Evaluating the Performance of Handoff Schemes in
Wireless ATM Networks
Farouk Smith, Neco Ventura
Dept. of Electrical Engineering, University of Cape Town
Rondebosch 7700, South Africa
Abstract—The performance of handover schemes in a Wireless The network architecture consists of three main parts as
ATM network is investigated and various handover schemes shown in the figure 1.
are discussed that have been proposed in the literature. The first part consists of ATM switches with standard
The impact of handoff schemes in terms of Service disrup- UNI/NNI and additional mobility support enhancements,
tion time, Handoff completion time, Buffer requirements at the
which form the mobility support ATM Network. These
Mobile Terminal (MT), Buffer requirements at the Access
Point (AP) and the excess bandwidth resources required dur- ATM switches are termed Mobility Enhanced Switches
ing handoff are addressed. (MES). The second part consists of ATM Access Points
The testbed to study the performance of the various hando- with Mobility Enhanced UNI/NNI and radio interface capa-
ver schemes is based on a gigabit ATM switch. The signalling bilities and lastly, Mobile Terminals with enhanced WATM
software of the ATM switch offers a direct many-to-many UNI.
communication, with connections that can be modified dy- One of the primary advantages of ATM is its ability to
namically. give QoS guarantees to connections. As mobile devices
With the proposed evaluation platform the authors will move between base stations, QoS re-negotiation may be
analyse the performance of the various handoff schemes and
required to maintain levels of service to the connections.
compare it to the analytical and simulation results obtained
from the literature. The aim of the handover protocol is to enable a wireless
terminal to move between access points seamlessly while
maintaining the negotiated QoS of its active connections.
The efficient handover and switching of the Mobile Termi-
nals (MT’s) virtual connections to the appropriate AP in the
The architecture proposed for wireless ATM is composed
wired network can achieve this.
of a large number of small transmission cells. A base station
The methodology for implementing hand over is one of
(also called an Access Point, AP) serves each of these cells.
the most important parts of WATM to support mobility ,
All the base stations in the network are connected via the
wired ATM network. As the cell size is reduced, hand over
The focus of the paper is on the performance of the new
rate also increases .
handoff schemes for integrated QoS based multimedia
Wireless ATM networks can be implemented by adding
services in WATM networks [2, 8]. Because new connec-
mobility support functions to fixed ATM switches. The
tions setup and signaling are necessary during handoff, it is
Wireless ATM architecture makes provision for the most
difficult to provide guaranteed QoS. The key requirement
important mobility support function, the handover protocol
for QoS is the re-negotiation of VC parameters during the
for inter and intra – switch handovers.
course of a connection.
The remainder of this paper is organized as follows: In
section 2, various handoff schemes are presented together
with their features, advantages and disadvantages. In section
3, performance issues are discussed. In section 4, the per-
formance analysis for the various handoff schemes is given.
Fixed ATM Network MES
In section 5 the evaluation platform is discussed. Finally,
section 6 concludes this paper.
2. HANDOFF REROUTING SCHEMES
There are two kinds of handoffs, namely intra-switch and
inter-switch handoff. In intra-switch handoff, the resulting
Mobile Terminal Wireless Link
route is optimal but in inter-switch handoff, the new con-
nection needs to be rerouted .
Handoff Schemes for rerouting can basically be classified
into four main groups. The full, incremental, multipath and
Fig. 1. Wireless ATM Network architecture
path extension re-establishment schemes.
The full re-establishment (FR) requires a completely new
The authors would like to thank Telkom SA, Siemens, National Research Foundation
path to be setup during handover . The incremental re-
(NRF) and The Department of Trade and Industry (DTI) for supporting this research
establishment (PR) scheme requires only that a partial new
path be setup and allows circuits to be re-used . This
scheme find the Cross Over Switch (COS) that is a cross All rerouting schemes have their advantages and disadvan-
point of the old path and the new path. The Multicast re- tages, but all are based on the following steps:
establishment (MR) scheme makes use of multicast in ATM
Networks , . A controlling switch establishes a con- Select a handoff switch. (Not used in path extension
nection to the current serving AP and all AP’s in the neigh- scheme), Setup a new sub-path between a handoff
borhood of the serving AP. When the MT moves to one of switch and a new AP. (Not necessary in multicast
the neighboring AP’s, data is immediately available. scheme), Release old sub-path between a handoff
Handoff is fast, but there is wastage of resources and could switch and old AP.
result in call blocking for other MTs attempting to connect
to the neighboring APs. In the path extension re- TABLE 2
NETWORK PARAMETERS FOR HANDOVER SCHEMES
establishment (PE) scheme the route is extended from the
original connection at the old Mobility Enhanced Switch Value
(MES) to the destination connection at the new MES, where
Number of hops from the
the new Access Point (AP) exist . After several handoffs, i) New AP to the COS Hnew
path looping occurs which results in a non-optimized path. ii) Old AP to the COS Hold variable
Hybrid Rerouting schemes have also been proposed that is a iii) New AP to the old AP Hctrl
combination of the connection extension and path rerouting through Signaling Channel
Bandwidth of the wired backbone network BWw 155Mb/s
schemes . This technique is superior to the other handoff Bandwidth of wireless link BWwl 2Mb/s
schemes because path extension reduces handoff delay and Latency of the wireless link Lwl 3ms
path (partial) rerouting increases resource utilization. The Latency of the wired backbone Lw 500µs
various schemes can also make use of radio hints (ER) in Protocol processing time for signaling mes-
sages PTfix 0.3ms
order to pre-establish connections before the old connections Protocol processing time to find the COS PTcos 2ms
are released, when the signal strength to the old AP becomes Time for a MT to acquire a wireless channel
weak . to a base station Tacq 2ms
Some of the features of the re-routing schemes are given Upper bound on the size of a signaling mes-
sage (bytes) Ssig 50
TABLE 1 Max size of a data packet (bytes) Sdata 53
SOME FEATURES OF HANDOVER SCHEMES Protocol processing time where admission
control is to be performed, in excess of fixed
Handover Schemes protocol processing time. PTadm 2ms
Path Easy to implement
Extension Path is extended for every handoff
No end point determination process re-
3. PERFORMANCE ISSUES
Handoff delay is the actual setup of the The service disruption time, time taken to complete a
path extension handoff, extra buffering needed to avoid data loss due to
Path looping occurs after several
rerouting and extra bandwidth resources required are im-
portant performance issues to be considered as they affect
the quality of service of the connections.
Results in inefficiency The five rerouting schemes in table 1 have been analyzed
Full Connection Re- Most optimal and simplest scheme by considering the signaling messages required to be ex-
routing Creates completely new route for each changed during handoff in a common network structure ,
. The performance measurements considered are (i)
Service disruption time, (ii) Handoff completion time, (iii)
Higher handoff delay
Buffer requirements at MT, (iv) Buffer requirements at AP
Maintains cell sequence through cell and (v) The excess bandwidth resources required during
Partial Rerouting Only a portion of the path is rerouted
Optimized path 4. PERFORMANCE ANALYSIS
Maintains cell sequence through cell
The parameters we intend to evaluate in this work is
forwarding based on the analytical and simulation analysis in  and
Multicast Rerouting Multicast data to all neighboring AP’s . The Service disruption time, Handoff completion time,
Faster handoff Buffer requirements at MT, Buffer requirements at AP and
Waste of resources the excess bandwidth resources required during handoff will
be obtained from the evaluation platform and is based on the
Hybrid Technique Path extensions and path rerouting ap-
following analytical analysis.
plied The parameters used in the analytical analysis are given
Lower handoff delay in table 2. It is assumed that the maximum throughput for a
Lower blocking connection is limited by the throughput of the wireless con-
nection. The analysis also assumes perfect delivery of sig-
naling messages. We compute and measure the time taken
in the Table 1 . for each of the signaling messages to be transmitted, for-
warded, and processed where necessary. Based on the time
Message (7) is the last packet indication in the wireless
S e rv e r link from AP1 to the MT. The time delay is denoted by T7.
Message (8) is the handoff completion and connection
release message from the MT to the old AP. Denote this
SW E time delay by T8.
Message (9) is the handoff completion and connection
release message from the old AP to the COS hop by hop.
SW D The time delay is denoted by T9.
Immediately after processing the last packet indication
message (7), the MT acquires a channel to the new AP,
SW C 5
AP2. This time delay is denoted by T10.
4 Message (11) is the acknowledgement from AP2 to the
2 MT. The time delay is denoted by T11.
Based on the time taken by these control messages, vari-
SW A SW B
ous performance measurements can be derived. Expressions
for five measures are given in  and . The disruption
time, Tdisrupt, is the time interval between the instant the
AP1 AP2 handoff completion command is received and the instant the
first data packet is received by the MT in the new path. This
10 includes the time to process all the signaling messages at the
switches and the transmission time of the first data packet.
The handoff completion time Tcomplete is the amount of
Fig. 2. The Partial Re-establishment Scheme with hints. time for all the rerouting to complete, i.e. from the time
when the MT issues a handoff initiation request to the old
taken by these signaling messages various performance AP to the time at which the connections to the previous AP
measures will be obtained. Some of the parameters used are is torn down or the connection established to the new AP.
hardware and network specific (Table 2). All events except for the acknowledgement transmissions
The handover schemes studied in this paper can be ana- occur sequentially. Thus, the completion time is the sum of
lyzed by considering the signaling messages required to be the times taken for each of the events during rerouting.
exchanged during handoff in a common network structure The amount of buffering required by the MT, BUFMT, is
. This is illustrated in Fig. 2 for the Partial Re- determined by the amount of time during which the MT
establishment scheme with radio hints, and can similarly be cannot transmit data on the wireless link. This includes the
done for the other handover schemes. time for the MT to greet the new AP to acquire a channel
When the MT detects the handoff occurrence in advance and the time for the new AP to acknowledge the greeting.
by the receive signal dropping below a set value, it sends a The amount of buffering required on the new AP for the
handoff request (1) to the current serving AP, AP1. The time downlink, BUFAP, down, is determined by the amount of data
needed for this message, T1, is the sum of the transmission that is transmitted in the new path before the MT is con-
time of a signaling message, the propagation time on the nected to the new AP.
wireless link, and the fixed protocol processing time on The excess bandwidth resources, Rexcess, used for a single
AP1. conversation during handoff is the amount of highest band-
After this AP1 communicates to AP2 through message width-space-time product among the two simultaneous
(2) the list of connections about to be re-established. Its total paths. The bandwidth space-time product is the product of
time is the end-to-end delay through the control channel the bandwidth provided to the connection, the length of the
between AP1 and AP2 plus the fixed protocol processing connection and the amount of time that the connection is in
time at AP2. This time delay is denoted by T2. existence [2, 8].
Immediately after sending message (2), AP1 begins the For evaluation of the analytical model it is assumed that
COS location algorithm process (3) to locate the COS be-
tween AP1 and AP2. The time taken to locate the COS is Hnew = Hold = (refer to table 2) for all the schemes
equal to the transmission and propagation time along each 2
hop, plus the fixed signaling processing time in each switch except the PE scheme. The values for Tdisrupt, Tcomplete,
along the path from AP1 to the COS. Denote this time delay BUFMT, BUFAP, and Rexcess depend on the specific hardware
by T3. being used and also the number of hops being considered.
Message (4) is a partial channel establishment from the Since the same hardware will be use to evaluate the various
COS to the new AP, i.e. AP2. Admission control tests are handoff schemes, comparisons can be made. Fig. 3, 4, 5 and
carried out at each switch along each hop, plus the fixed 6 show the performance parameters of the rerouting
protocol processing time at each switch along the path. De- schemes as obtained from the analytical analysis in graphi-
note this time delay by T4. cal form.
Message (5) is the acknowledgement of the partial chan- The multicast scheme and schemes that makes use of
nel establishment from AP2 to the COS. This signaling mes- radio hints have low service disruption time, whereas for the
sage retraces the path of the partial establishment, hop by partial, full and path extension schemes it increases with the
hop. Denote this time delay by T5. increase in the number of hops from the old/new AP to the
After the acknowledgement (5) the COS sends the last COS or between the old and new AP. The reason for this is
packet indication message to the old AP, AP1 hop by hop that for the MR and the PE schemes the handoff latency will
along the old path. This time delay is denoted by T6. not be affected due to multiple ATM switch handoff, since
the latency in the MR and PE schemes is determined only in at the new AP to buffer data. Hence, smaller buffer require-
the last hop between the MES and the MT. ments at the AP. This relationship can be seen by comparing
figures 3 and 5 with the exception of the ER scheme. The
reason for the exception is that depending on how far in
advance a hint is available before the MT moves, the extra
buffering required to complete the handoff may be signifi-
Service disruption time (Sec)
FR, PR cantly higher.
Buffer Requirements at AP (bits)
0.005 30000 FR, PR ER
1 2 3 4 5 6 7 8
Number of hops from the AP to the COS
Fig. 3. Service disruption time for different handover schemes 15000
Handoff completion time for the path extension method is
lower than any other scheme; a reason for this is given in the 1 2 3 4 5 6 7 8
previous paragraph. The scheme with hints has handoff Number of hops fromthe AP to the COS
completion time low compared to the full, partial and multi- Fig.5. Buffer requirements at AP for downlink during handover
cast schemes. The reason for this result is that in the FR
scheme a completely new path is setup from the MT to the Extra bandwidth resources required are the most for full
FH, hence the new connection setup time is directly propor- reconnection scheme and relatively constant compared to
tional to the number of switches between the MT and the the other schemes. The reason for this is that from the MT to
FH. The same can be said for the PR scheme, the only dif- the server (fig. 2) a new connection is setup and resources
ference being that the new connection setup time is propor- are used in each connection. Extra bandwidth resources are
tional to the number of switches between the COS and the not required for the path extension scheme since only the
MT. last hop is involved in the handoff process. The excess
bandwidth resources required for the MR scheme depends
on the number of simultaneous paths setup for multicast
purpose during handover. For the PR scheme the extra
Handoff completion time (Sec)
bandwidth resources required depend on the number of hops
between the COS and the MT and could be closely matched
ER to the FR scheme depending on the number of hops between
PR the COS and the APs.
Excess resources (bit links)
1 2 3 4 5 6 7 8
Number of hops from the AP to the COS
Fig. 4. Handoff completion time for different handoff schemes
The buffer requirements at mobile terminal for uplink 400000
data during handoff are the same for all schemes. This is 300000
because the amount of buffering required by the MT, 200000
BUFMT, is determined by the amount of time during which 100000
the MT cannot transmit data on the wireless link. This in- 0
1 2 3 4 5 6 7 8
cludes the time for the MT to greet the new AP to acquire a Number of hops from the AP to the COS
channel and the time for the new AP to acknowledge the Fig. 6. Excess bandwidth resources required during handoff
greeting. The message sequence is the same for all schemes
(messages 10 and 11, fig 2). The objective of the paper is to verify on an experimental
The buffer requirements at AP for the downlink are low- testbed the impact of the handoff schemes in terms of the
est for the MR and PE schemes. The reason for this behavior performance measurements mentioned above and compare
is due to the fact that the amount of buffering required on the results with those obtained analytically.
the new AP for the downlink, BUFAP, down, is determined by
the amount of data that is transmitted in the new path before 5. EVALUATION PLATFORM
the MT is connected to the new AP. This is directly related
to the service disruption time, since the service disruption The testbed consists of hardware and software compo-
time is the time interval between the instant the handoff nents and tools to monitor the handover operation. The test-
completion command is received and the instant the first bed would consist of at least two AP’s, a fixed host, a net-
data packet is received by the MT in the new path. Hence work node (WUGS) and a node controller [4, 14]. As seen
the lower the service disruption time, the less time is needed on Fig. 7, the network node consists of the Washington
University Gigabit Switch (WUGS)  intended for ex- However, packet losses using the wireless configuration
perimental research. The switching and control functionality are slightly higher. On further investigation, it was found
are separated. The Node Controller (NC) allows remote that the reason for the higher packet losses in the wireless
control, the modification of the signaling software for mo- link was due to occasional burst losses suffered in the wire-
bility purposes and a flexible environment for development less medium. It is for this reason that our choice of using
and testing. Ethernet is revalidated to eliminate the wireless specific
effects and concentrate on the losses due to mobility and
connection rerouting in isolation.
The protocol stack at the fixed host and the MT are dif-
ferent (Fig. 8). The fixed host uses AAL5 over ATM
whereas the MT uses UDP/IP over Ethernet. The AP’s per-
form the bridging function between the different protocol
stacks. In an end-to-end ATM system, the mobile host
NC would be responsible for reassembling the ATM cells.
At first the scenario with a single switch between the FH
and the AP’s will be considered. Later a more complex sce-
nario will be considered where one more switch can be lo-
cated between the fixed host and the MT, Fig 8, in order to
demonstrate the path extension scheme.
Points AP1 AP2 At this stage it is worthwhile mentioning why the use of
one or even two ATM switches in the experimental setup is
sufficient in evaluating the performance of the various
Fig. 7. Logical Testbed handoff schemes.
The effect of multiple switches on packet loss can be ex-
The evaluation platform for this work consists of the fol- plained as follows with regards to the various handoff
lowing hardware configuration. Five end stations, three schemes. The period during which the translation tables in
serving as AP’s, one as the fixed host and the last one serv- the COS are modified in order to reroute the connections,
ing as the Node Controller (NC), are connected via multi- has the biggest impact on packet loss, . Even when the
mode fibre to the WUGS switch ports. The two WUGS connections span multiple ATM switches, the single COS at
switches are also connected via multimode fibres. The Mo- which the translation tables are altered will be mainly re-
bile Terminal is connected to the AP’s via Ethernet to emu- sponsible for packet loss.
late the wireless access portion of the network. The reason for this is that the handoff schemes first up-
There are no wireless ATM cards available for experi- date all the switches in the new route and only then alter the
mentation. However, this is not an obstacle, as in this inves- entry at the COS (Section 4). Hence, it does not matter
tigation we are primarily interested in the effect of handover where in the connection the COS is located and the losses
on the backbone ATM network portion of the connection. suffered due to multiple switch rerouting will be similar to
Hence, experiments can be conducted with the Mobile Ter- our one hop experimental setup.
minal connected to the AP’s via Ethernet, Fig 8. This would The effect of multiple switches on handoff latency can be
allow us to focus on the consequences of connection re- explained as follows with regards to the various handoff
routing: loss, duplication, and reordering of packets. schemes. It can be noted that for the MR and the PE
Past experiments were performed to evaluate the per- schemes the handoff latency will not be affected due to
formance of handoff schemes using both Wavelan wireless multiple ATM switch handoff, since the latency in the MR
Ethernet cards and conventional wired Ethernet, . It was and PE schemes is determined only in the last hop between
found that the results of the performance evaluation are the MES and the MT. This can also be seen in fig. 3 where
similar using both configurations. the service disruption time is relatively constant with the
increase in the number of hops for the MR and PE schemes.
In the FR scheme a completely new path is setup from the
MT to the FH, hence the new connection setup time is di-
rectly proportional to the number of switches between the
MT and the FH (Section 4). The same can be said for the PR
scheme, the only difference being that the new connection
7x 8x 9x 10x 11x 12x 7x 8x 9x 10x 11x 12x 7x 8x 9x 10x 11x 12x 7x 8x 9x 10x 11x 12x
setup time is proportional to the number of switches be-
7 8 91011 12
7 8 9 1011
A 12 3456 A 12345 6
tween the COS and the MT. Hence, meaningful results can
1x 2x 3x 4x 5x 6x 1x 2x 3x 4x 5x 6x 1x 2x 3x 4x 5x 6x 1x 2x 3x 4x 5x 6x
A B A B
WUGS 1 WUGS 2
be obtained for the various performance parameters regard-
AP 1 AP 2 AP 3 less of the number hops involved in the handover procedure.
Bridge Bridge Bridge
The signaling protocols used in this experimental hando-
Physical Physical Physical Physical Physical Physical ver investigation are part of the WUGS environment, Fig. 9.
The Node Controller (Node/Switch Management) pres-
ents the same switch control API as the switch controller. It
allows several switches to be treated as one switch, i.e. the
node controller hides the details of how the individual
switches are connected together.
Fig.8. Experimental Testbed
The Connection Management Layer implements a dis- ment. Its signaling software offers a direct many-to-many
tributed control for a network with multiple nodes. Connec- communication and switching with connections that can be
tion Managers communicate with each other using a general modified dynamically.
signaling protocol. The authors expect to report results of the evaluation
The Call Management Layer implements a session ab- based on the experimental setup to verify the results ob-
straction (MCALL) used by signaling clients to request and tained analytically , , . Results will be presented at
manage connections in an ATM network. It allows general, SATNAC 2003.
dynamic, multi-connection, multicast sessions. The sessions
are supported through the UNI signaling protocol, Connec- 7. REFERENCES
tion Management Access Protocol, CMAP .  IEEE Journal on Selected Areas in Communications, Special Issue on
Wireless ATM, 1997.
Call Management  K. Keeton, B. Mah, S. Seshan, R. Katz and D. Ferari. Providing Con-
nection-Oriented Network Services to Mobile Hosts. August 1993.
 R. Yuan, S Biswas and D. Raychaudhuri. A Signaling and Control
Architecture for Mobility Support in Wireless ATM Networks, 1997.
 A. Festag, A. Wolisz, MOMBASA: Mobility Support – A Mulitcast-
Node/Switch Management based Approach. Proceedings of European 2000, Sept 2000.
 D. Kim and C. Toh. Mobile Multicasting in Wireless ATM Networks.
Mobile Networks and Applications 5 (2000) 103-110.
 J. Porter, D. Gilmurray, A. Massarella, and J. Naylon, Wireless ATM
handover requirements and issues. http://www.cam-orl.co.uk/radio/atm97-
Physical Network 0153.html.
Fig. 9. Signaling protocols in the WUGS environment  W.S.V. Wong, H.C.B. Chan, and V.C.M. Leung, Performance Evalua-
tions of Path Optimization Schemes for Inter-Switch Handoffs in Wireless
When a signaling client opens a MCALL, the connections
(could be several) are set up. The client manages the ATM Networks, July 2000.
MCALL and is designated the owner. Additional clients can  P. Marichamy, S. Chakrabarti and S.L Maskara. Performance Issues in
be added to the MCALL on request of the owner. Once a Rerouting for Handoff in Wireless ATM, 2000.
MCALL is opened, connections can be added, modified and  WUGS – Washington University Gigabit Switch, URL:
dropped to/from the MCALL. MCALL features can be set
according to the user needs. Signaling clients can be imple-
mented to perform rerouting of connections when handover  K. Cox, et al: Connection Management Access Protocol (CMAP)
occurs. Specification. URL:http://www.arl.wustl.edu/gigabitkits/kits.html
As already mentioned the performance measurements to  J. De Hart. Connection Management Software (CMSS) Architecture.
be considered with the experimental setup are (i) Service URL:http://www.arl.wustl.edu/gigabitkits/kits.html
disruption time, (ii) Handoff completion time, (iii) Buffer
 A. Acompora and M. Naghshineh, An Architecture and methodology
requirements at MT, (iv) Buffer requirements at AP and (v)
The excess bandwidth resources required during handoff. for mobile-executed handoff in cellular ATM networks, IEEE Journal on
It is intended to perform experiments with audio and low Selected Areas in Communications 12(8) (October 1994) 1365 – 1375.
bit rate video, in order to quantify the service disruption due  C. Chiasserini and R. Lo Cigno. Handover in Wireless ATM Net-
to the handover. Audio traffic can be characterized by its works: In-Band Signaling Protocols and Performance Analysis. IEEE
timing component. Every T milliseconds, a 160-byte packet
Transactions on Wireless Communications, Vol. 1, No. 1, January 2002.
will be transmitted by the source. For T = 20, this corre-
sponds to telephone quality audio of 64 kb/s. In a typical  A. Festag, L. Westerhoff and A. Wolisz. The MOMBASA Software
audio conversation, there will be silence periods during Environment – A Toolkit for Performance Evaluation of Multicast-Based
which no packets will be transmitted. If this occurs during a Mobility Support. Proc. of Performance Tools 2002, London, UK, April
handoff, there would be no disruption of traffic. It is for this 2002.
reason that we will make use of continuous traffic, so that
 R. Ramjee, T. La Porta, J. Kurose, D. Towsley. Performance Evalua-
we can detect all potential disruptions during a handoff.
The Video traffic can be characterized by a Constant Bit tion of Connection Rerouting Schemes for ATM-based Wireless Networks.
Rate source sending packets every 30 ms. This would corre- IEEE/ACM Transactions on Networking, 1998.
spond to a frame rate of approximately 30 frames per sec-
ond, the frame rate for full motion video.
In this paper, an evaluation platform has been presented
to study the handover schemes, and its components have
been fully justified to serve as a tool for the performance
evaluation of the various handoff schemes.
For experimental purposes a testbed will be setup, which
is based on an open, non-proprietary, networking environ-