workflow management in a business, and reduce or even eliminate the cost of moving ... mobile end-points is being developed at AT&T Bell Laboratories. ... Cellular telephone networks have extended the domain of telephone service over a.
Multimedia information processing in the SWAN mobile networked computing system Prathima Agrawal Eoin Hyden Paul Krzyzanowski Mani B. Srivastava John Trotter AT&T Bell Laboratories Networked Computing Research Department 600 Mountain Avenue, Murray Hill, NJ 07974
ABSTRACT Anytime anywhere wireless access to databases, such as medical and inventory records, can simplify workflow management in a business, and reduce or even eliminate the cost of moving paper documents. Moreover, continual progress in wireless access technology promises to provide per-user bandwidths of the order of a few Mbps, at least in indoor environments. When combined with the emerging high-speed integrated service wired networks, it enables ubiquitous and tetherless access to and processing of multimedia information by mobile users. To leverage on this synergy an indoor wireless network based on room-sized cells and multimedia mobile end-points is being developed at AT&T Bell Laboratories. This research network, called SWAN (Seamless Wireless ATM Networking), allows users carrying multimedia end-points such as PDAs, laptops, and portable multimedia terminals, to seamlessly roam while accessing multimedia data streams from the wired backbone network. A distinguishing feature of the SWAN network is its use of end-to-end ATM connectivity as opposed to the connectionless mobile-IP connectivity used by present day wireless data LANs. This choice allows the wireless resource in a cell to be intelligently allocated amongst various ATM virtual circuits according to their quality of service requirements. But an efficient implementation of ATM in a wireless environment requires a proper mobile network architecture. In particular, the wireless link and medium-access layers need to be cognizant of the ATM traffic, while the ATM layers need to be cognizant of the mobility enabled by the wireless layers. This paper presents an overview of SWAN’s network architecture, briefly discusses the issues in making ATM mobile and wireless, and describes initial multimedia applications for SWAN.
1. INTRODUCTION Wireless access technology promises to provide mobile users with ubiquitous and tetherless access to information. So far this promise has largely been realized in two specific domains: cellular telephony and wireless data networks. Cellular telephone networks have extended the domain of telephone service over a wireless last hop, while wireless data networks such as WaveLAN 18 for local area, Metrocom’s Ricochet for metropolitan area, and McCaw’s CDPD-based AirData for wide area do the same for users of TCP/IP data networks. Continual progress in wireless technology will, however, soon make it possible to economically
provide per user data rates of several Mbps, at least inside buildings. Already, the readily available radio modems operating in the 2.4GHz industrial, scientific, and medical (ISM) band offer 1 to 2 Mbps per channel data rates. When operated in a spatially-multiplexed indoor pico-cellular environments, a wireless network based on these radios can support a reasonable user density. Such wireless bandwidths, together with the emerging integrated service wired network infrastructure, will permit seamless delivery of multimedia information to a mobile user, at least indoors. Progress in packaging, display, and low-power circuits 8 has resulted in portable multimedia end-points 5 that seamlessly integrate into a user’s networked computing environment. To explore the synergy between multimedia information access, wireless access technology, and portable multimedia end-points, a mobile computing environment called SWAN is being developed in the Networked Computing Research Department at the AT&T Bell Laboratories. SWAN, which stands for Seamless Wireless ATM Networking, seeks to provide continual network connection to mobile heterogeneous ATM end-points in an indoor setting. The choice of end-to-end ATM was driven by our conviction that it is logical to extend ATM’s virtual circuit (VC) model over the wireless last hop, given that it appears increasingly likely that the core of future multimedia networked computing environments will be based on ATM cell switching. End-to-end ATM connectivity, of course, requires solutions to lower level problems such as VC connection and service quality management in the presence of mobility. SWAN’s solutions to these lower level problems have been presented elsewhere 12 . The focus of this paper is on the influence of the interplay between ATM and wireless on the higher level network model of SWAN. We also describe the application of SWAN for multimedia transmission and retrieval.
2. NETWORK COMMUNICATION ARCHITECTURE OF SWAN Figure 1 shows a high level view of the network communication model adopted by the SWAN 2 . A hierarchy of wide-area and local-area wired ATM networks is used as the back-bone network, while wireless access is used in the last hop to mobile hosts. In addition to connecting conventional wired server hosts and client end-points, the wired backbone also connects to special switching nodes called basestations. The basestations are equipped with custom wireless adapter cards, and act as gateways for communication between nearby mobile hosts (which are also equipped with wireless adapters) and the wired network. Our adapter cards, called FAWN 17 , provide a flexible interface to the wireless link. The geographical area for which a basestation acts as the gateway is a room-sized pico-cell. Network connectivity is continually maintained as users carrying a variety of mobile hosts roam from one cell to another. The mobile hosts themselves range from portable computers equipped with a suitable wireless adapter, to dumb wireless terminals 4 that have no or little local general-purpose computing resources. All mobile hosts in SWAN, however, must have the ability to participate in network signalling and data transfer protocols. A mobile in SWAN sends and receives all its traffic through the basestation in its current cell. SWAN uses end-to-end ATM, over both the wired network and the wireless last hops. This is in contrast to the use of connectionless mobile-IP in present day wireless data LANs. This design choice in SWAN was motivated by advances in video compression algorithms and availability of higher bandwidth RF transceivers which permit the transmission of packetized video to a mobile. Support for multimedia traffic over the wireless segment was a driving force in SWAN. Adopting the connection-oriented paradigm of ATM virtual circuits over
AT&T
BS
Portable Multimedia Terminals and PDAs
SWAN Media Server SW SW’
Wide Area ATM Fabric
SWAN Wireless ATM LAN/PBX
SWAN
BS
BS
BS
BS
BS
Wired Host AT&T
Personal Voice/Data Multimedia Embedded Intelligent Terminal Peripherals and Appliances Communicators
Figure 1: Network Communication Model of the SWAN Wireless ATM Network the wireless hop allows quality of service guarantees associated with virtual circuits carrying multimedia traffic to be extended end-to-end. Extending ATM’s virtual circuit paradigm all the way through to a mobile host necessitates continuous rerouting of ATM virtual circuits as a mobile host moves. The small cell sizes and the presence of quality of service sensitive multimedia traffic make this problem particularly important in SWAN. Virtual circuits carrying audio or video, as far as possible, need to be immune from disruptions as a mobile host hands-off from one basestation to a neighboring one. ATM signalling protocols need to accomplish the task of virtual circuit rerouting with minimum latency. Details of a signalling protocol for SWAN can be found in a separate paper 12 . Lower level protocol layers for wireless medium access, and for mobile hand-off from one basestation to another, must present minimal latency. The wireless hop in SWAN enables low latency hand-off, allocation of wireless resources to virtual circuits, medium access control (MAC), and air interface operation.
3. PROBLEMS IN MAKING ATM WIRELESS AND MOBILE A question that naturally arises in SWAN is what are the issues in implementing ATM in a wireless and mobile environment. After all, ATM was designed for an environment where the hosts do not move, and the transmission medium is a relatively error free and high speed point-to-point wired link. However, in a SWAN like environment, the hosts move and the transmission medium is a relatively noise and burst error prone shared medium of a modest bandwidth. The change from a wired static world to a wireless mobile world can have unexpected pitfalls as the experience of various researchers with TCP performance in a mobile-IP wireless environment has shown - packet loss in the wireless hops due to noise, burst error, or hand-off is falsely interpreted by TCP as being due to congestion, leading to a poor performance. Clearly, it is not obvious what
enhancements are needed for ATM to work well in a mobile and wireless setting. Understanding this research problem with a real system is in fact a raison d’etre for the SWAN system. Before describing the problems in ATM that need to be managed for successful use of ATM in a mobile and wireless network, it must be mentioned that there are strong positive reasons for why ATM is a good choice for such a network. First, fine-grained multiplexing as provided by ATM is well suited to slow speed links. Second, the virtual circuits (VC) of ATM provide a useful traffic flow id handle to use at the wireless link level. For example, the MAC layer can use the VC ids to meaningfully allocate and schedule the shared wireless channel resources. MAC layers in current IP or IPX based wireless data LANs cannot do such intelligent wireless resource allocation. Similarly, the link level error control mechanism can be suitably adapted on a per VC basis depending on the characteristic of the individual VC. Third, ATM’s notion of specifying per-VC quality of service (QoS) is quite useful in a wireless and mobile network carrying multimedia traffic. QoS specification can be used, for example, to give some connections higher priority during hand-offs, or to use rerouting policies tailored to the traffic characteristics. In short, wireless and mobile ATM suffers neither from the rigid synchronous structure of cellular voice and PCS systems, nor from the inherently best-effort delivery model of wireless data LANs. In addition to the reasons mentioned here, there is also the obvious reason that ATM is emerging as a scalable high-performance core of the wired multimedia networks. The issues that need to be successfully addressed in making ATM work in a SWAN-like network fall into two somewhat orthogonal categories: mobility related problems at the higher level, and wireless related problems at the lower level. Our specific solutions are presented elsewhere 3 , but here we summarize the problems themselves. From a mobility perspective, the key ATM issue is that of virtual circuit management in the presence of host mobility. Obviously, the VC route needs to be continually modified as the hosts at either end move during the lifetime of a connection. The rerouting must be done fast enough so as to cause minimal disruption to the transport layer. The signalling protocols for VC establishment, rerouting, and tear-down must function properly even in the presence of host mobility at both ends during the signalling phase itself. Any rerouting must maintain the sequence of ATM cells at the end-points. Clearly, there would appear to be “obvious” solutions to some of these problems from analogous problems in the connection-oriented cellular and PCS world. However, the scale of the problem here is different in many dimensions, rendering the naive solutions too inefficient. The hand-off frequency is much larger due to small cell size - imagine a person strolling down a hallway. Each terminal can terminate a large number of VCs, leading to a large number of VC reroutes on each hand-off. The statistical multiplexing inherent in ATM and multimedia traffic requires a QoS renegotiation at the new basestation, compared to the relatively simple frequency or time slot allocation in cellular and PCS basestations. Although wireless is a last hop issue, end-to-end ATM requires various components of the last hop to be aware of ATM traffic. From a wireless perspective, the key ATM issues are providing lower layer support for ATM QoS, and efficient transport of ATM over a slow noisy air medium. The first issue, QoS support, arises from the need for the MAC subsystem to participate in admission control at the time of VC admission, and renegotiation at the time of hand-off. Usually, MAC protocols are targeted at coordinating access of the shared air resource among multiple mobiles. With ATM, the MAC protocols need to schedule the air resource among multiple VCs at multiple mobiles. The second issue, efficient transport of ATM over the air, arises from the high header-to-payload ratio of ATM cells in an already low bandwidth air medium, and the high wireless link noise and burst error rate. To address the former, one can use header compression techniques between the basestation and the mobile, such as transparent replacement of 24-bit VPI/VCI by smaller wireless link connection ids, or by clustering multiple cells on the same VC into a MAC frame. To address the latter, link level error control
schemes are needed. Here, ATM VCs offer the advantage that link level error control schemes can be selected appropriately for each VC. Finally, besides the specific mobility and wireless issues, there is a third related problem of what services API should a mobile and wireless ATM network offer to native mode ATM applications. Current ATM APIs are tailored for a static wired environment, and only offer basic services for VC establishment, tear-down, and data read/write. For a SWAN-like network, the API ought to be able to hide wireless and mobility from applications that wish to operate transparently, and to send wireless and mobility related events to applications that wish to adapt to changing network conditions. For the latter class of mobility and wireless adaptive applications, the lower layers of the network, such as MAC, must allow the higher layers to register interest in specific low level events, such as an imminent soft-handoff, and be able to send events back to the higher layers and the application when the specified low level event occurs.
4. RELATED WORK Until now wireless and mobile networks have been restricted to cellular telephone networks carrying voice traffic, or to wireless data LANs. Recently however researchers have begun to seriously consider the possibility of delivering multimedia traffic in a wireless and mobile environment. For example, research projects such as Berkeley’s InfoPad 5 and UCLA’s WAMIS 9 , seek to provide a mobile user with audio, video, graphics, and data connectivity. When compared to such competing projects a distinctive aspect of SWAN is clearly the use of ATM, a choice which not only produces a homogeneous end-to-end network simplifying the architecture, but also has attributes such as QoS specifiable VCs which are quite useful in a wireless and mobile environment. Among other wireless multimedia systems, for example, InfoPad uses a TCP/IP backbone with a separate custom wireless hop transport mechanism between “dumb” terminals and basestations, while UCLA’s WAMIS, which uses no basestations, supports both TCP/IP and virtual circuits. Closer to SWAN’s ATM model, various researchers have recently addressed specific problems in using ATM in a wireless and mobile network. Raychoudhuri 14 looked at how a multi-service PCS network architecture might be based on ATM. Rajagopalan presented architectural options for integrating wireless access and mobility to existing ATM networks. At a lower level, Keeton 11 , Acampora 1 , Biswas 6 , and Toh 16 have addressed the problem of VC rerouting as a result of hand-offs. Indeed, Toh’s 16 measurements at Cambridge show that VC rerouting, when properly implemented, can indeed be quite fast. Chandler 7 described an air-interface for ATM cell transport in a CDMA wireless network. SWAN’s solutions to such low level problems have already been published elsewhere 12,15 , and form the algorithmic basis of our system implementation
5. SYSTEM DESIGN OF SWAN: HARDWARE AND SOFTWARE An initial system design of SWAN is complete and functional. Its focus is the local-area domain and the wireless last hop. The wireless last hop consists of basestations and mobiles. SWAN uses room-sized cells, each of which has a basestation equipped with one or more radio ports. The primary function of the basestation is to switch cells among various wired and wireless ATM adapters attached to the basestation. Generic PCs and Sun workstations are used as basestations by plugging in a wired ATM adapter card and one or more RF wireless ATM adapter cards. The mobiles, at the other end of the wireless hop, include portable computers with an adjunct ATM wireless adapter or multimedia terminals with embedded ATM wireless adapter.
From a hardware perspective, the key idea behind SWAN’s wireless last hop is the use of a single reusable ATM wireless adapter that interfaces to one or more digital-in digital-out radio transceivers on one side, and to one or more standard busses (PCMCIA, in our implementation) on the other side. The adapter, called FAWN 17 , provides flexible data processing resources in the form of FPGAs and a software-programmable embedded ARM610 RISC processor. The radio transceiver that is currently supported is an off-the-shelf 2.4GHz ISM band slow frequency hopping radio, although any radio providing multiple communication channels will suffice. The flexible and reusable FAWN adapter card provides a uniform mechanism for making devices “SWAN-ready”. The software view of the wireless hop is shown in Figure 2. On the basestations the three key modules are: (i) a kernel-resident cell routing and adapter interface module, (ii) a user-space-resident connection manager (CM) module, and (iii) a wireless adapter-resident medium access control (MAC) module. The first module primarily switches ATM cells amongst the various wired and wireless ATM adapters on the base stations; the second module participates in ATM level signalling to establish, tear-down, and reroute ATM VCs; and, the third module allocates and schedules wireless resources in response to signalling messages from the CM module, and implements low-level medium access control and inter-cell hand-offs by the mobile end-points. The use of PCs and workstations for basestations allows them to act as wired hosts as well, running application processes in addition to the three modules mentioned above. In essence, basestations in SWAN are computers equipped with banks of radios. The software view at mobile end-points is similar, except that a mobile end-point has only one ATM adapter (FAWN) and the connection manager, cell routing, and MAC modules at the mobile have somewhat different functionalities. While, from the hardware and software perspective, a basestation can have an arbitrarily large bank of radios, there are limits imposed by the spectrum allocation. There are a total of 83 frequency slots of 1 MHz bandwidth, and the radios hop among these slots according to FCC mandated rules. The current radios in SWAN give a
Connection Manager
Applications ATM Cells, or AAL Packets
ATM Cells
Applications ATM Cells, or AAL Packets
ATM Cell Routing, AAL, and Adapter Interfaces
RF ATM Adapter (FAWN)
ATM Cells
BASESTATION
ATM Cells
ATM Cell Routing, AAL, and Adapter Interface
ATM Cells
Wired ATM Adapter
Connection Manager
ATM Cells
Link Cells or Air Interface Packet
Figure 2: The Wireless Last Hop in SWAN
RF ATM Adapter (FAWN) MOBILE
nominal data rate of 625 Kbps in each 1 MHz slot. Each hopping sequence can be viewed as a communication channel, and these sequences must be sufficiently orthogonal to minimize collisions. It can be shown that the aggregate network capacity for co-located asynchronously-hopping communication channels is maximized when there are 83/4, or about 21, such channels. The particular hopping sequence family that we use in SWAN actually allow for a total of 22 weakly orthogonal sequences, which we then distribute (statically) among the basestation radio ports such that two radio ports with identical sequences are sufficiently far apart. In typical configurations SWAN basestations have up to 3-4 radio ports, each corresponding to 625 Kbps channel. Given the moderate per channel bandwidth of our radios, and our interest in multimedia applications, we typically operate with one mobile per channel in the current system. However, the MAC module also allows for multiple mobiles to share one channel, thus allowing a large number of low bandwidth users in a room. It must be emphasized that SWAN project has strived to take a radio-independent view of the system - even the reconfigurable wireless ATM adapter hardware was designed to be able to use different radios as its link interface. Any radio with a serial data-in data-out interface, and providing multiple software-controllable channels that can be spatially multiplexed, can be used in SWAN with a little effort. The choice of current radio was dictated to a large extent by pragmatic reasons, such as the availability of the radio in an embeddable card form as opposed to a black-box. As better embeddable radios become available to us, we will consider integrating them into SWAN. For example, in the next 1-2 years radios operating in the 5.2 Ghz ISM band are expected to become available, and would make natural candidates for a SWAN-like system. These radios will provide much higher per channel bandwidth of a few tens of Mbps, making it meaningful to multiplex several multimedia users on one channel. Indeed, one can even envision a SWAN system composed of heterogeneous radios providing a diversity of bandwidth, cell size, and access cost attributes, with the reconfigurable FAWN adapter providing a standard mechanism to integrate the radios into SWAN.
6. EXPERIENCE WITH AUDIO AND VIDEO TRANSMISSION IN SWAN Although SWAM is an on-going project, an initial version of the SWAN system is now operational. There are two modes of application level communication. The first mode provides end-to-end native mode ATM connectivity to applications running on mobile hosts. The ATM virtual circuits terminate at a mobile host after going across a wired ATM cloud composed of Fore’s ATM switches, a SWAN basestation, and a wireless ATM link. ATM connection manager processes at the basestations and the mobiles provide the signalling support for VC establishment and rerouting, and together with the MAC software embedded on the FAWN adapters implement a bandwidth based VC admission control at the time of connection set-up and hand-off. We have done preliminary experimentation with the ATM performance monitoring utility netperf and a version of video player nv ported to SWAN’s ATM API. In this paper we focus on the second mode of SWAN where TCP applications are run by using IP/ATM segmentation-reassembly modules at the mobiles and the basestations, together with IP forwarding at the basestations, and a reserved VC carrying IP traffic. In effect, this mode uses IP-over-wireless-ATM with the mobility being handled at the IP-level (via mobile-IP under Linux). Pending the development and availability of a richer suite of native mode ATM applications, much of our actual use of SWAN so far has been in this mode with conventional TCP/IP based multimedia (and other) applications such as nv, vat, and xmosaic, as the initial drivers of SWAN.
Nemesis Frontend Connection Manager
IP Packets
ATM Cells IP/ATM ATM ATM Cell Routing Segmentation + & Reassembly Cells Adapter Interfaces IP Packets Wired N/W Adapter
nv (video)
vat (audio) Connection Manager
IP/ATM Segmentation & Reassembly
ATM Cell Routing + Adapter Interfaces
ATM Cells RF ATM Adapter (FAWN)
Basestation (P5-90 / Linux) @ Murray Hill, NJ
ATM Cells
ATM Cells
Link Cells or Air Interface Packet
RF ATM Adapter (FAWN) Mobile (NEC Versa / Linux)
AT&T’s TCP/IP Network
Camera
Multimedia Server for Nemesis @ Holmdel, NJ
Microphone
SGI Indigo with Camera and Microphone for Live Audio/Video
Figure 3: Using SWAN for Transmission of Multimedia Streams We have experimented with several scenarios of multimedia information transmission in SWAN using the IPover-ATM mechanism. Figure 3 depicts some of the possibilities. In one such application the mobile user sees and hears live video and audio feed, being transmitted from a camera and microphone equipped SGI Indigo. The popular internet audio- and video-conferencing tools, vat and nv, are used in a host-to-host mode. The live feed can also be replaced by pre-recorded video streams, such as movie and cartoon clips, transmitted by nv_play from a remote host. In a related experiment, a mobile user can use vat alone to conduct a voice conversation with another mobile user, or with a user on a wired host. This is a nascent step towards our eventual goal of fully subsuming a multimedia PBX-like functionality in SWAN. Another application scenario that we have experimented with is to let a mobile users play, in real-time, video and audio streams being fetched using the Nemesis 1 0 multimedia service from a multimedia archives server containing audio, MJPEG video, and accompanying documents and viewgraphs of talks given at AT&T. Nemesis uses adaptive rate-control so that a
reasonable video quality (10-15 frames per second) is obtained at the mobile over the wireless link. The JPEG decompression is done in software at the mobile. The mobiles in the above experiments are FAWN-equipped NEC VERSA and AT&T Safari portable computers with built-in audio I/O and running the Linux OS. The wireless link has a raw bandwidth of 625 Kbps per mobile, which is divided equally into about 300 Kbps in each direction by the simple MAC module used in the IP-over-ATM experiments. The reliable TCP-over-wireless-ATM throughput measured in these experiments was about 227 Kbps in each direction, for a mobile in the same room as the basestation, and with radios operating at 10mW radiated (RF) power. While raw performance numbers are important, perhaps equally crucial is the subjective feel to a user. Even with the modest per-user bandwidth of our radios, the video and audio quality in the current implementation is reasonable for viewing stored talks and other video clips. This is particularly true with rate-adaptive applications such as Nemesis. The performance will only improve as we migrate to better radios. An open issue however is that of usage pattern - for example, whether the users will indeed watch a video stream while actively moving. The sheer physical bulk of typical PC laptops make it hard to watch a video clip while walking down a hallway. Therefore we do not yet have much realistic feel of such usage. However, simpler multimedia mobile terminal designed for on-the-move use are an attractive option. One such mobile terminal, called the Personal Multimedia Terminal (PMT) 4 , is also being integrated into the SWAN system by our co-researchers.
7. SUMMARY The SWAN project is exploring the seamless delivery of multimedia information to mobile users roaming in an indoor setting. This is being accomplished by a synergistic exploitation of wireless access and ATM virtual circuits. The first generation system is largely functional. Using IP delivery over ATM, we already have initial experience with live as well as stored video and audio transmission using nv and vat, and also with access to a multimedia archives server via Nemesis. We have also implemented end-to-end native mode ATM transport in SWAN. Our experiment with SWAN help highlights the strengths as well as the weaknesses of using wireless and mobile ATM in delivering multimedia traffic when compared to the IP-based wireless LANs or TDM-based PCS networks. The paper also described the research challenges encountered in making ATM wireless and mobile. While it is perhaps too early to tell whether ATM is overall the right choice for mobile and wireless networks, it is nevertheless clear that ATM offers quite important technical advantages in such networks.
8. ACKNOWLEDGMENTS The authors wish to than Abhaya Asthana, Mark Cravatts, Partho Mishra, and B. Narendran for their help and contributions to various aspects of the SWAN system.
9. REFERENCES 1. A. Acampora and M. Naghshineh. “An Architecture and Methodology for Mobile-Executed Handoff in Cellular ATM networks.” IEEE Journal of Selected Area in Communications, vol. 12, no. 8, October 1994. 2. P. Agrawal, et. al. “A Testbed for Mobile Networked Computing.” In Proceedings of 1995 IEEE International Conference
on Communications (ICC ‘95), pp. 410-416, June 1995. 3. P. Agrawal, P. Mishra, and M. Srivastava. “Network Architecture for Mobile and Wireless ATM.” Private communications, to be published. 4. A. Asthana, M. Cravatts, and P. Krzyzanowski. “An Indoor Wireless System for Personalized Shopping Assistance.” In Proceedings of IEEE Workshop on Mobile Computing Systems and Applications, December 1994. 5. B. Barringer, et. al. “Infopad: A System Design for Portable Multimedia Access.” Wireless 1994, pp. 382-396, Calgary, Canada, July 1994. 6. S. Biswas and A. Hopper. “A connection management scheme for a mobile radio LAN.” In Proceedings of the IEEE International Conference on Personal Wireless Communications, bangalore, India, August 1994. 7. D. P. Chandler et. al. “An ATM-CDMA Air Interface for Mobile Personal Communications.” In Proceedings of PIMRC’94, 1994. 8. A. P. Chandrakasan, and R. W. Brodersen. Low Power Digital CMOS Design, Kluwer Academic Publishers, 1995. 9. R. Jain, et. al. “PC-notebook based mobile networking: Algorithms, Architectures, and Implementation.” In Proceedings of 1995 IEEE International Conference on Communications (ICC ‘95), June 1995. 10. H. P. Katseff, and B. S. Robinson. “Predictive Prefetch in the Nemesis Multimedia Information Service.” In Proceedings of ACM Multimedia 1994, San Francisco, October 1994. 11. K. Keeton et. al. “Providing Connection-Oriented Services to Mobile Hosts.” In proceedings of the USENIX Symposium on Mobile and Location-Independent Computing, pages 83-102, Cambridge, Massachusetts, August 1993. 12. P. P. Mishra, and M. B. Srivastava. “Call Establishment and Rerouting in Mobile Computing Networks.” AT&T Bell Laboratories Technical Memorandum 11384-940906-13TM, September 1994. 13. B. Rajagopalan. “Mobility Management in Integrated Wireless-ATM Networks.” In Proceedings of the First International Conference on Mobile Computing and Networking (MobiCom ‘95), pp. 127-135, November 1995. 14. D. Raychaudhuri and N. D. Wilson. “ATM-Based Transport Architecture for Multiservices Wireless Personal Communication Networks.” IEEE Journal on Selected Areas in Communications, vol. 12, no. 9, pp. 1401-1414, October 1994. 15. M. B. Srivastava. “Medium Access Control and Air-Interface Subsystem for an Indoor Wireless ATM Network,” In Proceedings of the Ninth International Conference on VLSI Design, Bangalore, India, January 1996. 16. C-K Toh. “The Design and Implementation of a Hybrid Handover Protocol for Multi-Media Wireless LANs.” In Proceedings of the First International Conference on Mobile Computing and Networking, pp. 49-61, November 1995. 17. J. Trotter, and M. Cravatts. “A Wireless Adapter Architecture for Mobile Computing.” In Proceedings of 2nd USENIX Symposium on Mobile and Location Independent Computing, pp. 25-31, April 1994. 18. B. Tuch. “Development of WaveLAN, an ISM Band Wireless LAN.” AT&T Technical Journal, pp. 27-37, July/August 1993.
