connected calls from being
intermpted. An accepted connection that is requested by a
mobile terminal (MT) will he given pre-established virtual
connections to all the BSs in the tree so that the MT can freely
move within the tree. The VCT nehvork also monitors the
existing connections in the network in order not to exceed the
pre determined threshold limit. Moreover, it uses a scheme
that gives priority to the handoff calls where a percentage of
channels are resewed for handoff calls whenever the channels
in the BSs are busy.
A.
In ATM network, several sources will superimpose on a
single link such as trunk line that may carry hundreds of
videophone calls. Furthermore, the eficiency is .gained hy
relying on the statistical multiplexing effect sources, on the
.condition that enough sources are multiplexed and they are not
correlated.
The 'issue .of resource allocation to connection is. a
challenge, due to the lack of physical separation between
.resources used by different connection .in ATM based
networks. The resources are actually the carrier system, which
are for high capacity networks, designed for the digital
transmission of voice, video and data. In order to offer efficient
resource allocation, we could apply different bandwidth
allocation techniques for different types of IP traffics which
will be outlined in next section.
As we are focusing o n the provisioning QoS of IP trafic
over WATM, we proposed that IP traffic is classified as in
Differentiated Services (DiffServ) manner.
Statistical Multiplixing arid IP Traf/ic
Dif€Serv offers QoS by allowing priority scheduling to
facilitate the multimedia applications over the Intemet. Figure
2, shows the structure of DifBerv domain. The DifEerv
specifications refer to the forwarding treatment provided at a
router as per-hop behavior (PHB). PHB must be available at all
routers and normally PHB is the only part of DiffServ
implemented in core router.
Figure 2. DifBerv domain
The Intemet Engineering Task Force (IETF) has currently
specified two different PHBs known as Expedited Forwarding
(EF) and Assured Forwarding (AF) [SI. EF traffics are
normally given strict priority over the traditional best-effort
(BE) traffic inside the DifEew domain. However, each flow
has to specify tbe required bandwidth in advance so that the
appropriate resources can be reserved inside the network.
Furthermore, the edge router will police each flow and the non-
confonnant packets will either he dropped or shaped. Since A F
does not offer hard QoS guarantees, lETF has specified four
different AF class. Each class is assigned a certain amount of
bandwidth at each node. When the amount of traffic exceeds
this bandwidth, packets will be dropped according to their drop
precedence value.
In this paper, EF, AF and BE traffic will be represented by
CBR, VBR and ABR or non real-time sources respectively. It
is assumed that the BS functions similar to ingress node in
DiffServ domain while the wired ATM network does not
require PHB at all their nodes. It is also assumed that BS is
able to recognize EF, AF and BE as CBR, VBR and ABR
respectively.
Iv. BANDWIDTH ALLOCATION
WATM allows transfer of various types of multimedia
traffic and ensured guaranteed QoS. A CAC in WATM will
determine the amount of bandwidth to he allocated for each
source accepted into the network. The ATM Forum has
categorized the various types of source traffic into CBR, VBR
(rt-VBR and nrt-VBR), UBR and ABR: The bandwidth
allocation mechanisms differ for each time of demand. The
following sub-section explains the bandwidth allocation
techniques for CBR, VBR and UBR ad ABR
A. Peak Bandwidth for EF (CBR) Sources
A deterministic rate such as voice sources, usually hold one
unit of source for the whole duration of the connection. Figure
557

Page 3

3 illustrates the state-transition-rate diagram for m-server (time-
slot) loss system with Markov amval and service process.
(-744
8kB,21kB
5.7kB, 16.3kB
18.lkB, 37.6kB
Figure 3. State-transition-rate diagram for W d m
The probability that the systems having k calls, pk can be
obtained as follows:
(1)
B
loom
2ookB
300kB
27.62kB
19.66kB
71.88kB
24.32kB
17.98kB
54.76kB
23.21kB
17.42kB
49.04kB
call depart& rate. Hence, the fraction of time that all m
timeslots are busy,p, is determined as;
(W)
8kB, 21kB
5.7kB. 16.3kB
18.lkB, 37.6kB
( l l p ) m l m !
c(1/~)~
i=o
This probability expression is known as Erlang's loss
formula. Thus, Erlang's loss formula is generally derived &om
W m / m queue. This system has a blocked calls cleared
situation in which there are available m servers (timeslots).
Normally, CBR connection is given a fixed peak bit rate such
as 64kbls for voice [6]. The Erlang's loss formula also
indicates Bp which is equal the peak rate will be allocated for
EF sources.
ik!
P m = I
,B
look6 ZOOkB
300kB
22.63kB
37.52kB
40.87kB
22.49kB
17.42kB
40.60kB
22.41kB
17.36kB
40.44kB
B. Efective Bandwidth for AF Sources
Finding the effective bandwidth in ATM connection is
important in .order to maintain the QoS of the connection and to
ensure that the connections are used efficiently for variable bit
rate traffic. AF sources such as video traffic can be represented
by Gaussian distribution. The effective bandwidth for a
Gaussian i.i.d random process, x(n) can be calculated by using
the following equation [7].
where p is mean arrival rate and d i s the variance. 6 is
determined by using the following equation
,
'=[
Typically, ais set to 1. E is the loss probability and B is
the buffer size. Calculation is done by considering the
following parameters. a = 1, E = IO' and B varies from lOOkB
B
]
O<a<l
(4)
In(a) - In(&)
to 300kB. U and p will be shown in the table in the form of
(U#) in kB. (u,p) values are taken from reference [PI. Table I
shows the effective bandwidth for AF sources by applying
equations (3) and (4).
TABLE 1. EFFECTIYE BANDWIDTH
FOR VBR SOURCES
USING GAUSSLAN 1.I.D
The result shows that Gaussian i.i.d seems to be not
conservative in providing effective bandwidth. Therefore, a
more conservative effective bandwidth derivation has been
adopted based on self-similar traffic known as Pareto
distribution. Pareto distribution is suitable to represent modem
network traffic such as VBR, Web etc. The effective
bandwidth, BE can be determined by using Pareto distribution
as follows:
where, p is the mean arrival rate of the traffic stream in bps, a
is the variance coefficient which is calculated as the variance to
mean ratio, B is the buffer size, His the Hurst value and finally
E is the lost probability. Function of Hurst parameter, AH)
could be determined by using the following equation.
By using the same parameters and H = 0.85, the effective
bandwidth required by AF traffic calculated by using equations
(5) and (6) are shown in Table 11.
TABLE 11.
EFFEC~VE
BANDWOTH
DISTRJBL~ION
FOR VBR SOURCES
USMG PARETO
From the above results, it is proven that Pareto distribution
tends to be more conservative in allocating resources when
network traffic exhibits long-term memory. Therefore,
bandwidth that will be allocated in our network would he
based on the effective bandwidth obtained from the Pareto
distribution calculation.
C.
Best effort IP traffic represents many of the non real-time
applications such as data coming from LAN. The arrival and
service time of IP packets can be approximated to Markov birth
and death process in an infinite queuing system. The birth-
Mean Bandwidth for BE Sources
558

Page 4

death process of such queuing system can be. illustrated as
state-transition-rate diagram of Markov Chain in Figure 4.
I
h h h
I
Figure 4. State-transition-rate diagram for infinite queuing system
From the state-transition-rate diagram, the equilibrium
different equation for state k can be determined as follows [IO]:
'k-CPk-1 +P'rtlPk+l =('k +pk)Pk
(7)
wherei. is the call arrival rate, p is the call departure rate andm
is the probability that the systems with k members. Hence, pk
can be simplified as follows:
where
1
Since 1 < p, the summation will converge, therefore
From the stability conditions, the utilization, p should be
O P < l to ensure that po > 0. The steady-state probability of
finding k customen in the system is:
By applying Little's formula, the average delay, E[r] is
obtained from E[t] = E[n]/h, where E[n] is the average number
of customers in the system. Thus,
" P
E[f] = -
1-p :
Figure 5 shows that when p = 0, E[r] is exactly the average
service time that is expected by a customer. Therefore it is
reasonable to allocate mean bit rate, B, for BE traffic.
V.
SIMULATION
WORK
Simulation work has been carried out to observe the effect
of our proposed bandwidth allocation mechanism to the
network performance for AF sources. Figure 6 shows the
simulation network model.
Figure 6. Simulation Network Model
The three trpes of sources used are as described below:
Source 1 (EF): CBR traffic is based on UDP transport
protocol with a rate of 96Okbps and mean packet size
of 1500 bytes. This traffic is suitable to represent
voice application.
Source 2 (AF): represents VBR traffic based on UDP
transport protocol with a mean rate of 4lkbps and
mean packet aize equal to 1500 bytes. This traf€ic
would be suitahle to represent video application.
Source 3 (BE): represent ABFUUBR traftic for non
real-time traffic with rate of 32Okbps and packet size
equal to 1500 bytes.
In the simulation work, the buffer size is assumed 100kb,
link rate is using TI line equals to 1.544Mb/s and scheduling
mechanism is based on Priority Queuing (PQ). A guard
channel of a cepin percentage is reserved for handoff of on-
going calls. The three types of traffic arrived at the BS
according to deterministic, Gaussian and Pareto distribution
representing IP traflic of type EF, AF and BE. In normal case
IP traffic is treated as ABRiLTBR traffic in ATM network.
Figure 7 shows the end-to-end delay for AF sources using
the proposed bandwidth allocation technique while Figure 8
illustrates the end-to-end delay for AF sources that are treated
as normal BE traffic. In the later case the effective bandwidth
of AF sources is B,
*,
Figure 5. Average delay as a function ofp for Wl
559

Page 5

I
a *.a
I..
8..
I.)
sa
11.1
,*.e
/.I.
..a
..il.*
Figure 7. End-to-end delay for AF with proposed bandwidth allocation
’ .r --
study has shown that the performance of the network in terms
Figure 8. End-to-end delay for AF treated as BE
The two graphs show that OUT proposed bandwidth
allocation technique with PQ scheduling for AF traffic gives
almost constant low (about 0.3 second) end-to-end delay
compared to treating AF traffic as normal IP BE traffic.
VI. CONCLUSIONS
Call admission control (CAC) is an important mechanism
in provisioning QoS for 1P traffic over WATM network. This
paper has proposed a simple bandwidth allocation technique
for various types of IP traffic sources. Our efficient CAC
function uses an efficient bandwidth allocation technique as
well as efficient scheduling method. Peak bandwidth, effective
bandwidth and mean bandwidth are allocated to CBR, VBR
and ABR sources resuectivelv. We assume that IP trafic over
the WATM network is usinisimilar DiffServ technique at the
ingress node (base station). This is to ensure that the different
types of traffic are given priority. Thus, DiffServ PHB at the
ingress node such as EF, AF and BE is represented by CBR,
VBR and ABR sources resoectivelv in WATM. The simulation
of end-to-end delay is minimized as we allocate effective
bandwidth to the traffic.
REFERENCES
[I] S.H.S. Aiffiin, N. Fisal and M. Esa, “Quality-of-Service Performance in
Micra Cellular Network,” Proceedings of World Engineering Congress
1999, Malaysia, pp. 63-67.
A’’
S.H.S. Ariflin, N. Fisal and M. Esa, Smdy of Mobility Management in
Wireless ATM Network,” Proceedings of N C T 98, Universiti Pum
Malaysia, Malaysia, pp. 125-131.
M. Shafi et. al. “Wireless Communication in the twenty Fin1 Century: A
penpective”. Proceedings ofthe IEEE. Vol. 85. No. 10. October 1997.
R. S. Swain, “UMTS- A Zlst Cenhq System: A RACE Mobile Project
Line Assembly Vision”. htt~:llwww.vtl.tiitt~l~~MTSl~mt~.
S. Bakiras and V.O.K. Li, “Efficient Resource Management for End-to-
End QoS Guarantees in DiftServ Networks,” IEEE International
Conference on C?mmuni&ons, 2002. ICC 2002,. V01.2, pp. 1220-
1224.
S.H.S. Aiffin, “Quality of Service Provisioning in ATM Wireless
Network under Real-Time and Non Real-Time Connections,” Master of
Electrical Engineeing (Telecommunications)
Kejuruteraan Eleklik, Univeniti Teknologi Malaysia, 2001.
K. Nagarajan and G.T. Zhau, “A New Resource Allocation Scheme far
VBR Video Traffic Sources,” Conference Record of the Thirty-Fourth
Asilomar Conference on Signals, Systems and Computers, 2000, V01.2,
pp 1245-1249.
P. Orenstein, H. Kim and CLLau, “Bandwidth Allocation for Self-
Similar Traffic Consisting of Multiple Traffic Classes with Distinct
Characteistic,”lEEE Global Telecommwications Conference 2001.
GLOBECOM ’01, Vo1.4, pp. 2576-2580.
1 . Nonos, “On the use of Fractional Brownian Motion in the ,meaty of
Connectionless Networks,” IEEE Joumal on Selected Areas in
Communications, vol. 13(6), pp.1028-1038, August 1995.
[IO] L. Kleinrack, “Queuing System, Volume 1: Theory,” New York, John
Wiley, 1975.
Html
Thesis, Fakulti
[Z]
[3]
[4]
[SI
’
[6]
[7]
[8]
[9]
560