Mobile devices such as PDA and laptop computer come into wide use and the demand of IP-based ..... We also adapt Apple Darwin. Quicktime streaming server ...
Submitted to Network Research Workshop 2004 / 18th APAN Meetings
Seamless MPEG-4 Video Streaming over Mobile IP-enabled Wireless LAN Chul-Ho Lee, Dongwook Lee, and JongWon Kim Networked Media Lab. Dept. of Information and Communications Kwang-Ju Institute of Science and Technology (K-JIST), Gwangju, Korea +82-62-970-2274
{chlee, dulee, jongwon}@netmedia.kjist.ac.kr ABSTRACT In the mobile IP-enabled wireless LAN (WLAN), packet transfer is interrupted due to the handoff of a mobile node at the link and network layers, which results in burst packet loss. This transient behavior is hurting time-critical streaming media applications. Many ideas mitigating the interruption have been suggested such as reducing handoff latency and minimizing packet losses. Also, in the applications perspective, it is known that extending the prebuffering at a streaming client can mitigates the playback disruption due to handoffs. However, in the empirical experiment, seamless streaming still cannot be achieved due to the packet loss during handoff period in mobile agents. Moreover, the lack of study on the relation of the pre-buffering size and the handoff methods causes the over estimated pre-buffering. In this paper, targeting seamless streaming over the mobile IP-enabled WLAN, we extend the mobile IPv4 protocol and experiment on the MPEG-4 video streaming with the implemented mobile IP system. A packet forwarding with buffering mechanism is implemented to reduce the packet loss during link-/IP-layer handoff. Pre-buffering adjustment is performed based on the handoff transient time analysis [11]. The experimental results show that the proposed approach can reduce packet losses during a handoff and provide the feasibility of seamless MPEG-4 video streaming.
General Terms Performance, Design, Experimentation
Keywords WLAN, Mobile IP, MPEG-4 video streaming, Pre-buffering, and Packet forwarding with buffering mechanism
1. INTRODUCTION Mobile devices such as PDA and laptop computer come into wide use and the demand of IP-based multimedia services on mobile devices over wireless network increases rapidly. Mobile IP [1] is a protocol to provide IP mobility. In the mobile IP-enabled WLAN, mobile nodes can access Internet and communicate with correspondent nodes regardless of their location and time. The mobile nodes also can move on another network while accessing network. To provide the IP mobility, the basic mobile IP defines the handoff procedure between different sub-networks. In general, the handoff procedure of a mobile node consists of link-layer handoff and IP layer handoff. In the mobile IP-based WLAN, when a mobile node changes its associated access point (AP) to another AP in same sub-network, it suffers a latency caused by measurement, selection, and association processes for
link-layer handoff. Moreover, processing IP-layer handoff, it also spends time for discovering a new foreign agent (FA) and performing registration process of its new location. These handoff latencies cause interruption of packet transfer resulting in packet losses. This transient behavior is hurting time-critical streaming media applications; playback of streaming media can be stopped during handoff procedure due to the lack of received data. Many ideas mitigating the interruption have been suggested such as reducing handoff latency in mobile IP [3,4] and minimizing packet losses through a packet forwarding with buffering mechanism [5,6,7]. Also, in the applications perspective, it is known that extending the pre-buffering at a streaming client can mitigates playback disruption due to handoff, since the buffered packets in the streaming client can compensate the handoff delay. Although handoff latency is reduced through proposed handoff mechanisms, packet flows are still interrupted during link-layer handoff and thus seamless streaming still cannot be achieved [3,4]. The packet loss originated from the interruption can be canceled by the packet forwarding with buffering mechanism [5,6,7]. However, it is requested that the need of research about the influence of the packet forwarding with buffering mechanism on an application layer for seamless streaming, since those only concentrate on minimizing packet loss. In addition, those studies aren’t concerned with implementation issues and empirical experiments of the suggested handoff mechanisms. Moreover, the lack of study on the relation of the pre-buffering size and the handoff methods causes the over estimated pre-buffering. In this paper, targeting seamless streaming over WLAN, we extend the mobile IPv4 protocol and experiment on the MPEG-4 streaming with the implemented mobile IP system. A packet forwarding with buffering mechanism is implemented to reduce the packet loss during link-/IP-layer handoff. The causes of packet loss during handoff period are discussed and solutions for the packet loss are presented in the implementation perspective. Also, pre-buffering adjustment is performed based on the handoff transient time analysis [11]. The experimental results show that the proposed approach can reduce packet loss during handoff period and provide the feasibility of seamless media streaming. The rest of the paper is organized as follows. Section 2 describes the related problem and the proposed approach for seamless streaming over mobile IP-enabled WLAN. Section 3 explains the methodology and procedure of the implementation for used techniques. Experiment and discussion are presented in section 4. Finally we present conclusion and future work in section 5.
Submitted to Network Research Workshop 2004 / 18th APAN Meetings
2. SEAMLESS STREAMING OVER THE MOBIE IP-ENABLED WLAN To achieve seamless streaming over the mobile IP-enabled WLAN, we should consider the following limitations and solve related problems. First, the network bandwidth of WLAN is scarce and limited as compared to wired network. There also exist fluctuated packet transmission and burst packet loss due to the fading/shadowing of wireless channel and channel contention. Mobility issues including the handoff raise transient packet loss and disruption. There are some other limitations such as security and power consumption of wireless terminals. In this paper, we are concerned with solving the following handoff-related problems to achieve seamless streaming, since the transient behavior of handoff is fatal to streaming media even if the other issues are also important for seamless streaming.
there exists burst packet loss hurting streaming media due to handoff latency. In this paper, in order to settle this problem we implemented packet forwarding with buffering mechanism in a FA and performed pre-buffering adjustment in a streaming client.
− Packet forwarding with buffering mechanism in a FA − Sufficient pre-buffering in a streaming client
2.1 Problem statement In the mobile IP-enabled WLAN environment, when a mobile node moves another foreign network, there exists handoff latency due to following handoff properties. First, the mobile node can communicate with exactly one AP before and after handoff and it cannot communicate with an old FA during link-layer handoff. In addition, the registration process of IP-layer handoff can begin only after the link-layer handoff. This handoff latency is to perform the link-layer handoff including probe, authentication, and reassociation delays [2]. Without link-layer information, there exists delay before the mobile node discovers a change in the point of attachment through receiving agent advertisement message from the new FA. Also, the registration process takes some time to complete as the registration messages propagate to a home agent (HA) through the Internet. Accordingly, additional handoff latency is needed for the agent discovery and the registration process. As is stated above, total handoff latency is presented by time line for packet flows in Figure 1. When a mobile node locates in a foreign network, the HA intercepts packets that are destined a home address of the mobile node as proxy and sends those packets through IPinIP tunneling to care-of address (CoA) that indicates the termination point of a tunnel toward the mobile node. As the mobile node performs link-/IPlayer handoff to move on another foreign network, it cannot receive any packets toward the mobile node because these packets are destined old CoA and the mobile node cannot communicate with the old AP and the old FA during handoff period. Therefore,
(a)
Figure 1. Handoff latency.
2.2 Packet forwarding with buffering mechanism in a FA In order to minimize burst packet loss produced during handoff period, a packet forwarding with buffering mechanism (for smooth handoff) in a FA was proposed in [5]. Figure 2 indicates the procedure of the packet forwarding with buffering mechanism in a FA. First, a FA decapsulate and send packets toward a mobile node and also buffers these packets. When the mobile node moves on another foreign network, it appends a Previous Foreign Agent Notification extension to a registration request message as an extension of the registration process and sends them to a new FA. The new FA then sends a Binding Update message to the old FA
(b)
Figure 2. The procedure of packet forwarding with buffering mechanism in a FA.
Submitted to Network Research Workshop 2004 / 18th APAN Meetings as well as a registration request message to the HA. When the old FA receives a Binding Update message, it checks validation of that message; if that message is validated, it updates its binding cache, sends Binding ACK message to the mobile node through the new FA and re-tunnels buffered packets, along with any future packets tunneled to it, to the mobile node’s new CoA. Thus, burst packet loss generated during handoff period can be completely eliminated due to the packet forwarding with buffering mechanism in a FA.
The HUT Mobile IP and the MPEG4IP implementations for Linux are adopted as a mobile IP system and a MPEG-4 client, respectively [8,9]. Based on these implementations, we implement the process routine and related messages (such as Previous Foreign Agent Notification extension, Binding Update and Binding ACK [6]) for a packet forwarding with buffering mechanism into a basic mobile IP and also efficiently perform pre-buffering adjustment in a streaming client.
At this point, how many packets are buffered at an old FA is an important issue. It is related to handoff latency. If handoff latency can be reduced, buffering memory can also be smaller size. First, handoff latency for agent discovery can be simply reduced through more frequent agent advertisement and overlapped cell planning (i.e., the radio coverage of adjacent APs is overlapped.). Also, if a higher layer can know link-layer information and recognize link-layer handoff on a mobile node, handoff latency for the agent discovery can be more reduced. That is if a mobile node detects link-layer handoff, it can simply discover a change in the point of network attachment through broadcasting agent solicitation message. In addition, additional performance improvement can be achieved through hierarchical FAs management [3]. The FAs in a domain are organized into a hierarchy to handle local movements of mobile nodes with in the domain. Therefore, handoff latency is reduced for registration process because registration process with a HA isn’t always needed and mobility can be locally managed.
2.3 Pre-buffering in a streaming client In general, as playing streaming media, pre-buffering in a streaming client can be used to compensate network jitter and delay. It also can apply forwarded packets that were buffered in an old FA during handoff period to the playback of streaming media. However, if the amount of pre-buffering is not enough to overcome the discontinuance of data transmission generated during handoff period, although buffered packets are forwarded to a mobile node, these packets cannot be applied to the playback of streaming media and thus playback disruption is happened. Also, if the amount of pre-buffering goes to access, it produces the waste of memory usage for the pre-buffering and the delay of staring time for the playback. Therefore, to find the optimized pre-buffering time is also important issue.
3. IMPLEMENTATION The experimental environment to provide seamless MPEG-4 streaming over mobile IP-enabled WLAN is depicted in Figure 3.
Figure 3. Mobile IP-enabled WLAN environment. The implemented packet forwarding with buffering mechanism is to minimize burst packet loss due to handoff and the architecture is depicted in Figure 4. The core routine of packet forwarding with buffering mechanism is implemented in FA, as depicted in Figure 4 (b), and presented as following procedures. First of all, besides decapsulating tunneled packets and delivering them directly to a mobile node, a FA also keeps these packets in circular buffer. In order to transparently buffer packets passing through the FA with destination, the mobile node, we use Linux Divert socket [10]. It is also used to re-inject buffered packets that will be forwarded to the mobile node through a new FA. The Linux Divert sockets enable IP packet interception and injection on end systems as well as on routers. Packets are intercepted on an IP layer and are available for user processes outside the kernel via a modified version of raw sockets. Divert sockets rely on the IP firewall mechanism for packet filtering. At this time, the
(a) HA
(b) FA Figure 4. Implemented Mobile IP architecture.
Submitted to Network Research Workshop 2004 / 18th APAN Meetings circular buffer size and related buffering time are based on the consumed time to perform the worst case of link-/IP-layer handoffs. It is also related to find the optimized pre-buffering time in a streaming client and thus will be described later on. In addition, in order to make the tunnel between an old FA and a new FA for packet forwarding, we utilize the routing and IPinIP tunneling capabilities of the Linux operating system. In the old FA, the tunnel toward the new FA is established as an outgoing interface. Routing rules and tables are added for packets destined to a mobile node that will be re-tunneled to the new FA through the outgoing tunnel interface. Also, in the new FA, the tunnel between the old FA and the new FA is established as an incoming interface. Routing rules and tables are also added for a connection between the incoming tunnel interface and an outgoing Ethernet interface so that re-tunneled packets are delivered from the incoming tunnel interface to the outgoing Ethernet interface. Using this tunneling connection, buffered packets locating in the old FA can be re-tunneled to the new FA and all packets arriving at the old FA with destination, the mobile node, are immediately tunneled to new FA. Then, these re-tunneled packets are forwarded to the mobile node through the new FA. We adjust the pre-buffering time in MPEG4IP implementation to provide efficient space in the playback buffer for streaming media in a streaming client. A streaming client at a mobile node should be able to play streaming media without playback disruption while it consumes time for handoff latency including link-layer handoff latency, agent discovery time and the end-to-end delay of messaging to binding update among an old FA, a new FA and a mobile node (i.e., a mobile node cannot receive any packets during the handoff period). Thus, the streaming client should have the pre-buffering time to mitigate the playback disruption of streaming media that can be happened during the handoff period at least. At this point, the optimized pre-buffering time can be calculated as a function about link-layer handoff latency, the propagation delay of wireless link, the link delay between an old FA and a new FA, and the queueing delay and scheduling method for buffered packets in an old FA [11].
350 series APs and client adapters supporting IEEE 802.11b are used as wireless equipments in experiments. As was mentioned in the previous section, the HUT Mobile IP and the MPEG4IP implementations for Linux are adopted as a mobile IP system and a MPEG-4 client, respectively [8,9]. We also adapt Apple Darwin Quicktime streaming server as a streaming server in our experiments [9]. During experiments basic mobile IP function is only used without any link-layer information. To reduce handoff latency for agent discovery, we force a mobile agent to send out agent advertisement message regularly with the interval of 1 second (i.e., the lower boundary of the sending interval of agent advertisement messages in mobile IP specification) and configure that the radio coverage of APs between the adjacent different sub-networks can be overlapped. In addition, a mobile node can sends the registration request immediately after the discovery of a new mobile agent by policy setting in the HUT mobile IP implementation. In the experimental environment, measured handoff latency is approximately 2 second through the repeated experiments. This handoff latency is computed as the interval between the last packet transferred from an old FA and the first packet transferred from a new FA except mobile IP messaging. Based on this measured value about the handoff latency, we set the buffering time related to circular buffer size at 3 second with considering the variation of the handoff latency.
Table 1. Machines used in experiments. Machine MN HA oFA nFA Streamin g Server
Pentium III - Mobile CPU 733MHz RAM – 256MB Pentium III – 700MHz RAM – 512MB Pentium III – 800MHz RAM – 256MB Pentium III – 500MHz RAM – 128MB Pentium IV – 1.5GHz RAM – 512MB
Operation System Linux 2.4.20-8 Linux 2.4.18-3 Linux 2.4.18-3 (Divert socket support)
Figure 5. The performance comparison for UDP stream: the number of packet losses vs. the sending rate of UDP packets.
Linux 2.4.18-3 Linux 2.4.19
4. EXPERIMENT AND DISCUSSION We have measured the performances of UDP stream and MPEG-4 video streaming with our implemented packet forwarding with buffering mechanism in a FA and the pre-buffering adjustment at a streaming client under our mobile IP-enabled WLAN environment. Our experiment is restricted with performance measure as a mobile node is faced with handoff under our testbed. The experimental testbed is depicted in Figure 3. Besides machines used in experiments shown in Table 1, Cisco Aironet
4.1 Performance measure for UDP stream In order to validate the performance improvement for UDP stream, we measured on UDP packet loss during handoff period through following experiment. While a streaming sever periodically sends UDP packets with constant bit rate (CBR) to a mobile node, as the mobile node performs link-/IP-layer handoff to move on another foreign network, the performance of UDP stream can be measured by checking the number of packet losses generated during handoff period. The UDP packet size is constant as 1000bytes. Figure 5 shows the performance comparison for UDP stream between basic mobile IP system and our implemented mobile IP system using the packet forwarding with buffering in a FA as the sending rate of UDP packets increase. We can know that the implemented
Submitted to Network Research Workshop 2004 / 18th APAN Meetings mobile IP system has better performance relatively than the basic mobile IP system as sending rate increases. In the basic mobile IP system, the number of packet losses generated during handoff period increases according to the increment of sending rate. On the other hand, although the sending rate increases, there exist few packet losses (~ 5 packet losses) in the implemented mobile IP system, which will be discussed in the section 4.3. This means packet loss due to handoff can be almost eliminated by a packet forwarding with buffering mechanism in a FA.
4.2 Performance measure for MPEG-4 video streaming In order to evaluate the quality improvement for streaming media, we experimented on MPEG-4 video streaming with the sending rate of about 0.98 Mbps under the same environment as experiment for UDP stream. In the MPEG-4 video streaming experiment, we found that the forwarded packets cannot be applied to the playback of streaming media, as the pre-buffering time was set at under 2 second in a streaming client. That is if prebuffering time in a streaming client is insufficient, there exists playback disruption since buffered packets are forwarded to a streaming client after buffer underflow resulting from the discontinuation of data transmission during handoff period. Then, the pre-buffering time was set at 3 second based on the measure handoff latency and the study result of the handoff transient time analysis in [11]. Figure 7 indicates the performance comparison for MPEG-4 video streaming between basic mobile IP system and our implemented mobile IP system. We can find handoff period when any packets cannot be received in both Figure 7 (a) and (b). However, the level of the receiving rate of RTP packets doesn’t change after the handoff period in the basic mobile IP system, while the level of the receiving rate is higher for some time after the end of handoff period in the implemented mobile IP system since buffered packets are also forwarded as well as newly transmitted packets through a new FA. In the basic mobile IP system, there exists burst packet loss during handoff period even if handoff latency is shorter relatively and thus these packet losses affect video and audio quality degradation on the MPEG-4 video streaming. Also, additional quality degradation is generated from previous degradation and even the streaming client doesn’t recover the smooth playback of
(a)
the streaming media and synchronize between the video and audio of the streaming media for a long time. Even if the pre-buffering time is sufficient in a streaming client and thus the playback disruption of streaming media doesn’t occurs, there exists burst packet loss and non-acceptable video and audio quality degradation on the MPEG-4 video streaming. However, in the implemented mobile IP system, although there exist a few packet losses due to handoff, these packet losses are not burst but random and thus are acceptable to the playback of the streaming media. We could find the smooth playback of the streaming media overcoming the discontinuation of data transmission generated during handoff period through the packet forwarding with buffering mechanism in a FA and sufficient pre-buffering in a streaming client.
4.3 Discussion In the experiments on our implemented mobile IP system, we found out the two reasons of random losses due to handoff. The one is about the instantaneous amount of re-injected packets using Divert socket to forward buffered packets to a mobile node through a new FA. If an old FA re-injects whole buffered packets at a time, there exist several random losses. Therefore, we forced an old FA to send smoothly buffered packets to a mobile node through a new FA to minimize random losses related to the reinjection problem. The other is about the preparation of the tunnel establishment and routing entry for a mobile node in a FA. When a FA receives the registration reply from a HA, it confirms the request and connects an incoming tunnel interface to an outgoing tunnel interface. The FA also adds the routing rule and entry for a mobile node so that packets coming from an incoming tunnel interface go to a right outgoing tunnel interface. When a mobile node move on another foreign network, after registration process a HA changes the route about the mobile node to the new location of the mobile node and then all packets destined to the mobile node will be routed to this new path. At this point, if a new FA may not be ready to forward packets to the new path of the mobile node (i.e., it isn’t yet finished that the tunneling establishment or routing entry for the new path of the mobile node) depending on system performance and network latency, when some of the packets already comes from the HA to the new FA, these packets will be dropped. This means that some of the packets in a certain time period will be additionally dropped
(b)
Figure 6. The variation of the receiving rate of RTP packets in a streaming client.
Submitted to Network Research Workshop 2004 / 18th APAN Meetings during handoff period. We will solve this problem to provide better video and audio quality of streaming media although packet losses generated by this problem are only a few.
5. CONCULSION AND FUTURE WORK In this paper, we presented the causes of burst packet loss during handoff period and the proposed approach mitigating the playback disruption due to the packet loss in the implementation perspective to achieve seamless streaming over the mobile IPenabled WLAN environment. Because of the transient behavior of handoff, packet flows are interrupted and thus burst packet loss is happened. There thus exist the playback disruption and serious quality degradation of streaming media, as a mobile node performs link-/IP-layer handoffs to move on another foreign network. The proposed approach solves this problem by the implementation of the packet forwarding with buffering mechanism in a FA and pre-buffering adjustment in a streaming client to reduce burst packet loss and alleviate the playback disruption of streaming media. The experimental results show that the proposed approach can minimize packet loss during handoff period and provide the feasibility of seamless media streaming. Although the proposed approach is verified on a flat mobile IPv4 experimental environment, it can be extended to mobile IPv6 and a hierarchical mobile IP model. In future, first we will solve the problem described as an experimental result about the preparation of the tunnel establishment and routing entry for a mobile node in a FA. We also plan to extend the proposed approach to handoff between inter APs in same sub-network as well as handoff between different sub-networks (i.e., different foreign network). Finally, we will achieve the additional performance improvement of streaming media through mobility-aware streaming as well as the reduction of the handoff latency supported by link-layer information such as handoff occurrence and wireless channel information.
6. ACKNOWLEDGMENTS This research was supported by University IT Research Center Project.
7. REFERENCES [1] C. Perkins, “IP Mobility Support,” RFC 2002, October 1996. [2] A. Mishra, M. Shin and W. Arbaugh, “An empirical analysis of the IEEE 802.11 MAC layer handoff process,” submitted to ACM Computer Communications Review, 2002.
[3] D. Forsberg, J. K. Malinen, J. T. Malinen, T. Weckström, and M. Tiusanen, “Distributing mobility agents hierarchically under frequent location updates,” in Proc. of Sixth IEEE International Workshop on Mobile Multimedia Communications (MOMUC'99), November 1999.
[4] K. El Malki, “Low latency handoffs in Mobile IPv4,” Internet Draft, Internet Engineering Task Force, January 2004.
[5] C. Perkins and D. K-Y. Wang, “Optimized smooth handoffs in Mobile IP,” in Proc. of IEEE Symposium on Computers and Communications, July 1999.
[6] C. Perkins and D. Johnson, “Route optimization in Mobile IP,” draft-ietf-mobileip-optim-11.txt, September 2001.
[7] C. Blondia, N. Van den Wijngaert, G. Willems, and O. Casals, “Performance analysis of optimized smooth handoff in Mobile IP” in Proc. of the 5th ACM International workshop on Modeling, Analysis and Simulation of Wireless and Mobile Systems (MSWiM2002), September 2002.
[8] “Dynamics – HUT Mobile IP,” http://www.cs.hut.fi/Research/Dynamics, 2001.
[9] “MPEG4IP – Open Streaming Video and Audio,” http://www.mpeg4ip.net
[10] W. Kellerer, E. Steinbach, P. Eisert, and B. Girod, “A realtime Internet streaming media testbed,” in Proc. of International Conference on Multimedia and Expo, ICME 2002, August 2002.
[11] D. Lee and J-W. Kim, “Transient time period analysis of smooth handoffs in mobile IP networks and its application to media streaming,” in Proc. SPIE Applications of Digital Image Processing XXVI, August 2003.
