J-SAC-2.3.7.doc.doc

Abstract— Supporting transmission of stream media over wireless mobile networks is often difficult

because packets may be lost due to the rerouting of packets during handoff, and also because bursts of

packet loss may occur during handoff due to the disparity in the amount of available bandwidth among

different cells. In this paper, we propose an end-to-end multi-path handoff scheme that provides smooth

handoff for stream media in wireless networks with different bandwidth from cell to cell. In the proposed

scheme, multiple paths are established during handoff to reach a mobile destination node. The stream

media sources are equipped with an adaptive multi-layer encoder, and important layers in the encode

video stream are duplicated and transmitted over multiple paths during handoff. The effectiveness of the

proposed multi-path handoff scheme is verified and compared with existing schemes through extensive

simulations. The simulation results show that the proposed scheme provides higher throughput and better

quality for stream media.

Index Terms— Wireless networks, handoff, stream media, multi-layer video encoder, slow start,

congestion.

Manuscript received on Feb 14, 2003, revised on Sept 30, 2003. This work was supported by the National Science Foundation through grants ANI-0083074 and ANI-9903427, by DARPA through Grant MDA972-99-1-0007, by AFOSR through Grant MURI F49620-00-1-0330, and by grants from the University of California MICRO Program, California State University of Northridge Research and Sponsored Projects, Hitachi, Hitachi America, Novell,Nippon Telegraph and Telephone Corporation (NTT), NTT Docomo, Fujitsu, Korea Research Foundation through BK21, and NS-Solutions. This work has been presented in part at WMASH’03, San Diego, CA, Sept. 2003.

*Yi Pan and Tatsuya Suda are with the School of Information and Computer Science, University of California, Irvine, CA 92697 USA (e-mail: [email protected]; [email protected]. Phone: +1-949-824-4105. Fax: +1-949-824-2886)

Meejeong Lee is with the Dept. of Computer Science and Engineering, Ewha Woman’s University, 11-1 Daehyun-Dong, Seoul, Korea (120-750) (e-mail: [email protected]. Phone: +822-3277-2388).

Jaime B. Kim is with the Computer Science Department, California State University, Northridge, CA 91330 USA (e-mail: [email protected], phone: +1 818-677-3892).

*Yi Pan is the correspondence author.

An End-to-End Multi-Path Smooth Handoff Scheme for Stream Media

1

I. INTRODUCTION

Applications using stream media are becoming popular in wireless mobile networks. Providing smooth

handoffs for stream media without any transmission disruption and drastic quality degradation is a

challenging problem due to the following two factors: 1) rerouting of packets during handoff may result in

burst loss of packets (referred to as rerouting packet loss in this paper), and 2) disparity of available

bandwidths in cells may also result in bursts of packet loss during handoff (referred to as congestion

packet loss in this paper). In wireless mobile networks, when the data path is disconnected and the change

of the point-of-attachment to the network is required as a result of user movement, packets on the path

through the old point-of-attachment may be lost, degrading the quality of service for the stream media. In

addition, the amount of bandwidth available at the new point-of-attachment may be smaller than the

previous one, and the mobile node may experience congestion. This bandwidth disparity problem is

exacerbated as different wireless access techniques with difference link speeds, such as WLAN,

Bluetooth, and W-CDMA, are being deployed to provide ubiquitous connectivity. This mismatch in

available bandwidths may result in bursts of packet loss, if the stream media transmission continues

without appropriate rate adjustment in a new cell.

A number of mobility management techniques have been proposed in the literatures [1, 2, 3, 4, 5, 6, 7,

8, 9, 10, 11, 12, 13, 14, 15, 16]. Many of them address the smooth handoff of stream media in a QoS

enabled network [1, 2, 3, 4, 5], such as wireless ATM, but very few techniques have been proposed to

provide smooth handoff for stream media in a highly heterogeneous best effort network, such as the

Internet.

In this paper, we propose a novel scheme for smooth handoff of stream media in heterogeneous best-

effort wireless mobile networks. In order to avoid the drastic quality degradation, the proposed scheme

reduces the rerouting and congestion packet loss during handoff (1) by establishing multiple paths to the

mobile node and transmitting duplicate packets over the multiple paths to reduce the negative impact of

2

packet loss and (2) by probing the available bandwidth in the new cell and allowing stream media

applications to gradually adapt their transmission rates to the available bandwidth in a new cell.

The proposed scheme assumes that the stream media sources employ an adaptive multi-layer video

encoder (such as MPEG-1, 2 and 4 [17, 18]) that produces multiple encoded video streams (layers) of

differing importance from a video input. The adaptive multi-layer encoder adjusts the number of video

layers and the corresponding encoding rate of each layer based on the input from the lower layer

protocols. The proposed scheme also exploits the fact that cells in wireless networks overlap with each

other. In the proposed scheme, as a mobile node (receiving a stream media) moves from an old cell to a

new cell and enters a cell overlapping area, paths from the mobile node to the stream media source

through the new cell are established, while maintaining the already existing paths through the old cell.

With the proposed scheme, more important (video encoded) layers of the stream media are transmitted

redundantly through more number of paths, increasing the robustness to packet loss during handoff, while

the higher available bandwidths on some of the paths are exploited to transmit enhancement layers for

higher quality. For instance, the base layer of the stream media with high importance are replicated and

transmitted over all new and existing paths during handoff. In the proposed scheme, separate end-to-end

rate control is applied onto each separate path in order to adjust to a different amount of available

bandwidth on each path. The number of layers and encoding rate of each layer at the stream media

source are adjusted accordingly.

The overhead of the proposed scheme includes bandwidth required for the redundant transmission of

important layers of the stream media during handoff and computation of encoding rates at the stream

media source. The performance of the proposed scheme is investigated through extensive simulations.

Obtained performance measures include QoS related metrics for the stream media (such as throughput and

packet loss ratio), overheads of the proposed scheme such as reduced transmission efficiency due to

redundant transmission, and the scalability with respect to the number of paths used during handoff.

3

The paper is organized as follows. Section II surveys related work. Section III describes the proposed

multi-path smooth handoff scheme in detail. In Section IV, simulation models are described, and in

Section V, numerical results are presented to investigate the performance of the proposed scheme.

Concluding remarks and possible future work are described in Section VI.

II. RELATED WORK

In the literatures, different mobility management techniques have been proposed for QoS enabled

networks (e.g., wireless ATM and IntServ supported networks) and best effort networks (e.g., the

Internet).

Many existing mobility management techniques proposed for QoS enabled networks use resource

reservation protocols to reserve some amount of bandwidth on the new data path for mobile nodes. These

techniques [1, 2, 3, 4, 5] all require extensive QoS support in the network to keep soft states of resource

reservation for each active flow at network devices. Given that end-to-end QoS support is hardly

available in a highly heterogeneous, large-scale network as the Internet, these techniques may not be

applicable to such networks.

There are few mobility techniques proposed for stream media in the Internet. Existing mobility

management techniques proposed for the Internet may be classified into two categories: network layer

mobility management techniques [6, 7, 8, 9, 10, 11, 12] and transport layer mobility management

techniques [13, 14, 15, 16]. None of these existing technologies, however, considers the requirements of

the stream media transmission specifically. For instance, most of the existing network layer mobility

management techniques only remove rerouting packet loss during handoff, while the bandwidth disparity

among the cells is not considered. Thus, with these schemes, congestion loss may still occur in the new

cell during handoff. Most of the existing transport layer mobility management techniques are proposed for

window-based congestion control and use retransmission to recover packet loss. The transmission rate of

the stream media is adjusted to the available bandwidth in a new cell either through a slow start or a

4

multiplicative decreasing procedure. This rate adjustment process usually is slow and causes rate

disruption or significant packet loss. Because of this downside, together with the high fluctuation of the

transmission rate caused by oscillation of the window size in a window-based congestion control

algorithm, existing transport layer mobility management techniques based on TCP extensions may not be

directly applied to stream media.

Recently, some researchers investigated the stream media transmission in best effort CDNs (Content

Distribution Networks) and proposed alternative schemes where mobility support is provided at the

session layer. A common design of stream CDNs includes an overlay network of proxies deployed at the

edges of a network and a data center to distribute the contents and redirect the user requests [20, 21, 22].

Requested streams are cached at local proxies. The transmission rate of the stream media is adjusted at the

proxies through filtering or transcoding the incoming streams based on the locally available resource. In

these schemes, mobility is handled at a session level. A mobile node moving into a new cell is redirected

to a new proxy, and a new session is established between the mobile node and the server through the new

proxy. Then, a mobile node starts receiving packets from the new proxy [22]. The deployment of an

overlay network of proxies can be excessive and may only be justified for popular contents. Session set-up

and rate adjustment (e.g., transcoding) delays during handoff may also cause interruptions in the

transmission of stream media.

The scheme proposed in this paper maintains multiple paths between the sender of the stream media and

the receiver (i.e., a mobile node) during handoff. Recently, some new transport layer protocols that

maintain multiple connections between the sender and the receiver have been proposed in the literatures.

They are SCTP (Stream Control Transport Protocol) [23], multi-path TCP [24], and p-TCP [25]. Unlike

the proposed scheme that prevents packet loss through redundant transmission and applies rate-based

congestion control, all existing schemes are mainly proposed for reliable data transmission and have one

or both of the following drawbacks for the stream media: 1) packet loss is recovered through

5

retransmission, and it introduces unpredictable delays in packet delivery and may break the real-time

requirements of the stream media transmission, 2) window-based congestion control is used, and it causes

high fluctuation in transmission rate and may not achieve smooth rate adjustment for stream media.

III. SCHEME DESCRIPTION

In this section, the proposed multi-path handoff scheme for stream media is described in detail. In

Section III. A, overall system architecture of the proposed scheme is presented, and detailed descriptions

on architectural components are given in subsections III. B through III E. Overhead of the proposed

scheme is discussed in subsection III F.

A. System Architecture

The proposed scheme acquires multiple paths from the sender of the stream media to the receiver (i.e., a

mobile node) during handoff, estimates the available bandwidth on each path through an end-to-end rate

control algorithm, calculates the number of video layers and the target encoding rate of each layer, and

assigns different layers to different paths for transmission.

In the proposed scheme, new paths are established as soon as a mobile node enters an overlapping area,

before the mobile node reaches the middle point of the overlapping area, unlike in many existing handoff

schemes, making it possible to start using the new paths earlier.

The proposed scheme inserts four components in the transport protocol layer at the sender and the

receiver: a path management module at both the sender and the receiver, a multi-path distributor module

at the sender, a pair of rate control modules at both ends for each path (at the sender and at the receiver),

and a multi-path collector module at the receiver. Figure 1 illustrates the overall system architecture of the

proposed scheme.

The path management module at each end of the transport protocol manages the currently available

paths, using COA binding update messages in the mobile IP protocol [6]. (COA binding update messages

indicate the existence of multiple paths during handoff.) The path management modules report the

6

existence of multiple paths to the multi-path distributor (at the sender) and the multi-path collector (at the

receiver). . The path management module also assigns a rate control module to a new path and removes a

rate control module from an old path during handoff. The rate control modules for each path perform end-

to-end feedback-based rate control. This end-to-end feedback-based rate control includes an initial

probing phase using a slow start algorithm to estimate the available bandwidth on a new path. Based on

the estimated available bandwidth on a new path provided by the rate control modules and the notification

of existence of multiple paths from the path management module at the sender, the multi-path distributor

module calculates and reports the number of video layers and the target encoding rate for each layer to the

multi-layer encoder at the sender. The multi-path distributor module also assigns different video layers to

appropriate paths based on the differing importance of video layers. Note that the multi-path distributor

may assign the same video layer to multiple paths, depending on the importance of the video layer. The

multi-path collector module at the receiver accepts incoming packets from multiple paths, filters and

reorders them before passing them to the decoder.

B. Path Management

During handoff, to allow the path management module at the sender to maintain multiple paths

simultaneously, the mobile IP simultaneous binding [6] and route optimization [7] options are used. The

simultaneous binding option allows the receiver (i.e., a mobile node) to simultaneously register multiple

COAs, and the route optimization option allows the sender to be informed of the current COA

registrations.

COA binding update messages report the existence or loss of a COA. COAs that a mobile node (i.e., the

receiver) is registered with is used to identify a path from the sender to the receiver. When a new COA is

reported, the path management module assigns a rate control module to the new path and notifies the local

multi-path distributor/collector of the new path; when a loss of a COA is reported, the path management

module removes a rate control module and notifies the local multi-path distributor/collector of a loss of a

7

path.

C. Rate Control Module

A rate control module is installed at the sender and the receiver, immediately after a new path is notified

of by the path management module. We choose a TCP Friendly Rate Control (TFRC) rate control

algorithm [26] at the rate control modules to estimate the available bandwidth on a path. Rate control on a

path consists of two phases. In the first phase (called the probing phase), a slow start algorithm is used to

detect the available bandwidth on a new path. In the second phase (called the congestion avoidance

phase), once the initial estimation of the available bandwidth is done in the first phase, the available

bandwidth on the path is estimated through a predefined equation to avoid congestion.

The rate control module at the sender sets the transmission rate on each path at the estimated available

bandwidth of the path, and applies a token bucket algorithm to enforce that the actual transmission rate

conforms to the estimated available bandwidth. Note that in the congestion avoidance phase, base layer

packets are always allowed to be transmitted, while enhancement layer packets are subject to token bucket

control. However, since the estimation of the available bandwidth on the path in the probing phase may

not be accurate, all packets (including base layer packets) are subject to token bucket control to avoid

significant congestion loss on a new path. Packets that are in excess of the available bandwidth are simply

discarded. The rate control module at the receiver receives the packets and notifies the multi-path collector

of packet arrival on the corresponding path. The rate control module at the receiver also sends a packet

loss ratio report to the rate control module at the sender periodically with the time interval of the round

trip time (RTT) for the path.

D. Multi-path Distributor Module

In order to calculate the number of video layers and target encoding rates of different layers for the

multi-layer video encoder, the multi-path distributor module performs a VLPA (Video Layer-Path

Adaptation) algorithm to decide 1) the number of video layers and the target encoding rate for each layer

8

and 2) the paths to transmit the packets for each layer. In the following description, represents the

cumulative video stream rate from layer 0 (the base layer) to (and including) layer i, and (= )

represents the encoding rate of layer i.

VLPA algorithm decides the number of video layers based on the number of paths in the congestion

avoidance phase and the available bandwidths on the paths. Note that the VLPA algorithm only considers

the paths in the congestion avoidance phase, not in the probing phase, because the slow start algorithm in

the probing phase can potentially cause highly dynamic changes in estimation of the available bandwidth

on a path, and thus, in the number of video layers and the encoding rates to use. This may lead the system

unstable. The number of video layers is set to the number of distinct transmission rates on the paths in the

congestion avoidance phase. (For instance, if the transmission rates on the paths are 1Mbps, 2 Mbps,

2Mbps, and 3Mbps, the number of video layers is set to 3). The cumulative rate up to and including layer i

(ri) is set to the ith transmission rate among the paths in the congestion avoidance phase, sorted in

ascending order. (For instance, if the transmission rates on the paths are 1Mbps, 2 Mbps, 2Mbps, and

3Mbps, cumulative encoding rates are r0 = 1Mps, r1 = 2Mbps, and r2 = 3Mbps.) Targeted encoding rate for

each layer is calculated as the difference between the two consecutive cumulative rates ri and ri-1 (

). In an exceptional case when the path with the highest transmission rate is in the probing phase,

the VLPA algorithm also considers that new path in calculation of video layers. An additional video layer

is added and the cumulative rate of the entire stream is set to the current transmission rate on that new

path, allowing utilization of extra bandwidth on the new path. This video encoding rate calculation is

executed periodically, and the updated number of video layers and the encoding rates are reported to the

multi-layer video encoder. (In the simulations in Section IV, the time interval between two consecutive

video encoding rate calculations is set to be one video frame time (i.e. 40ms).)

The VLPA algorithm also determines the paths to transmit each layer through the following policy:

1. Video layer i is transmitted redundantly through all the paths whose transmission rates are greater

9

than or equal to cumulative rate .

2. If a path is in the probing phase (i.e., a new path), its transmission rate is not used to determine a

cumulative rate. As a result, the new path may have a transmission rate between two consecutive

cumulative rates and . In this case, both layers (layer i and i+1) are assigned to this path to

assure that the available bandwidth on this new path will be fully used in the probing phase.

Note that with the above path assignment policy, the base layer (i.e., layer 0) is assigned to all paths, and

the more important a video layer is, the more number of paths are assigned to that layer. A more detailed

algorithm can be found in [28].

E. Multi-path Collector Module

The proposed scheme transmits duplicate packets through multiple paths simultaneously, and thus, out-

of-order and/or duplicate arrivals of packets may occur. The multi-path collector module at the receiver is

responsible for buffering and reordering of packets from multiple paths. It also filters out the redundant

packets before delivering packets to the application.

For stream media applications, receiving the video stream on a real time basis is critical. Therefore,

when a packet is missing, the multi-path collector only waits for the packet for a predetermined time

interval. If the missing packet does not arrive within the time interval, the multi-path collector delivers its

buffer content to the application with the unfilled holes in packet sequence.

F. Overhead of proposed scheme

The overhead of the proposed scheme is in two folds: reduction in transmission efficiency due to

transmission of duplicated video packets and transmission of control packets associated with the proposed

scheme, and processing of the proposed scheme at the sender and receiver.

Reduction in transmission efficiency impacts the number of users supported in the network, and this

aspect is evaluated in detail in Section V.

Processing of the proposed scheme is mostly at the multi-path distributor module running the VLPA

10

algorithm and rate control module running the TFRC algorithm. The dominating computation in VLPA

algorithm is to assign paths to each video layer. This requires checking each path against each layer i and

takes O(m*n) steps, where m is the number of paths (available during handoff) and n is the number of

video layers. Since n is always less than or equal to m in the proposed scheme, the VLPA computation for

multi-path cases requires O(m2) more steps than for single path cases. The dominating computation in the

TFRC algorithm is to identify packet loss at the receiver and to compute the transmission rate at the

sender. Each computation takes the constant number of steps per RTT (the round trip time) for a given

path [26]. Therefore, computational cost for m paths is approximately O(m) times the computational cost

for a single path with the shortest RRT among all paths. The number of paths during handoff (m) is

usually very limited, and thus, the increase of processing in the proposed scheme is limited.

IV. SIMULATION MODEL

The performance of the proposed scheme is evaluated through extensive simulation using OPNET [27].

The purpose of the simulation is two-fold: first to investigate the performance of the proposed scheme

with various network parameters, and second to compare the proposed scheme with other existing

schemes.

Existing schemes chosen to compare against the proposed schme use a single path from the stream

media source to the mobile node (i.e., the receiver). With only one path reaching the mobile node during

handoff, stream media handoff may be performed in an end-to-end manner or using proxies. Hereinafter,

we will refer to them as a single-path scheme and a proxy-based scheme, respectively. In the single path

scheme, upon receiving a new COA binding update message from mobile IP during handoff, the stream

media sender stops sending video packets to the base station of the cell that the mobile node is leaving

(old base station) and starts sending video packets to the base station of the cell the mobile node is

entering (new base station). ***I am here.**** An adaptive multi-layer encoder and TFRC rate control

are applied at the sender in a single path scheme to provide a fair comparison with the proposed scheme.

11

In addition, another version of single path scheme using mobile IP fast forwarding option is simulated to

represent single path end-to-end handoff with network layer mobility management techniques. In this

version, the video packets on the fly toward the old base station will be forwarded to the new base station

during the handoff, in order to avoid re-routing loss. The above two versions of single path schemes will

be referred to as single path without forwarding scheme and single path with forwarding scheme in the

rest of the paper, respectively.

Proxy-based scheme models the stream media transmission in a generic CDN. In proxy-based scheme,

mobility is handled at a session level as we mentioned in stream CDN (Section II). We assume a perfect

scenario in which the stream media is always available in the buffer of the new proxy and the proxy

transcodes the stream media according to available bandwidth. The stream media transmission is resumed

by re-transmission of the last incomplete video frame through the new proxy in a handoff. Session set-up

and transcoding delay in the new proxy are ignored. For proxy-based scheme, the proxy typically exists

very close to the mobile node and has very short round trip time to the mobile node. Under such a

situation, TFRC is not applicable since TFRC rate control algorithm does not scale to very short round trip

time due to heavy overhead incurred by the feedback per round trip time. Therefore, we assume a

centralized resource controller will let the proxy know the available bandwidth in the new cell

immediately.

Our simulation experiments concentrate on analyzing a single handoff instance since a more general

scenario with a sequence of handoffs consists of individual handoffs with different parameter settings. In

most experiments, a handoff occurs between two overlapping cells since it is the most common case.

Scenarios with more than two overlapping cells are used to illustrate the growth of overhead with the

increasing number of paths used in handoff.

Figure 2 shows the network configuration used in our simulations. The wireless domain includes base

stations for two neighboring cells, physical links connecting base stations with the wireless gateway, and a

12

wireless gateway connecting the wireless domain to the wired backbone network. For the proxy-based

scheme, a proxy is assumed to be located at each base station. Wireless links are 802.11b WLAN 11Mbps

links with a bit error rate between 110-3 to 110-5 (average is 210-5). The coverage radius of the base

station for each cell is 300 meters, and the distance between two neighboring base stations is 500 meters.

In each cell, a background node is placed to simulate the background traffic. All wired links in the

simulation have 155Mbps link capacity, 110-12 bit error rate, and 10s of propagation delay as default

settings.

In order to evaluate the overhead of the proposed scheme with respect to the number of paths used

during a handoff, network configurations with multiple cells are used (See Figure 3.) All parameter values

are same as that assumed in Figure 2, except for the distance between the wireless base stations to allow

more number of overlapping cells. In one of the configurations (Figure 3 (a)), base stations are placed on

a grid, and the distance between the two adjacent base stations is assumed to be 500 meters. This allows a

maximum of 4 paths in the overlapping area. In Figure 3 (b), every three neighboring base stations form

a triangle with equal-length edges of 500 meters.

The video source traffic is generated using a source adaptive multi-layer encoder model presented in

[19]. The video packet size is fixed to 1024 bytes.

To simulate cells with different amount of available bandwidth in the wireless network, background

traffic is introduced. Background traffic in each cell is generated by a Poisson process, and it is

transmitted from the base stations to the background nodes.

To investigate the impact of various network parameters on the performance of the proposed scheme,

following network parameters are varied:

available bandwidth in the cell the mobile node is entering,

round trip time from the source to the destination

speed of mobile node movement, and

13

number of overlapping cells

We summarize the parameters we use in different sets of simulations in Table 1, 2, and 3 below. In the

tables, old cell refers to the cell that mobile node is leaving and new cell refers to the cell the mobile node

is entering. As shown in Table 2 and 3, we run the simulation in two cases when changing round trip time

and speed of mobile node movement. In Case I, the mobile node moves from a cell with lower available

bandwidth to a cell with higher available bandwidth. In Case II, the mobile node moves from a cell with

higher available bandwidth into a cell with lower available bandwidth, in which congestion happens.

In the simulation, the available bandwidth is varied by changing the mean of the Poisson background

traffic volume in the new cell. The round trip time is varied by tuning the propagation delay of the first

link from the Corresponding Node to Backbone Router in the simulation network model (Figure 2). The

speed of a mobile node is varied from 4.5mph to 140mph during handoff in the simulation. The trajectory

of a mobile node is simulated using the trajectory model provided in the OPNET simulator [27]. Number

of overlapping cells is varied from 2 to 4, depending on the configuration of wireless networks.

We measure the performance during the handoff period to show the difference of the proposed scheme

with other schemes. The handoff period is defined as the time period during which a mobile node stays in

the cell overlapping area.

The performance of handoff schemes is evaluated with respect to two aspects: the quality of service and

the overhead. To measure the quality of service of the video stream, throughput, packet loss ratio, and

goodput in terms of video frame rate per second are measured.

The overhead of the proposed scheme is measured as reduction in transmission efficiency and network

capacity. The transmission efficiency is defined as the ratio of the number of unique application packets

received to the total number of packets transmitted during the handoff period. The network capacity is

defined as the number of users supported in a fixed bandwidth multi-cell wireless network. A reduction

In the simulation, available bandwidth in the new cell is calculated as the link speed minus the volume of background traffic. The actual throughput, however, is much lower due to the overhead of the MAC layer and link errors.

14

ratio of network capacity using multi-path handoff is measured in a comparison to network capacity using

the proxy-based scheme.

V. NUMERICAL RESULTS

In this section, simulation results are presented to evaluate the performance of the proposed scheme. In

all the figures presented in this section, SP_NF denotes single-path handoff scheme with no forwarding;

SP_FF denotes single-path handoff scheme with forwarding; PROXY denotes proxy-based scheme;

MPATH denotes the proposed scheme; and MP_Base denotes the base layer performance of the proposed

scheme. Each result is an average of multiple simulation runs under the same set of parameters.

A. Quality of Service Metrics

1) Throughput

Figure 4 shows the throughput of the different handoff schemes as a function of the available bandwidth

in the new cell. For the proposed scheme, the throughput of the whole video stream is determined by the

maximum available bandwidth of the two neighboring cells, whereas the base layer throughput is

determined by the minimum of the two. Both the proxy-based scheme and the single path schemes show

lower throughput than the whole video stream throughput of the proposed scheme because those schemes

could only utilize the available bandwidth of a single path at a time. Among the proxy-based scheme and

the single path schemes, the latter show lower throughput when new path has higher available bandwidth

since the single path schemes are slow to acquire the higher available bandwidth in the new cell. Recall

the single path schemes depends on the congestion avoidance steps for rate increment whereas the proxy-

based scheme can almost immediately adjust the stream rate to the higher available bandwidth in the new

cell.

Figure 5 shows the throughput of the different handoff schemes as a function of the round trip time. It

illustrates the impact of propagation delay to the proposed scheme. As shown in Table 2, we present

results from two simulation cases: 1) Case I, the available bandwidth in the new cell is higher and 2) Case

15

II, the available bandwidth in the new cell is lower. Except for the proxy-based scheme, which does not

depend on the end-to-end feedback for stream rate adjustment, throughput decreases as the RTT increases.

This is mainly due to the rate control equation used in TFRC algorithm [26] to decide the steady state

throughput r: (1), where MSS and p represent the packet size and the loss rate,

respectively.

For the proposed scheme, throughput reduction for RTT increment is more significant in Case I. In Case

I, the transmission rate on the new path during the probing period is lower than the actual available

bandwidth in the new cell, and thus, as more time is spent for probing with increasing RTT, the

throughput during handoff becomes lower. As explained for Figure 4, since the proposed scheme adjust

the whole video stream rate to the maximum available bandwidth among all paths for the whole period of

handoff, throughput of the proposed scheme maintains the highest value among all compared schemes

except for the cases where RTT is greater than 100ms in Case I.

Figure 6 shows the throughput of the different handoff schemes as a function of speed of mobile node

movement. The parameter setting used in this figure is summarized in Table 3. Note the length of handoff

period is reduced as the speed of mobile node movement increases. Figure 6, therefore, shows the impact

of reduced length of handoff period on the throughput of the proposed scheme. With a reduced length of

handoff period and fixed RTT value, the number of round trips that TFRC can perform during the handoff

is reduced. In Case I, therefore, both the proposed scheme and single path schemes, which are all subject

to the TRFC rate control in obtaining the higher available bandwidth in the new cell, show throughput

reduction due to less rounds of rate increment executed during the handoff as the speed of mobile node

movement increases. In Case II, the throughput of the proposed scheme remains constant for increasing

speed of mobile node movement. Recall the whole video stream rate of the proposed scheme during

handoff is determined by the highest bandwidth available among all neighboring cells. Since the available

bandwidth in the old cell is higher in Case II, the performance variation related to the probing in the new

16

cell does not affect the throughput of whole video stream during handoff. The throughput of single path

schemes also remains almost constant in Case II. Since TFRC rate reduction progresses a lot faster than

rate increment, the impact of reduced handoff period is a lot smaller in Case II and the throughput changes

for increasing speed of mobile node movement is not significant compared to Case I. Similar to RTT, the

throughput of proxy-based scheme is not affected by speed of mobile node movement. Its rate adjustment

is not based on TRFC, and thus, the length of handoff period does not have impact on its performance. In

summary, it is shown that the proposed scheme achieves the highest throughput in almost all cases of

speed of mobile node movement.

2) Packet Loss Ratio

There are three types of packet loss during the handoff period: re-routing loss, congestion loss, and

probing loss. Packets on the fly to the old cell are lost when mobile node stops receiving packets from the

old cell, and we call this type of loss as re-routing loss. When a mobile node moves from high bandwidth

cell to low bandwidth cell, loss may occur if the stream transmission rate is not reduced promptly and this

type of loss is referred as congestion loss. Probing loss occurs if the slow start procedure is used to

estimate the rate on a new path. Loss of packets occurs during the last RTT of the slow start probing

phase, and the probing loss increases as RTT and available bandwidth on the probed path increase.

We measured the ratio of packet loss caused by each of the above three different sources. Figure 7

through 9 shows the packet loss ratio of different handoff schemes as a function of available bandwidth in

the new cell, RTT, and speed of mobile node movement respectively. The parameter setting shown in

Table 1, 2, and 3 are used for Figure 7, 8, and 9 respectively. Regardless of different parameter settings,

the packet loss ratio of base layer video in the proposed scheme is kept virtually 0 in all cases due to the

redundant transmission and per-path rate control. The proxy-based scheme also shows very low ratio with

respect to all types of loss in all cases because a very ideal environments are assumed for the proxy-based

scheme in our simulation, i.e., a local proxy that is ready with the content always exists in the new cell and

17

the transmission rate can be adjusted to the available bandwidth instantly. The packet loss ratio of the

whole video stream in the proposed scheme is very close to this optimal proxy-based scheme in most of

the cases. The main source of loss in the proposed scheme is probing. The probing loss occurs during the

last RTT period of slow start step, and the amount of probing loss increases as the available bandwidth in

the new cell and/or RTT increases. Including probing loss, speed of mobile node movement does not

affect the absolute amount of loss for any type of loss. As the speed of mobile node movement becomes

faster, though, the length of handoff becomes shorter and the total amount of packets transmitted during

handoff is reduced, which results in a higher packet loss ratio. The proposed scheme does not suffer re-

routing loss since the mobile node keeps receiving through both cells while it is transiting handoff area,

and the congestion loss is also kept very low due to its per-path rate control.

Single path schemes have the highest packet loss ratio in most of the cases. Both of them suffer

congestion loss, but no probing loss. The single path without forwarding suffers re-routing loss also, but

the single path with forwarding scheme removes most of the re-routing loss as it is expected to do. Due to

the difference in re-routing loss, the single path with forwarding always has a lower packet loss ratio. The

congestion loss is mainly affected by the degree of difference in available bandwidth of the cells. When

the new cell has smaller available bandwidth, the congestion loss starts to grow and it grows as the

bandwidth difference becomes larger. The length of RTT also affects the congestion loss according to the

following equation for the packet loss ratio p, which is derived from the TFRC rate control equation given

in Equation (1): (2). That is, the packet loss ratio is inversely

proportional to the square of RTT. Re-routing loss in the single path without forwarding increases as the

RTT increases because re-routing loss occurs while registration for the new path, which takes half of a

RTT, is progressing. Comparing the single path schemes and the proposed scheme, it is shown that the

amount of loss caused by probing in the proposed scheme is not as significant as congestion and re-

routing loss in the single path schemes.

18

3) Video Frame Rate

The video frame rate is defined as the average number of successfully received video frames per second,

and it is measured as goodput to illustrate the smoothness of video stream perceived by users. It is

assumed that the target frame rate is 25 frames per second.

Figure 10 shows the transient behavior of video frame rate during the handoff period. Two cases are

simulated, and the parameters set for these two cases is summarized in Table 4. In Case I, no congestion

occurs in the new cell since it has higher available bandwidth than the old cell. As a result, all four

schemes show relatively small fluctuations in the video frame rate. In Case II, on the other hand, the two

single path schemes show a drastic drop in the video frame rate due to congestion loss in the new cell. In

the single path schemes, the TFRC at the sources are not aware of the handoff and keep sending at a high

rate which was available in the old cell, and this results in congestion loss. The proposed scheme shows

smooth frame rate even in Case II due to proper rate control in the new cell and redundant transmission of

base layer packets. Since the session set up and the transcoding delays, which usually are incurred in a real

proxy handoff and may cause stream disruption, are ignored in our simulation, proxy-based scheme also

shows smooth performance in both cases.

B. Overhead Evaluation

1) Data Transmission Efficiency during handoff

The main cost paid to achieve higher performance in the proposed scheme is the bandwidth waste due to

redundant packet transmissions during handoff. Data transmission efficiency is defined as the ratio of the

number of unique application packets received to the total number of packets transmitted. For the

proposed scheme, the amount of redundant transmissions is mainly determined by the number of

overlapping cells and the available bandwidth in those cells, and thus, these two parameters are varied in

this simulation. In the proposed scheme using VLPA algorithm, the greatest reduction in transmission

efficiency occurs when all overlapping cells have the same amount of bandwidth. In this case, identical set

19

of packets (base layer packets) are transmitted through all neighboring cells. Therefore, to test the lower

bound of the transmission efficiency of the proposed scheme, we assume that all other cells, except the

one that we change the available bandwidth, have the same bandwidth of 7.8Mbps in Figure 3 (a).

Figure 11 shows the data transmission efficiency of different handoff schemes as a function of the

available bandwidth in new cell. In the proposed scheme, the transmission efficiency becomes worse as

the number of paths increases. However, the decrease of transmission efficiency is marginal compared to

the decrease from single path to 2-path case, since the area that allows the mobile node to acquire more

than two paths during handoff is very limited. Regardless of the number of paths, the transmission

efficiency is measured lowest when all cells have the same amount of available bandwidth (i.e., 7.8

Mbps). Among all other schemes, proxy-based scheme has the best performance and single path with

forwarding scheme is a little better than single path without forwarding scheme.

2) Network Capacity

The redundant transmission during handoff causes higher resource consumption in the wireless network

and may result in reduction of network capacity, which is defined as the maximum number of users

supported in a wireless network. We first evaluate this reduction factor in an analytical model, and then

show the simulation results to verify our analysis.

We analyze the reduction of network capacity in the worst case in which all cells have the same amount

of available bandwidth for each user. Therefore, the transmission efficiency of the proposed scheme is the

worst and results in greatest reduction in network capacity. Given a typical multi-cell wireless network

configuration as in Figure 12, we analyze the network capacity of a multi-cell wireless network with a set

of notations given in Table 5.

Without loss of generality, we use the marked hexagonal cell in Figure 12 for the analysis. For single

path cases, all users in the hexagonal cell use only the resource provided in this cell. Therefore, the

network capacity is calculated using following equation: (3). By considering the

20

resource used by users in the 2-path and 3-path area of the neighboring cells in the proposed scheme, we

get: (4) and (5). Detailed analysis is available

in [28].

We also run a simulation using the model in Figure 3 (b) to verify the analysis. In the simulation, mobile

node moves randomly in the multi-cell wireless network. The mobile node gets 400kbps in each cell. 10

30-minute long random traces are used and for each random trace, proxy-based scheme and the proposed

scheme using 2 and 3 paths in each handoff are applied. We measure the ratio of total packet transmission

in the proposed scheme to total packet transmission in the proxy-based scheme. This ratio xm-path shows the

increasing factor of resource consumption of a single user using the proposed scheme. Assume every user

in the wireless network using in proxy-based scheme, each user now uses

bandwidth in the proposed scheme. Therefore, we can calculate the network capacity using m paths in the

proposed scheme as: . Together with Equation (4) and (5), we have

and . Given the multi-cell network configuration in Figure 3(b),

we can calculate that the reduction factor of network capacity is 0.7692 for 2-path case and 0.7657 for 3-

path. Figure 13 shows a good agreement of simulation results and analytical results.

VI. CONCLUSION AND FUTURE WORKS

In this paper, a novel multi-path smooth handoff scheme for stream media is proposed. The proposed

scheme can provide smooth handoff for stream media in wireless networks with different bandwidth from

cell to cell. Through extensive simulations and careful analysis of the numerical results, we’ve evaluated

the performance of the proposed scheme. Our proposed scheme protects important application data (i.e.,

base layer video packets) through redundant transmissions on multiple paths, avoids congestion and

provides smooth stream rate and video quality adaptation regardless of bandwidth disparity during

handoff. It is also shown the performance of the proposed scheme is comparable to proxy-based handoff

scheme in many situations while it does not need excessive deployment of proxies in the Internet. The cost

21

of the proposed scheme is also carefully evaluated in terms of transmission efficiency during the handoff

and network capacity reduction. The result shows as the number of paths used in the handoff increases,

the network capacity will not decrease greatly although the transmission efficiency during the handoff

decreases, since 1) the size of overlapping area for handoff period is limited; 2) the area with more than

two available paths during the handoff is further limited by the configuration of multi-cell wireless

network.

To study the user-perceived video quality, we also test the proposed scheme together with a real wavelet

multi-layer encoder. The test results are promising and we plan to study more with other encoding

techniques, such as multi-descriptive coding, and their applications in stream CDN [29]. We plan to

investigate the extension of our scheme for multi-descriptive encoded stream in the future wireless

networks.

REFERENCES

[1] C. K. Toh, “The Design and Implementation of a Hybrid Handover Protocol for Multi-Media Wireless

LANs”, in Proc. of ACM Mobicom’95, page 49-61, Berkeley, CA, Nov. 1995

[2] K. Lee, “Adaptive Network Support for Mobile Multimedia”, in Proc. of ACM Mobicom’95, page 62-

74 Berkeley, CA, Nov. 1995

[3] A. Talukdar, B. Badrinath, A. Acharya, “Rate Adaptation Schemes in Networks with Mobile Hosts”,

in Proc. of ACM Mobicom’98, page 169-180, Dallas, Texas, Oct.1998

[4] S. Choi, K. G. Shin, “Comparison of Connection Admission-Control Schemes in the Presence of

Hand-offs in Cellular Networks”, in Proc. of ACM Mobicom’98, page 264-283, Dallas, Texas,

Oct. 1998

[5] A. Campbell, “QOS-aware Middleware for Mobile Multimedia Communications”, in Multimedia

Tools and Applications, Kluwer Academic Publishers, vol 7, page 67–82, 1998

[6] C. Perkins, Ed., “IP Mobility Support for IPv4”, RFC3344, Aug. 2002

22

[7] D. Johnson and C. Perkins, “Route Optimization in Mobile IP,” Internet Draft, draft-ietf-mobileip-

optim-09, work in progress, Feb. 2000

[8] H. Soliman, C. Castelluccia, K. Malki, L. Bellier , “Hierarchical Mobile IPv6 mobility management ”,

Internet Draft, draft-ietf-mobileip-hmipv6-07.txt, work in progress, Oct. 2002

[9] A. Campbell, J. Gomez, S. Kim, A. Valko, C. Wan, Z. Turanyi, "Design, Implementation and

Evaluation of Cellular IP", IEEE Personal Communications, Vol. 7, No. 4, page 42-49, Aug. 2000

[10] R. Ramjee, K. Varadhan, L. Salgarelli, S. Thuel, S.Y. Wang, and T. La Porta, "HAWAII: A

Domain-based Approach for Supporting Mobility in Wide-area Wireless Networks", IEEE/ACM

Transaction on Networking, Vol. 10, no. 3, page 396 - 410 Jun 2002

[11] K. El-Malki and H. Soliman, “Fast Handovers for Mobile IPv6”, Internet Draft, draft-ietf-

mobileip-fast-mipv6-00.txt, work in progress, Feb. 2001

[12] A. Helmy, "A Multicast-based Protocol for IP Mobility Support", in Proc. of 2nd ACM

International Workshop on Networked Group Communication (NGC 2000), page 49-58, Palo Alto,

CA, Nov. 2000

[13] A. Bakre and B. Badrinath, “I-TCP: Indirect TCP for Mobile Hosts”, In Proc. of 15th IEEE

ICDCS, page 136-143, Vancouver, Canada, May 1995

[14] H. Balakrishnan, S. Seshan, and R. Katz, “Improving Reliable Transport and Handoff Performance

in Cellular Wireless Networks”, ACM Wireless Networks, Vol 1, No. 4, page 469-481, Dec. 1995

[15] T. Goff, J. Moronski, and D. Phatak, “Freeze-TCP: A True End-to-End Enhancement Mechanism

for Mobile Environments”, In Proc. of IEEE INFOCOM’00, Vol 3, page 1537-1545, Tel-Aviv,

Israel, Mar. 2000

[16] K. Brown and S. Singh, “M-TCP: TCP for Mobile Cellular Networks,” In Proc. of ACM

SIGCOMM Computer Communication Review, Vol 27, No. 5, page 19–43, Oct. 1997

23

[17] MPEG ISO/IEC Joint Technical Committee, “Short MPEG-2 description” iso/iec jtc1/sc29/wg11

nMPEG00, Oct. 2000

[18] MPEG ISO/IEC Joint Technical Committee, “Mpeg-4 overview,” iso/iec jtc1/sc29/wg11 n4668,

Mar. 2002

[19] B. Vickers, C. Albuquerque and T. Suda, "Source-Adaptive Multi-Layered Multicast Algorithms

for Real-Time Video Distribution", IEEE/ACM Transactions on Networking, vol 8, No. 6, page

720-733, Dec. 2000

[20] R. Rejaie and J. Kangasharju, “Mocha: A Quality Adaptive Multimedia Proxy Cache for Internet

Streaming”, in Proc. of ACM NOSSDAV’01, page 3-10, Port Jefferson, NY, Jun. 2001.

[21] CDN Sota, [online], http://www.telin.nl/Middleware/cdn/state-of-art/cdnsota_toc.html

[22] T. Yoshimura, Y. Yonemoto, T. Ohya, M. Etoh, and S. Wee, “Mobile Streaming Media CDN

enabled by Dynamic SMIL,” in Proc. of International World Wide Web Conference, page 651-

661, Honolulu, HA, May 2002

[23] R. Stewart, Q. Xie, K. Morneault, C. Sharp, et al, “Stream Control Transmission Protocol”,

RFC2960, Oct. 2000

[24] D. S. Phatak, T. Goff, “A Novel Mechanism for Data Streaming Across Multiple IP Links for

Improving Throughput and Reliability in Mobile Environments”, in Proc of IEEE INFOCOM’02,

Vol 2, page 773-781, Jun. 2002

[25] H-Y. Hsieh and R. Sivakumar, “A Transport Layer Approach for Achieving Aggregate

Bandwidths on Multi-homed Mobile Hosts”, in Proc. of ACM Mobicom’02, page 83-94, Atlanta,

GA, Sept. 2002

[26] M. Handley, J. Padhye, S. Floyd, J. Widmer, “TCP Friendly Rate Control (TFRC): Protocol

Specification”, Internet Draft, draft-ietf-tsvwg-tfrc-05.txt, work in progress, Oct. 2002

[27] OPNET Modeler, [online] http://www.opnet.com/products/modeler/home.html

24

[28] Y. Pan, M. Lee, J. Kim, T. Suda, “An End-to-end Multi-Path Smooth handoff Scheme for Stream

Media”, technical report, [online]

http://netresearch.ics.uci.edu/ypan/MPATH_strm/MPATH_REPORT.pdf

[29] J. Apostolopoulos, T. Wong, W-T. Tan, S. Wee, “On Multiple Description Streaming with Content

Delivery Networks”, in Proc of IEEE INFOCOM’02, Vol 3, page 1736-1745, NY, Jun 2002

LIST OF TABLES

Table 1: Parameters for Different Available Bandwidth

Table 2: Parameters for Different Round Trip Time

Table 3: Parameters for Different Speeds of Mobile Node Movement

Table 4: Parameters to test Video Frame Rate

Table 1: Parameters for Different Available Bandwidth

Table 2: Parameters for Different Round Trip Time

Table 3: Parameters for Different Speeds of Mobile Node Movement

Table 4: Parameters to test Video Frame Rate

25

Table 5: Notations in Network Capacity Analysis

LIST OF FIGURES

Fig. 1: System architecture of multi-path handoff scheme

Fig. 2. Simulation network model

Fig. 3. Simulation models for overhead evaluation

Fig. 4. Throughput vs Available Bandwidth in the New Cell

Fig. 5. Throughput vs Round Trip Time

Fig. 6. Throughput vs Speed of Mobile Node Movement

Fig. 7. Packet Loss Ratio vs Available Bandwidth in the New Cell

Fig. 8. Packet Loss Ratio vs Round Trip Time

Fig. 9. Packet Loss Ratio vs Speed of Mobile Node Movement

Fig. 10. Video Frame Rate in Handoff

Fig. 11. Transmission Efficiency in Handoff

Fig 12. A typical configuration of a multi-cell wireless network

Fig 13. Reduction of Network Capacity

26

Table 5: Notations in Network Capacity Analysis

27

Fig. 2. Simulation network model

Fig. 1: System architecture of multi-path handoff scheme

28

Fig. 3. Simulation models for overhead evaluation

Fig. 4. Throughput vs Available Bandwidth in the New Cell

Fig. 5. Throughput vs Round Trip Time

29

Fig. 6. Throughput vs Speed of Mobile Node Movement

Fig. 7. Packet Loss Ratio vs Available Bandwidth in the New Cell

Fig. 8. Packet Loss Ratio vs Round Trip Time

30

Fig. 9. Packet Loss Ratio vs Speed of Mobile Node Movement

Fig. 10. Video Frame Rate in Handoff

Fig. 11. Transmission Efficiency in Handoff

31

Fig 12. A typical configuration of a multi-cell wireless network

Fig 13. Reduction of Network Capacity