Accelerating Mobile Ip Hand-offs Through Link

Transcript

Accelerating Mobile IP hand-offs through Link-layer Information, An Experimental Investigation with 802.11b and Internet Audio N. A. Fikouras, A. J. Könsgen and C. Görg Department of Communication Networks, University of Bremen ComNets - FB 1, Kufsteiner Straße NW1, 28359 Bremen, Germany {niko, ajk, cg}@comnets.uni-bremen.de Abstract In this study the Fast Hinted Cell Switching movement detection method for Mobile IP is introduced. This method assumes that the Link-Layer is capable of delivering information to Mobile IP regarding the identity of the local mobility agent. This tends to negate the need for Mobile IP movement detection as well as agent discovery and leads to accelerated Mobile IP hand-offs. Moreover, the existence of Link-Layer information renders the presence of periodic mobility agent broadcast advertisements as unnecessary which enables a more efficient utilisation of the network capacity. The performance of the proposed solution is experimentally compared against that of Eager Cell Switching that is considered the fastest of traditional Mobile IP movement detection methods. The scenarios investigated involve straight line movement patterns and Mobile IP hand-offs between Foreign Agents that provide wireless network connectivity through IEEE 802.11b Wireless LAN standard, compliant hardware. The IEEE 802.11 SSID field was utilised to contain information important to Mobile IP such as the identity of the local mobility agent. The overlying communication involves a unidirectional GSM audio stream directed to the mobile node. Through the experimental investigation it is determined that the proposed solution outperforms its Eager Cell Switching counterpart by managing faster Mobile IP hand-offs and minimum packet loss of the active audio stream. Introduction The recent years have brought an infiltration of Internet services in all aspects of modern life. This development has given place to a new trend in fixed networks, whereby data traffic has been overtaking voice traffic. It is expected that this tendency will expand into wireless networks. This expectation becomes all the more important as forecasts predict that as early as 2003 almost half of the Internet population will consist of mobile access devices [1]. Furthermore, the expansion of the Internet and its deployment for a wide range of applications have indicated the Internet Protocol (IP) as a potential bearer for voice traffic. In this manner telecommunication services can be provided through data communication services. The Internet Protocol was not designed with any considerations for mobility. Mobile IP (MIP) [2] is an extension to the basic protocol design that enables mobile nodes to roam while not interrupting established communications and while remaining reachable on a permanently allocated IP address. MIP introduces several new overheads, including IP layer hand-offs. MIP hand-offs always accompany a Link-Layer (LL) hand-off where mobility has been traditionally managed and kept transparent from higher layers. However, due to layer independence and lack of communication between the LL and MIP layers, the latter is required to rely on network layer based mechanisms in order to perform movement detection as well as local agent discovery. During MIP hand-offs the mobile node is unable to send or receive traffic and therefore it is said to suffer network service disruption. Research [3] has indicated that the interruption introduced by the MIP movement detection process alone may range up to 3 seconds. However, when delivering telecommunication services such as telephony over IP, an interruption of that scale may not be acceptable. Further research [4] has shown that MIP hand-offs can be accelerated by establishing a communication between the MIP and lower layers that enables MIP to realise changes such as LL hand-offs that would otherwise have to be determined through alternative time-consuming means. In this study it is shown that MIP hand-offs between networks with the IEEE 802.11b Wireless LAN (WLAN) standard can further be accelerated by extending the information communicated between 802.11b and MIP. For the purposes of this study, the 802.11b Service Set Identifier (SSID) was used to communicate to MIP, information about its current network location, such as the identity of the local Mobility Internet Internet Network C Network A CN HA FA Network B FA Next Position FA FA Network A FA AP AP AP Network B Cell 1 Next Position Cell 2 Cell 3 movement Figure 1 - The Mobile Internet Protocol Agent. As will be shown, the proposed approach negates the need for MIP movement detection and leads to significantly faster MIP hand-offs. In the following sections, the MIP is briefly introduced with a focus on MIP movement detection and IP layer hand-offs. This is followed by the introduction of a new movement detection method that relies on LL information to accelerate MIP hand-offs. Its performance is compared through an experimental set-up against that of a traditional MIP movement detection method, namely Eager Cell Switching. The scenarios investigated involve straight line movement patterns with an overlying audio stream that remains active during MIP hand-offs. For the audio stream, the popular GSM encoding was selected. The Mobile Internet Protocol The Internet Protocol (IP) was originally designed without any mobility support. The routing mechanisms of IP assume that a node (host or router) maintains a point of attachment to the Internet, indicated by its IP address. Should a mobile node change its point of attachment and move to a new location incompatible with its IP address, it would be unable to send or receive traffic. The MIP is an extension to the basic protocol design for Internet mobile host support. MIP provides a functionality similar to the post-office forwarding scheme in order to provide network connectivity to roaming mobile hosts. MIP introduces three new network entities, namely the Home Agent (HA), Foreign Agent (FA) and Mobile Node. Alternative MIP configurations (MIPv6) may omit the FAs by distributing their functionality amongst the Mobile Node and the network infrastructure. movement movement Figure 2 – IP layer mobility Every Mobile Node is permanently allocated an IP address in its Home Network. Every such network is required to host a HA. Every time that a Mobile Node moves, it is required to register its current point of attachment to the Internet with the HA. For every registered Mobile Node the HA is required to act as a proxy in the home network, intercept all incoming traffic from a given sender – termed as correspondent node (CN) - and redirect it through packet encapsulation to the Mobile Node’s most recently registered location (Figure 1). Mobile IP hand-offs Node Mobility dictates the need for Mobile Node hand-offs. In general, whenever a Mobile Node leaves a network and enters another, it is required to perform a hand-off. A MIP hand-off occurs whenever a Mobile Node moves between two IP networks (subnets) that maintain a Mobility Agent (HA or FA). In Internetworks such as the one presented in figure 2 the Mobile Node has to perform a MIP hand-off every time it moves between networks A and B. In this case the MIP hand-off accompanies an 802.11b hand-off, as node movement also signifies a location switch between wireless cells 1 and 2 and their corresponding Access Points (AP). However, hand-offs between wireless cells 2 and 3 are not followed by a MIP hand-off because they exist within the same IP network. Whenever a Mobile Node moves it is required to detect its movement by discovering a change in its environment and register its new location (the serving Foreign Agent) with its Home Agent. From this description it is determined that the duration of a MIP hand-off is the sum of the individual processes of: 1. Link-layer hand-off, 2. MIP Movement Detection, 3. MIP Registration. For the purposes of this study, IEEE 802.11b WLAN networks have been assumed. Therefore, in the investigated experimental scenarios, a MIP hand-off is always preceded by an IEEE 802.11b hand-off. As will be shown, 802.11b hand-offs may require significant time intervals to complete. detection delay is equivalent to the time required for the Mobile Node to receive the first agent advertisement upon entering a network. The reaction of the Mobile Node to this event depends on its movement detection method. For a discussion of different MIP movement detection methods, please refer to [4]. The MIP registration involves the exchange of a registration request and reply between the Mobile Node and HA. Its duration is equivalent to the round trip traversal time between the aforementioned entities including processing delays that are platform and implementation specific. In order to reduce the physical distance that is traversed by the MIP registration signalling, further extensions have been proposed that introduce mobility agent hierarchies and enable localised registration management. For more information on hierarchical MIP, please refer to [5]. In this study, the investigated scenarios involve straight line movement patterns. [4] indicates that in such conditions the movement detection method with optimum performance is the Eager Cell Switching (ECS) [6]. The need for MIP movement detection and its impact on MIP hand-offs is discussed in the following section. Movement Detection In the previous sections it was identified that MIP hand-offs always follow an 802.11b hand-off between different IP networks. However, due to layer independence, the event of an 802.11b hand-off can not be communicated to MIP which in turn has to rely on alternative means to determine this information. For this, MIP requires that all Mobility Agents advertise their existence in fixed periodic intervals. A Mobile Node discovers agents by receiving their advertisements. Moreover, a Mobile Node determines its position by evaluating these advertisements. That is, the receipt of a new agent advertisement from an "undiscovered" mobility agent is perceived as an indication of movement into a new network. Similarly, the loss of contact with a "discovered" agent is perceived as an indication of movement out of a network. From this, it can be derived that the movement The ECS movement detection method assumes frequent location changes and therefore dictates immediate MIP hand-offs upon discovering a new Mobility Agent. This assumption is valid for all movement patterns that follow straight line traces or those with a small rate of change in their direction of movement. A finite state machine that describes the operation of ECS is illustrated in figure 3. It can be seen that the Mobile Node proceeds directly to Agent Selection and registration as soon as an unknown agent advertisement is received. However, the time interval between the location switch and the time of receipt of the initial agent advertisement can take on significant values. [1] dictates that the smallest advertisement period may not be smaller than 1 sec. As such, the agent discovery interval may randomly take on values between 0 and 1 sec resulting to an average of 0.5 sec. However, [7] dictates that an optimum agent discovery performance can be achieved by reducing the advertisement period to 0.1 sec. Figure 4 has been acquired through simulations with the Network Simulator 2 (NS) [8]. It illustrates the effect of various MIP agent advertisement periods and sizes on the link throughput of an 914MHz Lucent WaveLAN DSSS radio interface. Marked points illustrate the link throughput capacity for a given advertisement period and size with respect to the Unknown Agent Advertisement Known Agent Advertisement Link Disruption Serving Agent Advertisement 200 Throughput (KB/sec) Agent Selection Serving Agent Lifetime Expires 210 190 180 99% 170 99% 160 97.5% 95% 150 90% 140 Registration Request Registration Unsuccessful Registration Bad Link / Link Layer Hand-off Registration Successful Figure 3 – ECS Finite State Machine 90% 80% 48 112 176 Ad ve rtis 240 em en 304 tS iz e 368 (b y tes 432 ) 95% 75% with respect to throughput for advertisement size 48 bytes and period 1 sec 130 Idle 97.5% 80% 75% 496 0 0.1 0.2 0.8 0.7 0.6 0.5 ds) 0.4 (secon 0.3 t Pe riod isemen Adv ert 0.9 1 Figure 4 – Effect of Advertisement size and period to link throughput Unicast Agent Advertisement Agent Selection Registration Request Serving Agent Lifetime Expires Registration Unsuccessful Registration Link Layer Hint & Broadcast Solicitation Link Disruption Bad Link / Link Layer Hand-off Serving Agent Advertisement Idle Agent Selection Registration Request Serving Agent Lifetime Expires Link Disruption Registration Unsuccessful Registration Bad Link / Link Layer Hand-off Serving Agent Advertisement Idle Registration Successful Link Layer Hint & Registration Request Registration Successful Figure 5 – HCS Finite State Machine Figure 6 – FHCS Finite State Machine smallest possible advertisement size and largest possible advertisement period as defined by [1]. A trade-off between advertisement based movement detection and link utilisation can be observed. As advertisement periods take on smaller values that lead to optimum movement detection, the link throughput drops significantly. proposed to extend the amount of information communicated from the LL to MIP and to include information about the environment such as the identity of the local Mobility Agent. Through this, the need for movement detection as well as the need for agent discovery and selection can be negated. Agent Selection is the state in which advertisements are evaluated and an appropriate agent is selected prior to registration. In popular MIP implementations [9], Agent Selection is based on a local Mobility Agent list that maintains an entry for every discovered Mobility Agent. During Agent Selection, the list is traversed and the agent with the longest outstanding lifetime is selected. As will be experimentally shown, Agent Selection may acquire significant time periods. Proposed Solution From what has been presented, agent advertisement based movement detection methods, such as ECS, are bound to perform with respect to the agent advertisement period while affecting the efficiency of their network. Alternatives can be realised by overruling the fundamental assumption of layer independence. The establishment of a communication between lower layers and MIP for the communication of information such as the completion of LL handoffs can negate the need for agent advertisements as the means of movement detection. [4] introduces Hinted Cell Switching (HCS) as a method based on LL information to perform faster movement detection. The operation of HCS is presented in figure 5. The finite state machine illustrates that as opposed to ECS, where a Mobile Node is required to wait for the elapse of the advertisement period prior to receiving an agent advertisement, HCS enables the node to request for an advertisement upon receiving a LL hint. This is achieved through the broadcast of an agent solicitation that forces all neighbouring Mobility Agents to respond with an immediate unicast advertisement. However, in environments with a large number of roaming nodes that are required to broadcast solicitations with every location change, the effect illustrated in figure 4 may again be encountered. For this reason, it is In the previous section an optimised solution to the MIP problem of movement detection and agent selection was highlighted. Figure 6 illustrates the finite state machine that describes the operation of the proposed solution. It can be observed that through LL information that provides the identity of the local Mobility Agent, the Mobile Node may directly proceed to Registration and bypass the broadcast solicitation as well as the Agent Selection required by HCS. The proposed solution constitutes an improvement of the HCS and therefore has been termed as Fast Hinted Cell Switching (FHCS). In the following sections the experimental set-up of this study is presented, followed by a description of the experiments undertaken and the acquired results. Experimental Set-up For the purposes of this study ECS and FHCS were prototypically implemented. Their performance was compared with respect to the network service disruption introduced by each movement detection method. The latter has been identified with the help of an audio stream that pertains with every MIP hand-off. The experimental testbed is illustrated in figure 7. It consists of three IP networks with mobility support, each one indicated by the presence of a Mobility Agent (Home or Foreign). The Mobile Node has been permanently allocated an IP address from the network where the HA resides. However, the mobile node maintains its point of attachment to the network via one of the available 802.11b Access Points that are connected with each of the Foreign Agents. All Mobility Agents, as well as the router and correspondent node are AMD Thunderbird PCs at 800 MHz. The Mobile Node is a Samsung notebook with an Intel Pentium III at 600 MHz. Wirelined network connectivity has been provided through Fast Ethernet while for the wireless connectivity the IEEE 802.11b compliant Cisco Aironet Access Points and PCMCIA cards that operate at the unlicensed 2.4 GHz ISM band, were used. All machines have been installed with the Linux Mandrake 7.2 distribution and the Linux Kernel 2.2.17. MIP support was provided with the Sun Labs MIPv1.2 implementation [9] that was extended to provide support for ECS and FHCS as they were presented in previous sections. For the ECS experiments the agent advertisement period was set to 1 sec, the shortest value indicated by [1]. During all experimental trials the Mobile Node maintained a single uni-directional audio stream communication originating from the correspondent node. However, all conclusions drawn for this communication should be applicable for bi-directional communications as well as for communications that originate from the Mobile Node. All communications were monitored with the help of the “tcpdump” [10] network analyser software. For the production of the audio stream the RAT [11] audio conferencing software was used that provides support for various audio encoding schemes. For the purposes of this study the popular GSM encoding scheme was chosen. Assuming GSM encoding, RAT produces a 29 kbps stream (encoded audio with header information) that is transported over RTP (Real Time Protocol) in packets of 73 octets length and an average inter-packet delay of 20 ms. It is noted that the RTP protocol is based on UDP Correspondent Node Switch Router Home Agent Foreign Agent Foreign Agent Access Points 802.11 (User Datagram Protocol), but provides facilities such as packet sequence numbers and the RTCP (Real Time Control Protocol) that are important for time sensitive applications. For the performance measurements presented in this study preference was given to UDP over TCP due to the fact that TCP utilises flow control and recovery mechanisms that do not react well when confronted with the service disruption introduced by MIP hand-offs. For more details on TCP performance over MIP please refer to [12]. For each movement detection method a single MIP hand-off between the Foreign Agents was performed. During the hand-off, the Mobile Node maintained a single active audio stream communication without additional background traffic. For each MIP hand-off the length of the respective 802.11b hand-off was measured. In addition, the individual delays of MIP movement detection and registration were identified. Finally, the packet trace of the communication at the time of the hand-off was acquired. The latter was used to determine the extend of packet loss suffered during the MIP hand-off. In order to determine the duration of the 802.11b hand-off, facilities already available from the Cisco Aironet Linux driver were used. More specifically, the publicly available driver provides access to internal information about the status of the network through the Linux “proc” filesystem. In order to realise the inter-layer communication required by FHCS, several information were acquired through the “proc” filesystem. More specifically, the SSID value of 802.11b that is normally used to identify the infrastructure network was used to contain the IP address as well as the MAC address of the local Mobility Agent. Periodically, the MIP software was required to check the value of the SSID through the “proc” filesystem. Any change was interpreted as indication of movement and lead to an immediate response as dictated by FHCS with the information provided by the new SSID. Moreover, through the available information, the status of the wireless network was observed in order to provide a trace of the internal operations undertaken during MIP hand-offs. It has been identified that the operating system provides information about the following four distinct communication states: 1. Connected. This state indicates that the Mobile Node is synchronised on a given Figure 7 – Experimental Testbed frequency as well as maintains an association with the respective 802.11b Access Point. 2. Disconnected. This state indicates that the Mobile Node is not synchronised nor maintains an association. The Mobile Node is scanning the frequencies for an Access Point beacon. 3. Synchronised. This state indicates the the Mobile Node is synchronised to a frequency but does not maintain a valid association yet. 4. State Transition. In this state the Mobile Mode is switching between any of the aforementioned states. During State Transition the operating system enters uninterruptible sleep. The lack of access to the WLAN firmware has hindered a more in depth investigation of the 802.11b behaviour during hand-offs. For this, it is noted that the focus in this study is more on the performance of MIP movement detection methods and less on that of 802.11b. Experimental Results Figures 8 and 9 illustrate the packet traces of two audio stream communications during MIP handoffs with ECS and FHCS support respectively. The linear increase of the packet sequence numbers before and after the MIP hand-off indicate that the unreliable character of the UDP protocol protects communications from unwanted side-effects when suffering periods of extended service disruption. In figure 8, the sum of the MIP movement detection and advertisement evaluation delays takes on values that span to more than half a second. It is noted that this value is in accord with the expectations presented during the description of the ECS in earlier sections. It is further noted that for the ECS trial an agent advertisement period of 1 sec was assumed. The network service disruption introduced by the 802.11b hand-off, in both trials, is surprisingly larger than that of MIP. Through monitoring of the internal hardware state it was possible to determine that the communication suffers initial packet loss, long before any change to the internal state is observed. That is, the Mobile Node is suffering severe packet loss even though it internally remains in the “Connected” state. The time point in which the first state transition occurs has been indicated in the packet trace figures. It is suspected that any packet loss that precedes the state transition is associated with an initialisation of internal buffers during state transition. As a result, any change of state would lead to the loss of packets that were earlier successfully received, stored for processing but never delivered to higher layers. The lack of access to the WLAN firmware hindered further investigation into this matter. However, the performance of 802.11b hand-offs is reserved as the subject of future research. Figure 9 illustrates that the inter-layer communication proposed by FHCS enables the direct delivery of information from the LL to MIP regarding the identity of the local Mobility Agent and therefore negates the need for movement detection and agent selection. This is verified by the fact that the LL hand-off is almost directly succeeded by the registration process. The value of the FHCS movement detection delay was identified to be a little more than a single millisecond, as is presented in figure 11 where the individual delays for each of the processes undertaken during a MIP hand-off are illustrated. It is noted that in order to enable the comparison between the two methods an additional slice was included in the FHCS piechart that indicates the improvement in Mobile IP hand-off with Fast Hinted Cell Switching Mobile IP hand-off with Eager Cell Switching 160 160 GSM Audio Stream GSM Audio Stream 140 Communication Progress (Packet Sequence Number) Communication Progress (Packet Sequence Number) 140 Registration 120 Advertisement Evaluation 100 LL hand-off 80 Movement Detection State Transition 60 40 20 120 Registration 100 80 LL Hand-off State Change 60 40 20 0 0 0 0.5 1 1.5 2 2.5 3 Time (sec) Figure 8 – Packet Trace of MIP hand-off with ECS 0 0.5 1 1.5 2 2.5 3 Time (sec) Figure 9 – Packet Trace of MIP hand-off with FHCS Evaluation Registration 20,001 ms 61,256 ms State Transition 22,394 ms Movement Detection 525,146 ms State Transition 14,808 ms Performance Improvement 772,437 ms Frequency Scan 347,144 ms Frequency Scan 335,192 ms State Transition 23,102 ms Association 2,382 ms State Transition 21,643 ms Figure 10 - MIP hand-off with ECS support Registration 91,4 ms Association 6,566 ms Movement Detection 1,131 ms Figure 11 - MIP hand-off with ECS support performance over its ECS counterpart. Moreover, it is identified that due to lack of access to the hardware firmware it has been unable to establish a direct relation between the communication packet loss during 802.11b hand-offs and the internal hardware state. In the contrary, the MIP hand-off performance realised the expectations presented during the description of the ECS and FHCS movement detection methods. field was utilised to contain the identity of the local mobility agent. Through an experimental investigation that involved a GSM audio stream and Mobile IP hand-offs between two IEEE 802.11b WLAN networks, it was identified that the proposed solution did not require agent advertisements and through Link-Layer information performed almost instantaneous movement detection and agent discovery. Conclusions References Due to lack of communication between the Mobile IP and lower layers where mobility is traditionally managed, Mobile IP movement detection methods have to rely on periodic Mobility Agent advertisements to determine their location. However, optimum movement detection performance require large broadcast advertisement rates. From this a trade-off is made apparent between optimum movement detection performance and the efficient utilisation of the network capacity. Moreover, advertisement based movement detection methods require significant time periods to complete during which the mobile node is unable to send or receive traffic. Through experimental investigation it was identified that the Eager Cell Switching method required more than a half a second in order to perform movement detection with an agent advertisement period of 1 sec. However, when delivering telecommunication services such as telephony over IP, an interruption of that scale may not be acceptable. For this a new movement detection method was introduced, namely Fast Hinted Cell Switching, that relies on the capacity of the LinkLayer to communicate to Mobile IP information about its network environment, such as the identity of the local Mobility Agent. This interlayer communication negates the need for additional Mobile IP mechanisms for movement detection and agent selection and therefore enables faster Mobile IP hand-offs. For the interlayer communication, the IEEE 802.11 SSID [1] European Telework Online, WWW-site (12.12.2000), http://www.eto.org.uk/eito [2] C. E. Perkins, ‘IP Mobility Support’, RFC 2002, Oct. 1996. [3] N.A.Fikouras, K. El Malki, S. R. Cvetkovic, ‘Performance Analysis of Mobile IP Handoffs’, Asia Pacific Microwave Conference, Dec. 1999. [4] N. A. Fikouras, C. Görg, ‘Performance Comparison of Hinted and Advertisement Based Movement Detection Methods for Mobile IP Hand-offs’, 7th European Conference on Fixed Radio Systems and Networks. Sept. 2000 [5] K. El Malki, N. A. Fikouras, S. R. Cvetkovic, ‘Fast Handoff Method for Real-Time Traffic over Scaleable Mobile IP Networks’, draftelmalki-mobileip-fast-handoffs-01.txt, Work in Progress, June 1999 [6] C. E. Perkins, ‘Mobile IP, Design Principles and Practices’, Addison Wesley, (1998). [7] R. Caceres, V. N. Padmanabhan, ‘Fast and Scalable Wireless Handoffs in Support of Mobile Internet Audio’, ACM MONET Journal, Vol. 3 (4) Dec. 1998. [8] NS. UCB/LBNL/VINT Network Simulator ns – (version 2), http://www-mash.cs.berkeley.edu/ns [9] Sunlabs MobileIP, ftp://playground.sun.com/pub/mobile-ip [10] Tcpdump, WWW-site (5.3.2001), http://www.tcpdump.org [11] RAT. UCL Robust Audio Tool, http://wwwmice.cs.ucl.ac.uk /multimedia/software/