Providing Mobile LAN Access Capability for Devices Shih-Yen Chiu artis@os.nctu.edu.tw Hsung-Pin Chang hpchang@cis.nctu.edu.tw Department of Computer and Information Science National Chiao Tung University Hsinchu, Taiwan, ROC Ruei-Chuan Chang rc@cc.nctu.edu.tw Abstract In this paper, we design and implement a system with mobile LAN access capability. First, by adding a virtual network interface card (NIC) driver between IP and protocol stack, users can use their devices to access the LAN resources via the access points. Besides, to accommodate with the mobility characteristics in network, we also propose a new handoff scheme. Instead of passively waiting for a handoff indication by the expiration of link supervision timer, in this paper, we detect a handoff actively and promptly by measuring the RSSI (Receive Signal Strength Indication) value. In addition, our proposed handoff scheme eliminates the inquiry procedure and only requires paging procedure during a handoff. We have implemented a prototype of system with mobile LAN access capability by our proposed scheme. Experimented results show that the handoff duration is significant reduced than the normal scheme. 1. Introduction With the tremendous progress of personal mobile communication technologies, wireless communication devices are becoming smaller and cheaper to be portable and affordable for a wide range of end users. is such a wireless communication standard that provides cheap but powerful communication technology [4-5]. As stated in [4], network is intended following the vision of a truly low-cost, low-power, single chip radio-based cable replacement. Therefore, the specification is proposed to get ride of the cumbersome cable connections and configuration procedures to facilitate on-demand connectivity among devices. Nevertheless, devices would also be connected to local area networks (LAN) via access point. Thus, clients not only get wireless access to the information provided locally but even access to the resource resided in the internet, such as printing, file storage etc. In this paper, we design and implement a system with mobile LAN access capability. Although a LAN access profile is provided in specification [5], however, a RFCOMM layer must be implemented to emulate a serial line on top of protocol stack. Then, the Point-to-Point protocol (PPP) is performed over RFCOMM to allow TCP/IP operating on the network. In other words, the cable replacement scenario and dialup network mechanism are combined to provide with LAN access capability. However, this would result in an extra overhead since the introduction of two protocol layers, PPP and RFCOMM. Besides, the dialup network cannot deal with roaming as the link of PPP is assumed static. As the mobility characteristics of devices, users may move between different access points. Therefore, how to minimize the handoff duration poses a challenge in the design of a system. As mentioned above, LAN access profile does not take the user mobility into consideration since, in dialup network, the link is assumed fixed. Thus, in [2], the authors proposed a handoff approach that uses role switch scheme to minimize the handoff duration. Nevertheless, in their proposed scheme, the handoff duration involves inquiry and paging procedures and, unfortunately, the inquiry procedure is very time consuming. Besides, they rely on the expiration of supervision timer to detect a possible handoff. However, it is difficult to select a suitable supervision timer value. Furthermore, their system is built on an emulation system that the link is emulated over an Ethernet connection. In this paper, we first design and implement a system with LAN access capability. Instead of using the LAN access profile, we insert a virtual NIC driver between IP and protocol stack (). Thus, the device can pretends itself as an Ethernet adaptor to the upper layer protocols and applications. Consequently, legacy applications can access the LAN resources over devices without any modification. Besides, to accommodate with the mobility characteristics of users, a new handoff scheme is proposed that only demands paging procedure during handoff. As a result, handoff duration can be significantly reduced. In addition, in previous approach, they reply on the expiration of link supervision timer to detect a possible handoff. However, it is not easy to select a suitable value of link supervision timer. In this paper,
instead of passively waiting the handoff procedure to be invoked by the timeout of supervision timer, we make use of the RSSI (Receive Signal Strength Indication) mechanism to actively and promptly detect a handoff once the link strength is detected weak. The remainder of this paper is organized as follows. In Section 2, we give a brief introduction of technology. Section 3 reviews the related work. The implementation of virtual NIC driver and proposed handoff scheme are presented in Section 4. Section 5 gives the experimental results. Finally, Section 6 summarizes this paper and provides future work. 2. Overview is a low-power, low-cost and short-range wireless communication technology in the 2.4 GHz ISM (Industrial, Scientific and Medical) RF band. uses FHSS (Frequency Hopping Spread Spectrum) scheme with hoping rate of 1,600 hops per second to minimize the effects of signal interference. The transmission range is 10 meters and can be extended up to 100 meters by providing a power amplifier. can offer a speed up to 1Mbps. To identify the identity of a device, each device has a 48-bit BD ( Device) address, which has the same length as the MAC (Media Access Control) address of IEEE 802.x family. Communication between devices follows a strict master-slave scheme. Each master device can have up to 7 active slaves and forms a so-called piconet. Between each master-slave pair, two different links can be provided. One is the SCO (Synchronous Connection Oriented) and the other is the ACL (Asynchronous Connectionless Link) link. The SCO link is typically used for voice communication and ACL link is used for data communication. For ACL links, a slave can transmit packets only after the master sends a packet addressed to it. Note that, slaves cannot transmit packets directly, i.e., the communication between slaves must go through the master device indirectly. Further introduction of technology can be seen in [3, 8-9]. 2.1. Specification The 1.1 specification, which is released in February 2001, consists of two parts: core and profiles. 2.1.1. Core Specification. Figure 1 shows the protocol stack. The RF (Radio Frequency) defines the physical characteristics of the RF link, e.g., channel arrangement, permissible transmit power levels, and receiver sensitivity level. The baseband specification defines the device discovery, link formation, and synchronous and asynchronous communication with peer host. To provide a reliable wireless link, fast ARQ (Automatic Repeat Request), CRC (Cyclic Redundancy Check) and FEC (Forward Error Correction) are combined with the frequency hopping scheme in Baseband to detect and resolve packet errors or loses during transmission. A pplication RFCOMM Audio Data SDP HCI Link M anager Basband RF Control LM P Figure 1. Protocol Stack The Link Manager uses the LMP (Link Manager Protocol) to manage the physical link by negotiating and configuring the parameters of a physical link. The (Logical Link Control and Adaption Protocol) is responsible for providing logical links to the upper layer protocols. The tasks it deals with include SAR (Segmentation and Reassembly), protocol multiplexing and negotiation of QoS (Quality of Services) parameters. HCI (Host Command Interface) is the interfaceindependent command interface between hosts and devices. Thus, the software stack on the host can communicate with the hardware via the HCI commands. However, if all of the protocols are included in a single chip, e.g. the control chip of a wireless keyboard, the can directly control the Link Manager to manage the physical link and HCI layer is unnecessary. The RFCOMM protocol emulates a serial line and thus legacy protocols and applications that use the COM port for communication can use the devices with no or little modification. 2.1.2. Profile Specification. After the specification of core network protocol stack, a variety of applications can then uses the services offered by the stack to provide devices with diverse functionality and capacity. For interoperability issues, SIG thus defines 13 profile specifications to facilitate interoperability [5]. In this section, we focus on the LAN access profile.
ME Application TCP/UDP PPP IP SDP RFCOMM LMP Baseband LMP PPP Network PPP SDP RFCOMM Baseband ME LAN Figure 2. LAN access Profile Application TCP/UDP IP LAN As stated above, uses RFCOMM to emulate a serial line, i.e., RS-232 cable connection, on top of protocol stack to allow legacy applications continuous operating over the devices. LAN access profile utilizes this feature and then introduces PPP to run on top of RFCOMM since PPP expects a serial line interface from the lower layer. Figure 2 shows the protocol stack in the LAN access profile. The left is a host, the middle is a access point and the right is a machine in the local area network. As PPP provides a packet-oriented interface to the upper layer, TCP/IP protocol stacks can be supported on top of PPP. In other words, TCP/IP protocol stack utilizes the underlying PPP and RFCOMM as the link and physical layers for transmission. Notably, RFCOMM just emulate a serial line while the actual physical transmission is via the wireless network. Thus, LAN access profile incorporates the dialup network with the protocol stack to achieve LAC access capability. 2.2 Buletooth Discovery and Connection Procedure The Baseband Specification defines point-to-point connection establishment as a two-step procedures [4]. First, the inquiry procedure is used to discover other devices. Then the paging procedure is subsequently used to synchronize these two devices. Once the paging process is completed, the device moves to the connection state. The inquiry procedure is an asymmetric process; it involves two types of nodes each performing different actions. We call these two types of nodes as sender and receiver. To find a device, the sender will move into the inquiry state that generates an inquiry hopping sequence and broadcasts inquiry messages. Notably, the hopping sequence is based on the local device s clock and the chosen inquiry access code. In contrast, to be discoverable from sender, receivers will periodically enter the inquiry scan state that hop according to the inquiry scan hopping sequence, which is also derived from the local clock and inquiry access code. Once the receiver performing the inquiry scan receives the inquiry message, it replies with a response message after a random backoff to avoid collision from other devices. The response message includes the receiver s BD address and internal clock setting, which is used by the following paging procedure. Similar to the inquiry procedure, paging procedure is also an asymmetric process. The sender enters the paging state while the receiver enters the paging scan mode. Different from the inquiry procedure, the page message is unicasted to the receiver since the receiver s BD address and clock setting has been obtained from the inquiry procedure. In contrast, the sender must broadcast the inquiry message. As a result, the reply to a page message does not require a random backoff delay. After the completion of paging procedure, these two devices are synchronized and move to the connection state. As stated in the Bleutooth specification, the inquiry state may have to last for 10.24 seconds unless the inquiry collects enough response and determines to abort the inquiry state earlier [4]. The formula of how to derive the value of 10.24 can be seen in [4, 13]. In order words, in an error-free environment, inquiry procedure would at most spend 10.24 seconds to discover the desired devices, which is unacceptable during handoff. In contrast, if the paging procedure is activated immediately after the inquiry procedure, which is the normal case during connection establishment, the duration of a paging procedure is around 1~2 second [4, 13]. Thus, compared with the inquiry procedure, the duration of paging procedure is quite short. 3. Related Work In 1999, Albrecht et al. proposed the mobility issues on OSI layer 3 for a network [1]. In the proposed scheme, they use Cellular IP [12] to solve the micro mobility problem and Mobile IP for macro mobility. In 2000, they further proposed the solution to the layer 2 mobility [2]. When a device wants to anticipate a piconet, it performs the inquiry and paging procedure to find the access point and synchronize with it. On the contrary, the access point has to periodically enter inquiry scan mode and page scan mode to respond to the new anticipated device. Therefore, during the establishment of a connection, a mobile host acts as a master while access points act as slaves (Here, we assume the sender is a master and receiver is a slave). This is because that the inquiry and paging states take a substantial amount of time compared to the inquiry scan and paging scan states and,
during this time, none of the slave devices can send or receive data. Thus, if the access point acts as a master, this would result in a long silence time interval that none of any slave host can send/receive packets. After the connection is established, a role switch is performed that the mobile host becomes a slave and the access point becomes the master to coordinate the communication in its piconet. However, in their proposed scheme, hosts will face a substantial amount of link broken time since the inquiry procedure is very time consuming. Besides, they rely on the link supervision timer to detect a possible handoff. Once the timer is expired, they assume a handoff is occurred. Nevertheless, it is difficult to determine a suitable value for the timer. A too small value would tear down a connection too early while a too large value would regard a broken link as still alive and increase the handoff duration. Since device discovery is a time-intensive phase of the connection establishment procedure, in [13], the authors proposed the use of IrDA (Infrared Data Association) to accelerate the connection establishment. In their proposed scheme, an IrDA connection is first established between two devices both equipped with IrDA capability. Then, the device discovery information is exchanged via the established IrDA connection. As a result, the device can bypass the time-intensive device discovery procedure. However, IrDA has its limitation in that it is short distance and requires line-of-sight between devices. Besides, both access point and devices must both equip with IrDA capability. 4. Design and Implementation In this section, we present the design and implementation of providing mobile LAN access capability for devices. Section 4.1 introduces the mechanism to provide LAN access capability for users. In Section 4.2, the proposed handoff algorithm is presented. 4.1. Providing LAN Access Capability for Devices To reduce the extra overhead of protocol processing in PPP and RFCOMM, in this paper, a more efficient approach is adopted that runs the TCP/IP protocol stack directly on top of protocol stack, which is similar to the solution in [2]. By adding a virtual network interface card (NIC) driver, which is called BTH ( Ethernet) in the paper, between IP and, TCP/IP BTH(MH) HCI Linux Protocol Stack Virtual NIC Driver Protocol Stack TCP/IP BTH(AP) HCI Figure 3. The proposed protocol stack for LAN access capability the IP packets can then be sent/received by the device. Figure 3 shows the integrated protocol stack. BTH thus hides the specific details from TCP/IP and pretends itself as an Ethernet device driver, while the actual communication is via the network. Therefore, IP packets are sent to the BTH and BTH uses the logical link provided by to transmit these packets. According to the different roles between devices and access points, the function of BTH between device and access point is distinct and is called as BTH(MH) and BTH(AP) respectively. For BTH(MH), once a packet is received, instead of triggering a interrupt, the will call the callback function which is registered before by the BTH(MH) to indicate the BTH(MH) the arrival of a new packet. Then, the BTH(MH) calls the receive function provided by to read the packet. If a packet is sent from upper layer, the BTH(MH) just call the send function provided by to transmit this packet. For BTH(AP), besides the function provided in BTH(MH), it must read the address field and decide how to forward this packet. If the packet is destined to a machine in LAN, the access point will forward the packet to the local area network. In contrast, if the packet is addressed to one of the devices, the access point forwards this packet to the corresponding device. If the packet is addressed to the access point itself, the packet is processed as described in BTH(MH). As a result, our approach gives a more efficient way to provide LAN access capability for device. 4.2. Handoff Scheme Before describing the proposed handoff algorithm, we first disclose the composition of handoff duration in a network. The duration of a handoff can be split into three phases:
I. Detect a broken link that indicates a possible handoff. II. Search for a new access point to reconnect by the inquiry procedure. III. Synchronize to the new access point by paging procedure. Paging procedure is necessary for synchronization with the new access point. Besides, as stated in Section 2, the duration of paging procedure is quite short compared with inquiry procedure. Thus, we strive to eliminate the time to detect a broken link and the inquiry procedure. is not intended to behave as a cellular system that provides fast and seamless handoff scheme. There is no beacon mechanism that can tell mobile clients in which cell they currently reside. Thus, in [2], a device recognizes a possible handoff passively if a link is detected broken by the expiration of link supervision timer. However, it is difficult to select a suitable value for link supervision timer. A too small value will cause a short silence to be considered as a handoff and perform the unnecessary paging and inquiry procedures. In contrast, a too large value will result in a late response to a handoff. provides a RSSI mechanism for determining the current link s signal strength. Thus, instead of passive waiting for the link supervision timer expires, we detect the handoff actively by periodically querying the RSSI value of a link in the access point. Once the value is smaller than a predefined threshold, an ongoing handoff is then detected. Besides, inquiry procedure is time consuming and, if it can be omitted, handoff duration can be significantly reduced. Figure 4 shows the steps of a handoff. Once a handoff is detected by access point, it disconnects the link with the mobile host (1). Then, this access point broadcasts the BD address of the host to the neighboring access points (2). After that, the neighboring access points tries to reconnect to the roaming host by the paging procedure (3).Once a paging succeeds, this access point will inform other access points to stop their paging procedures. (1) Access Point 1 (2) Mobile Device (3) Access Point 2 5. Experimental Results In this section, the performance of the system with LAN access capability is evaluated. The experimental platform is shown in Table 1. Both the access point and mobile host run on the Linux Kernel 2.2.19. Besides, the Linux Kernel of access point is patched with Linux Ethernet Bridging for forwarding functionality. 5.1. Handoff Duration The handoff time has a significant impact to the protocols (e.g. TCP/IP) and user applications above protocol stack [6]. A shorter handoff time can result in a better system performance. The handoff duration is measured from the disconnection of a link to the reconnection of the link. We ignore the time of broadcast handoff messages from one access point to the other access point since it incurred very little overhead. The average handoff time is 2.583 seconds. 5.2. TCP Performance in the Network Second, we measure the TCP performance in the Network by Netperf benchmark [10-11]. Figure 5 shows the TCP processing performance (transaction/second) between a device and a access point. The y-axis represents that ratio of send packet size to response packet size. In Figure 6, we measure the response time of ping application. Although the handoff time is 2.583 in average, however, the TCP will encounter a further degraded performance. This is because that the slow start and exponential backoff design methodology of TCP. Thus, researchers are strives to propose schemes to improve TCP performance during handoff. Some literatures addressed this issue can be seen in [6-7]. Hardware Specificaiton Access Point 1 Access Point 2 Table 1. The experimental platform. CPU Ethernet Network Adaptor Celeron 566 (RAM 128 MB) Pentium 120 (RAM 64MB) Mobile Host Pentium II 300 (RAM 32MB) Ne2000 Compatible Network Adapter Ne2000 Compatible Network Adapter N.A. Module Ericsson Module(RS232) Ericsson Module(RS232) Ericsson Module(RS232)
Figure 5. Throughput improvement under different number of random requests Figure 6. Throughput improvement under different number of sequential requests 6. Conclusions and Future Work In this paper, we design and implement a system with mobile LAN access capability. First, to resolve the drawbacks of LAN access profile, we implement a virtual NIC driver to provide users with LAN access capability. Then, a handoff scheme is proposed that eliminates the inquiry procedure, which is the most time consuming procedure during handoff. Besides, instead of using the supervision timer to detect a possible handoff, the proposed handoff scheme uses the RSSI mechanism to detect an ongoing handoff actively and promptly. However, in our proposed scheme, the paging procedure is performed in the access point and the data transmission activity is ceasing during this duration since the access point acts as a master in its piconet. In contrast, if a module having the role switch capability is available, the paging procedure would be performed in the mobile stations as proposed in [2]. Nevertheless, the Ericsson module does not have the role switch ability. Besides, once a handoff is detected by an access point, all the neighboring access points must perform paging procedure to page the roaming mobile client. However, since paging requires nearly all of a device s bandwidth, the access points are not able to serve any other client during this time. As a result, the planning of access points should avoid one access point is directly connected by many other access points. Since a master can only support up to seven active slaves, to support more active slaves, a access point (acting as a master) can equip with more transceiver. However, as the FHSS scheme adopted in the Blutooth, these transceiver would interfere with each other. Our feature work would derive a hopping scheme that eliminates the interference between adjacent transceiver. Acknowledge We would like to thank Da-Wai Chang, Ming-Xian Jiang and Rong-Jyh Kang for the help of implementation of protocol stack. References [1] M. Albrcht, M. Frank, P. Martini, M. Schetelig, A. Vilavaara, A. Wenzel, IP Services over : Leading the Way to a New Mobility, Proceeding of the 24 th Conference on Local Computer Networks, Lowell, MA, October 1999, pp. 2-11. [2] Simon Baatz, et. al., Handoff Support for Mobility with IP over, Proceeding of the 25 th Conference on Local Computer Networks, November 2000, pp. 143-154. [3] Chatschik Bisdikian, An Overview of the Wireless Technology, IEEE Communications Magazine, Dec. 2001. [4] SIG, Specification of the System Core, 2001. [5] SIG, Specification of the System Profiles, 2001. [6] K. Brown, S. Singh, M-TCP: TCP for Mobile Cellular Networks, ACM Computer Communications Review, Vol. 27, No. 5, 1997, pp. 19-43. [7] R. Caceres, L. Iftode, Improving the Performance of Reliable Transport Protocols in Mobile Computing Environments, IEEE Journal on Selected Area in Communications, Vol. 13, No. 5, June 1995, pp. 850-857. [8] Jaap C. Haartsen, The Radio System, IEEE Personal Communicaitons, Vol. 7, No. 1, February 2000, pp. 28-36. [9] J. Hartsen et al., : Visions, Goals, and Architecture, ACM Computer Communications Review, Vol. 4, No. 2, 1998, pp. 38-45. [10] Hewlett-Packard Company, Netperf: A Network Performance Benchmark, February, 1995. [11] http://www.netperf.org/ [12] A. G. Valko, Cellular IP: A New Approach to Internet Host Mobility, Computer Communication Review, Vol. 29, No. 1, January 1999, pp. 50-65. [13] Ryan W. Woodings, Derek D. Joos, Trevor Clifton, Charles D. Knutson, Rapid Heterogeneous Ad Hoc Connection Establishment: Accelerating Inquiry Using IrDA,