Seamless Media Streaming over Mobile IP-Enabled ...

Seamless Media Streaming over Mobile IP-enabled Wireless LAN Dongwook Lee∗ , Chul-Ho Lee† , and JongWon Kim† ∗ Ultra

Fiber Optic Networks (UFON) Research Center Media Lab. Department of Information and Communications, Gwangju Institute of Science and Technology (GIST), Gwangju, 500-712, Korea, email:{dulee,chlee,jongwon}@gist.ac.kr

† Networked

Abstract— In the mobile IP-enabled wireless LAN (WLAN), packet transfer is interrupted due to the handoff of mobile node (MN), which results in burst packet losses. This transient behavior hurts time-critical streaming media applications. In case of one-way streaming media applications, it is well known that pre-buffering at the receiver side is effective in overcoming network fluctuations. However, determining the required margin of buffering is difficult in the Mobile IP network, since it depends on the efficiency of adopted link-/IP-layer handoff options. In this paper, we introduce a scheme that helps estimate the required pre-buffering level more accurately by considering both handoff duration and transient packet losses. For experiment, we implement a packet buffering and forwarding mechanism (a.k.a. smooth handoff) to reduce the packet losses during the link/IP-layer handoffs. Also, pre-buffering adjustment is performed based on the handoff latency experimentally measured and analytically obtained by the handoff transient time analysis. The experiment and simulation results shows that the proposed scheme can provide an appropriate guideline on the buffer parameters and thus can facilitate the seamless streaming over the Mobile IP networks. Keywords: Seamless media streaming, Mobile IP, Handoff transient time, and Pre-buffering.

I. I NTRODUCTION Mobile and wireless technologies have accelerated widespread use of multimedia services to mobile computers. In streaming applications, media streams have to be transmitted continuously, overcoming the fluctuation of network resources. The delay, jitter, and busty packet losses are usually addressed by adopting sufficient data buffering at the clients prior to playback [1]. This buffering, which is called ”pre-buffering”, can smooth network variations and give retransmission opportunity for lost packets. Under mobile networks, available bandwidth is scarce and even fluctuating severely. In addition, transmission itself is paused when a handoff happens. In fact, several mechanisms are proposed to reduce the blackout period when transmission is blocked due to handoff [2], [3]. However, these schemes are limited since they require special arrangements such as MAC bridge, additional resources, and corresponding signaling. Even with the fast handoff situations described in [4], packet losses are still present owing to the weak signal strength around a handoff. Thus, for streaming applications in mobile network, it is a good choice to use the pre-buffering technique to overcome possible shortage of buffer during a handoff. However, inaccurate and conservative

choice on the required buffering margin can waste limited latency budget, resulting in overall quality degradation. In this paper, we introduce a seamless streaming framework by estimating the accurate pre-buffering level to compensate handoff latency and by adapting rate-shaping scheme to overcome the shortage of bandwidth around a handoff. We calculate the handoff latency from the application point of view by extending the our previous work [5] in the Mobile IP networks with smooth handoff [6]. After that we implement a preliminary version of the proposed framework extending the basic Mobile IPv4 system. Network simulations as well as media streaming experiments over Mobile IP-enabled WLAN show that the proposed scheme can provide an appropriate guideline on the buffer parameters and thus can provide seamless streaming. II. S EAMLESS M EDIA S TREAMING F RAMEWORK A. Problem statement Handoff latency: In the Mobile IP-enabled WLAN environment, when a mobile node (MN) may move to another foreign network, it experiences handoff latency due to following handoff processes. First, the MN can communicate with only one AP at each time and it thus cannot communicate with an new AP before handoff. The MN should find a new AP and reattach to it during link-layer handoff procedure. Thus, linklayer handoff latency consists of probe, authentication, and reassociation delays. In addition, the registration process of IP-layer handoff can begin only after the link-layer handoff. Moreover, without link-layer triger, there exists more delay before the MN discovers change in the point of attachment by receiving agent advertisement message from the new FA. Also, the registration process takes some time to be completed, since the registration messages should propagate to a home agent (HA). Thus, network-layer handoff latencies includes agent discovery and registration delay. As stated above, total handoff latency is presented by time line for packet flows in Fig. 1. Packet loss: When a MN locates in a foreign network, its HA can intercept packets as proxy that go toward a home address of the MN. It then sends those packets through IPinIP tunneling to the care-of address (CoA) that indicates the termination point of a tunnel toward the MN side. As the MN

0-7803-8785-6/05/$20.00 (C) 2005 IEEE

Initial Buffering

oFA (Old Foreign Agent) nFA (New Foreign Agent)

Monitoring & Estimation

Binding Update message

Monitoring

HA Media Server

12 8

Buffering oFA

Target Buffer Level (BT) Estimation

11 10

7

Wired Network

9

6 5

4

3

Pre-buffering

Periodical Updating

Handoff

2

nFA

Buffering Policy Selection

IEEE 802.11b LAN 1

Packet Buffering

handoff Mobile node

(a) Handoff scenario Fig. 3.

Fig. 1.

Handoff latency of Mobile IPv4.

Streaming Server

Streaming Client

Video Archive Rate Adaptation Packet Scheduler

Fig. 2.

802.11 WLAN

Feedback Handling Server Buffer

Mobile IP Network FA

Display

Feedback

FA

FA

FA

FA

Pre-buffering Management

FA Mobile Node (MN) 802.11 WLAN

Monitoring & Estimation

Decoder Client buffer

Seamless media streaming framework in Mobile IP networks.

performs the link-/IP-layer handoffs to go to another foreign network, it cannot receive any packets because these packets are destined to old CoA and the MN cannot communicate with the old AP during the handoff period. Therefore, there exist burst packet losses hurting streaming media during the handoff period. B. Proposed Framework Fig. 2 shows the seamless media streaming framework under the Mobile IP networks [7], where IEEE 802.11 devices are configured as a WLAN infrastructure. A streaming application on a MN receives packets from a media server, while keeping an amount of packets in the client buffer to overcome resource fluctuations of network: available bandwidth, delay, and loss. The streaming server reacts to the feedback informed by the streaming client and performs quality adaptation, and packet schedule. The streaming client sends it’s status information to the server, which includes current buffer occupancy, receiving rate, error rate, and etc. Many studies on the relationship between feedback and reaction have been reported in [9]. In this work, we are focusing on the movement of client, handoff, which interrupts media delivery and spoils client’s streaming budget. Under the Mobile IP networks, a MN that currently plays the streaming content can move beyond the reach of foreign agent (FA) and initiate a handoff procedure. During the handoff, to register a new location of the MN, the MN sends ‘Binding Update’ (BU) message to both a corresponding node (CN) and a home agent (HA) [6]. After the CN (or the HA) receives the BU, it changes the route path from the old route destined to the old FA (oFA) (old stream) to the new one destined to the new FA (nFA) (new stream). The packets in the old route are forwarded to the nFA via

(b) Pre-buffering procedure Pre-buffering scenario.

the oFA according to the smooth handoff procedure. Thus, to receive the packets from the old stream, the MN should wait for an additional time as well as the L2 and L3 handoff delay. Moreover, it is known that the change of route path causes the packet sequence disruption [5]. In our framework, a handoff transient time is estimated before the handoff happens. To estimate the handoff transient time, the mobile client monitors network conditions such as link delay, flow rate, and queue status of the neighbor FAs. The handoff protocol and related signaling procedure are analyzed to get the handoff transient time. After estimating the handoff transient time, the streaming client tries to prepare enough data to compensate the estimated interruption time during the handoff. To acquire the target buffer level, which is estimated based on the given handoff transient time, the streaming client tries to collect packets, while keeping playback of the received media packets. There are two choices to boost required target buffer level: 1) increasing sending rate at the streaming server and 2) decreasing the playback speed at the streaming client. The choice depends on the policy of prebuffering management module. It is easily inferred that the required pre-buffering level will be changed along the time, since the link delays between FAs and HA or CN as well as traffic conditions are changing. Thus, it is needed to keep track of the exact required buffer level before a handoff. In Fig. 3(b), the estimation and target buffer level adjustment are performed periodically. To summarize, the pre-buffering scenario consists of target buffer level estimation and target buffer level adjustment as shown in Fig. 3(b). In the following section, we will analyze the handoff procedure to get handoff transient time which will be used to estimate target buffer level for pre-buffering. III. TARGET BUFFER LEVEL ESTIMATION FOR PRE - BUFFERING Handoff transient time estimation In this section, handoff transient time and out-of-order packet period are estimated, which is based on the results of previous work [5]. Under the Mobile IPv4 networks, a MN has a time period when it can not send and receive packets during handoff. We define this blackout period as the STP (silence time period). Also, we define the UTP (unstable time period) during when the packet sequence could be mis-ordered. The HTP (handoff time

0-7803-8785-6/05/$20.00 (C) 2005 IEEE

Handoff Initialization

Old stream

New stream 7Mpbs Throughput (Mbps)

Case 1

STP HTP

HA

Old stream

New stream

Case 2

Cp(t)

Cn(t)

4Mpbs

Lost data Rs(t): Source Sending Rate at t

STP

to

tn

1Mpbs

t0 t1 t2 t3

Case 3

tf

oFA

2Mpbs

UTP

HTP

UTP

STP

nFA

Handoff initiation

HTP

(a)

(b)

Fig. 6.

Time

Handoff Completion

Handoff scenario based on channel throughput.

Fig. 4. Possible packet orderings observed at the MN during the handoff (for the traffic from the HA to the MN). BIND MN

tAD

tw

tn

Qn

HA

reception. This attachment delay is set by tAD = tL2 + td . According to the delay of each message flows shown in Fig. 5, the Tf new , Tf old ,Tl old can be described as follow.

NEW_FIRST

NEW_FIRST

BIND nFA

nFA

tn

MN

tw

Qn

(a) BIND MN

tAD

tw

BIND nFA

tf

Qn

OLD_FIRST oFA

tf

Qo

OLD_FIRST nFA

MN

tw

Qn

Tl

(b) BIND MN

tAD

tw

BIND nFA

Qn

tn

LAST HA

to

LAST oFA

Qo

tf

LAST nFA

Qn

tw

MN

(c)

Fig. 5. Message flows and associated link delays for various packet streams (for the case when the MN moves from the to the nFA): (a) the first packet of the new stream (for Tf new ), (b) the first packet of the old stream (for Tf old ), and (c) the last packet of the old stream (for Tl old ).

period) is the whole handoff period. Depending on the location of oFA and nFA, the timing between the old and new streams can be classified into three cases. As shown in Fig. 4, we can associate the transient time periods (i.e., STP, UDP, and HTP) according to link delays in nFA, oFA, and HA. For example, the ‘Case 1’ illustrates the situation where a MN moves to the far-away (in network routing sense) nFA from its HA. Thus, by setting the time when the MN leaves the oFA to zero, the STP, UTP, and HTP can be denoted by new ),

(1)

U T P = max(0, Tl old − Tf new ), HT P = U T P + ST P + |Tf new − Tf old |,

(2) (3)

ST P = min(Tf

old ,

Tf

where Tf new is the time when the first packet of new stream arrives to the MN and Tf old and Tl old are the time when the first packet and the last packet of the old stream are delivered to the MN, respectively. The link delays between HA - oFA, HA - nFA, and oFA - nFA are denoted by to , tn , and tf , respectively. A handoff is started when a MN moves from the oFA. After the departure, there is a delay until the moved MN receives a Router Advertisement (RA) message from a nFA. This delay consists of handoff latency in a link layer and an agent discovery delay. We denote the L2 handoff delay and the agent discovery delay as tL2 and td , respectively. The propagation delay of wireless link is denoted by tw . Queueing delay of a packet in nFA, oFA, and HA is denoted by Qo , Qn , and Qh , respectively. The MN can recognize the change of attachment after L2-handoff completion and RA packet

old

Tf new = tAD + 2tw + Qn + Qc + 2tn , Tf old = tAD + 2tw + 2tf + Qo + Qn ,

(4) (5)

= tAD + 2tw + 2Qn + tn + to + Qo + tf .

(6)

Packet loss estimation Handoff is performed based on the received signal strength of received data. Fig. 6 describes a throughput profile when a MN moves to a nFA from a oFA. When a MN is moving to a nFA from an oFA, the signal strength (and related transmission throughput) from the oFA decreases [8]. It is clear that the transmission throughput of IEEE 802.11b is smallest when the MN is around handoff. The streaming client should consider the throughput degradation around a handoff as wall as the handoff processing delay described in previous section. Streaming application may not guarantee required bandwidth (so called the source sending rate at time t, Rs (t)), around a handoff. Accordingly, when the throughput is less than Rs (t), the packets sent by a streaming server will be lost. The throughput of the oFA channel is less than Rs (t0 ) at t0 and the MN starts a handoff at t1 . While the handoff is finished at t2 , the channel throughput of the nFA is less than Rs (t2 ) until t3 . The total packet loss caused by handoff can be divided into three losses: pre-loss, post-loss, and handoff-loss. The pre-loss, Lpre , and the postloss, Lpost , are packet losses during periods before and after the handoff, respectively. The handoff-loss, Lhandof f , is data losses during handoff duration. Then, the total loss, Ltotal , can be presented by Ltotal = Lpre + Lhandof
t
t f + Lpost , where Lpre = t01 (Rs (t)
− Cp (t))dt, Lpost = t23 (Rs (t) − Cn (t))dt, and Lhandof f = t∈ST P Rs (t)dt, where Cp (t) and Cn (t) is the throughput function of channel in the oFA and in the nFA at time t, respectively. Channel adaptation by feedback The packet losses caused by decreased signal strength can be reduced by adopting a rate shaping method [9]. Using the feedback sent by the client, the streaming server can adjust Rs (t) to the rate constrained by channel condition. For example, the streaming server reduces Rs (t) to Cp (t0 ) at t0 in Fig. 6. Generally, after a handoff, the received signal strength from the nFA is bigger than that from the oFA. Thus, Cp (t1 ) Tretx (LB ) = tRT T .
n tF

10M, 15ms 10M, 10ms

Playback Time (seconds)

Minimum buffer level for seamless handoff We now describe how to find the minimum pre-buffering level that a streaming client can tolerate a handoff. The video and audio data are first packetized at the server. For simplicity we assume that the media data is compressed such that one video frame and associated audio fit into one service frame. The service frame is segmented into n packets and the packets are transmitted at a constant rate. The packets are stored in the receiver buffer until the buffer reaches the target buffer size BT arget . When media playback is started, n packets are collected to reconstruct one service frame. Playback time of each service frame corresponds to inter-frame spacing, tF , at the sender. The buffered packets can be played during Tp B and it can be denoted by Tp =
T arget  · tF , where
x n represents the largest integer smaller or equal to x. Once the receiver faces buffer underflow, it stops playing media and stores incoming packet to the receiver buffer until the receiver buffer size reaches BT arget . The target buffer size is calculated according to the burst error length [1]. They figured out that the minimum buffer size, Bloss , to recover all lost packets can be written as

IEEE 802.11b LAN FA

10M, 10ms Media Server

8

12

16

20

24

Simulation Time (seconds)

(7)

(b) Fig. 8.

(a) Buffer consumption rate and (b) Playback time with respect to

where tRT T , LB , Tretx (x) represent the round trip time, the L2 handoff latency where the BH is 330 packets (33 frames). burst error length, and retransmission time of all x packets, respectively. The packet losses caused by decreased signal strength calculate the B T arget . In Eq. (1), the Tf old is selected as around handoff can be overcome by rate shaping controlled the STP in our simulation. Thus, L2 handoff latency (t ) is a L2 by client’s feedback information. Also, the smooth handoff main control variable to calculate the B T arget . The L2 handoff of Mobile IP eliminate the handoff-loss. Then, the streaming latencies of each FA are configured variously between 0.5 application can only consider the STP to calculate a pre- and 1.5sec. The IEEE 802.11b PHY is operated with 11Mbps buffering level. If we add the condition to Eq. (7), the target in normal condition. However, we change the PHY mode to buffer level, BT arget = BH , which can tolerate a handoff and 1Mbps in the oFA before 0.5sec of handoff initiation. And provide seamless playback is at the end of handoff, the PHY mode continues 1Mbps data  rate until 0.5sec later and goes to 11Mbps. In our simulation,  Tretx (Ltotal ), w/o rate shaping BH Tretx (Lhandof f ), with rate shaping =
(8) the 10% additional margin is added to the BT arget . The  ST P n maximum buffer limit is 10% bigger than the BT arget . Only tF , with rate shaping & smooth handoff. the streaming traffic is emitted to the link, and other traffic IV. S IMULATION AND E XPERIMENT are not introduced in the simulation, which makes queueing Our simulation is performed to show that the pre-buffering delay at FAs ignorable. level estimation method is validate. We also extend the HUT Fig. 8(a) represents the buffer consumption ratio with reversion of Mobile IP system to test the correctness of the spect to the L2 handoff latency variation under the BT arget seamless streaming framework. is 330 packets(tL2 = 1.0sec). The buffer drain rate is linearly increases according to the tL2 . When the tL2 is 1.5sec, 1.0sec, A. Simulation and Discussion and 0.5sec, 330 packets, 326 packets, and 268 packets in The simulation was based on the Network Simulator (NS- the buffer are drained respectively. However, when the tL2 is 2) [10]. Fig. 7 shows the simulation scenario and parameters. 1.0s without the rate shaping, the whole packet, 330 packets, Under the rate shaping option, the STP is main variable to are consumed without buffer underflow. The results are meet

0-7803-8785-6/05/$20.00 (C) 2005 IEEE

Fig. 9.

Fig. 10. nism.

Experiment Environment.

Foreign Agent Architecture for Buffering and Forwarding Mecha-

the result of the Eq. (8) well. The media playback time of a streaming client is shown in Fig. 8(b) where the target buffer level is 330 packets and maximum buffer limit is 363 packets, which are estimated for 1sec of L2 handoff latency. When the L2 handoff latency is 1sec, the playback is not interrupted but continuously served. However, we can observe that playback is interrupted whenever a handoff happens under it’s L2 handoff latency is 1.5sec. B. Experiment and Discussion The HUT version of Mobile IP [11] and the MPEG4IP implementations [12] for Linux are adopted for our experiment. On top of the HUT Mobile IP, we add a packet buffering and forwarding mechanism (a.k.a. smooth handoff protocol) to reduce packet losses during handoff, since the HUT version of Mobile IP does not have the smooth handoff protocol. Packet buffering and forwarding mechanism: To implement no packet losses during handoff, every packets transmitted to a MN via wireless channel are stored to the circular buffer of a current FA for a while. The FA keeps recently transmitted packets to its buffer. When a MN initiates handoff, some packets cannot be reached to the mobile node, but the copies of those packets are stored in the FA’s buffer. Receiving the BU message, the FA begins to forward the stored packets to the new FA of the MN. The stored packet are discarded from the buffer, when a buffered time exceed a time threshold. The time threshold of buffering time of a packet should be longer than a link/network handoff time. Also, the buffer size should be big enough to store all packets during handoff period.

The buffering and forwarding mechanism can eliminate the packet losses caused by handoff. However, the drawback of this method is that a successfully transmitted packet can be forwarded to the new FA, when the buffering time threshold are longer than an actual link/network handoff time. However, to preserve no packet loss during a handoff, the buffering time should be exactly equal or longer than the handoff time. Fig. 10 shows the FA architecture of the packet buffering. In order to transparently capture packets passing through the FA, we use Linux Divert socket [13]. The captured packet are stored in local circular buffer. Pre-buffering adjustment of the MPEG4IP player: From the Eq. (8), the pre-buffering time can be calculated as a function of link-layer handoff latency, the propagation delay of wireless link, the link delay between the old and new FA’s, and the queueing delay. The pre-buffering adjustment of the MPEG4IP player is performed only once at a beginning stage based on the handoff latency experimentally measured and analytically obtained by the previous section. Experiment results and discussion:We tested the performance of MPEG-4 video streaming over the modified Mobile IP system. The experimental testbed is depicted in Fig. 9. Cisco Aironet 350 series APs and client adapters supporting IEEE 802.11b are configured. The extended HUT Mobile IP and the MPEG4IP implementations for Linux are used. We also adopt the Apple Darwin Quicktime streaming server as a streaming server [12]. We set the period of agent advertisement broadcasting to 1sec, which is the lower boundary of the sending interval of agent advertisement messages in Mobile IP specification. We also configure that the radio coverage of APs between the adjacent different sub-networks can be overlapped. Handoff latency is computed as a time interval between the last packet transferred from the old FA and the first packet transferred from the new FA (except for the Mobile IP signaling). To measure required link delays both of wireless link delay (tw ) and link delay between old FA and new FA (tf ), we added the probing packet and measurement functions in the HUT MN and agent. In our testbed, we measured each delays several times and take the maximum values among them: tf = 1.84ms, tw = 2.09ms, tL2 = 700ms, td = 1sec, tAD = 1.7sec. According to Eq.(1), we set ST P to 1.708sec1 . Our experiment does not fully implement the proposed framework shown in Fig. 3 and it is a preliminary work. In our experiment, we simply use a fixed pre-buffering time estimated by maximum measurement result of STP. To evaluate the quality improvement for streaming media, we experiment MPEG-4 video streaming with the sending rate of about 1 Mbps and the frame rate is 25f rames/sec. To show seamless playback of streaming, we measure the sequence number of RTP packet and the decoded frame sequence in the following different cases: implemented proposed framework case using the packet buffering and forwarding mechanism 1 In the Mobile IPv4, STP can be greatly changed by t , which is a random d variable from 0 to 1sec.

0-7803-8785-6/05/$20.00 (C) 2005 IEEE

with pre-buffering adjustment and original handoff case with different pre-buffering level. The experimental result of the implemented proposed framework case is depicted in Fig. 11. When we applied our measurement result to the pre-buffering time, we observed that the playback did not interrupted during handoff. We can find that burst packet losses generated during STP of 1.2sec2 can be canceled out by the packet buffering and forwarding mechanism and those forwarded packets could reach the MN before the buffer of streaming client becomes underflow by the sufficient pre-buffering in the MPEG-4 client. However, in the experimental results of original handoff case represented in Fig. 12 and 13, we can know that the media playback is blocked by the burst packet losses generated during STP because the any packets cannot be received during the STP. In addition, since the pre-buffering time of 1sec is smaller than STP of 1.37sec in Fig. 12, there exists buffer underflow at the streaming client. Thus, additional quality degradation is generated after the handoff and even the streaming client doesn’t recover the smooth playback of the streaming media and doesn’t synchronize between audio and video of the streaming media for a long time. From this result, we can know that the pre-buffering adjustment, as well as the packet buffering and forwarding mechanism to prevent burst packet losses during handoff, is important to overcome buffer underflow and severe quality degradation.

(a)

(b)

Fig. 13. Original handoff with 2sec pre-buffering: (a) Received RTP sequence and (b) Decoded frame sequence.

sate the handoff latency. We calculated the handoff latency in application point of view under the Mobile IPv4 environment. The packet losses caused by decreased signal strength around handoff can be overcome by rate shaping controlled by client’s feedback information. Also, the smooth handoff of Mobile IP reduces the handoff-loss. Thus, the streaming application can only consider the STP to calculate the target buffer level. The simulation and experiment results show that the handoffaware streaming has no playback discontinuity while keeps a minimal pre-buffering level. For the future work, we will fully implement and experiment the proposed framework for seamless media streaming in the Mobile IP-enabled WLAN. ACKNOWLEDGEMENT This research was supported by grant R05-2004-000-109870 from the Basic Research Program of the Korea Science and Engineering Foundation. R EFERENCES

(a)

(b)

Fig. 11. Packet buffering and forwarding mechanism with pre-buffering adjustment: (a) Received RTP sequence and (b) Decoded frame sequence.

(a)

(b)

Fig. 12. Original handoff with 1sec pre-buffering: (a) Received RTP sequence and (b) Decoded frame sequence.

V. C ONCLUSION We introduced the seamless streaming framework by estimating the accurate buffer level for pre-buffering to compen-

[1] E. Steinbach, N. Farber, and B. Girod, ”Adaptive playout for low latency video streaming,” in Proc. of IEEE ICIP‘01, 2001. [2] H. Yokota, A. Idoue, T. Hasegawa, and T. Kato, ”Link layer assisted mobile IP fast handoff method over wireless LAN networks,” in Proc. of ACM MobiCom ’02, Atlanta, Georgia, Sep. 2002 [3] T. Zhang, J. C. Chen, and P. Agrawal, ”Distributed soft handoff in all-IP woreless networks,” in Proc. of 3GWireless01, San Francisco, May 2001. [4] R. Koodli, “Fast handovers for mobile IPv6,” Internet Draft, Internet Engineering Task Force, March. 2003. [5] D. W. Lee and J. Kim, “Out-of-sequence packet analysis in mobile IP handoff and its enhancement,” in Proc. of 3G Wirelesss ‘02, San Francisco, CA, May 2002. [6] C. Perkins and D. Johnson, ”Route optimization in Mobile IP,” draft-ietfmobileip-optim-11.txt, Sep. 2001. [7] C. Perkins, “IP Mobility Support,” RFC 2002, Internet Engineering Task Force, Oct. 1996. [8] J. D. Pavon and S. Choi, ”Link adaptation strategy ofr IEEE 802.11 WLAN via received signal strength measurement ,” in Proc. of IEEE ICC‘03, 2003. [9] G. J. Conklin, et. al, ”Video coding for streaming media delivery on the Internet,” IEEE Trans. on Circuits and Systems for Video Technology, vol.11, no.3, pp.269-281, 2001. [10] NS-2 Homepage, http://www.isi.edu/nsnam/ns/. [11] (2001) Dynamics - HUT Mobile IP, [Online] Available: http://dynamics.sourceforge.net/ [12] B. May. (2003) MPEG4IP - Open Streaming Video and Audio. [Online]. Available: http://www.mpeg4ip.net/ [13] W. Kellerer, E. Steinbach, P. Eisert, and B. Girod, ”A real-time Internet streaming media testbed,” in Proc. of International Conference on Multimedia and Expo (ICME 2002), Aug. 2002.

2 Usually, an actually measured STP is less than the estimated STP, which is chosen as the worst cast of measurement.

0-7803-8785-6/05/$20.00 (C) 2005 IEEE