For a wireless mobile multimedia network to be implemented, an underlying architecture must be in place to support the stringent demands of users and applications. In today's networks, there is one layer in the protocol stack which is becoming the most ubiquitous and an obvious piece to the overall puzzle. That piece is the Internet Protocol (IP) of the TCP/IP protocol suite.
IP provides the addressing and routing services needed in a TCP/IP based packet network. IP is ubiquitous in the packet switching arena and hence, it may make good business sense to extend its capabilities to support mobile hosts with the needs of voice, video, data and public Internet access. In fact, much work has been completed to define the necessary architecture via the IETF Mobile-IP working group 2. A number of Internet Request for Comments (RFCs) and Internet Drafts (I-Ds) have been written and proposed to help further the development and deployment of mobile IP computers and mobile IP enabled networks.
Although still very much a work in progress, mobile IP for data is well underway. Using mobile IP for multimedia such as voice and video, however, is still a relatively new terrain. A number of issues crop up in being able to support multimedia applications while mobile, particularly where real-time applications such as two-way voice conversations are concerned.
To understand the characteristics of the Mobile IP specifications, let us review what it means to be mobile.
First, being mobile means that your physical location is not fixed. That is, you may access the network from multiple, disparate places. Distances spanned while being mobile may be only a few feet or hundreds of miles.
Second, being mobile means that you can also "roam". To roam is the ability to stay connected to the network as you move. This may have several implications on the design of the network infrastructure. For example, how do you supply coverage for the roaming area? Do you install multiple base stations or maintain one high capacity base station for all users in the coverage area? How will you handle "hand-off" if a roaming mobile moves to a new base station? This last issue can supply sufficiently enough headaches for a mobile IP implementation alone. We will try to address some of the major issues shortly.Third, to be mobile may also mean you are often out of service. If you move into an area where there is no coverage, your network connection will be lost. Where most of today's IP computers are directly connected via a reliable LAN cable or phone line, mobile computers may experience significant accessibility problems with their physical medium and network provider due to the very nature of mobile computing technologies.
The requirement for Mobile IP comes from the limitations of traditional IP addressing and IP routing. Each host in an IP network maintains a unique IP address. This unique IP address not only uniquely identifies a host, but it also is used to find a path to this host from others in the IP network. The concept of a network and host portion of an IP address would assume that a mobile IP computer would either have to receive a new address when it moves between IP subnets or that it needs to supply new routing information for the network on which it now resides. Although either solution may be workable in small IP networks, higher level protocols may refuse to function properly and it would eventually be difficult to scale in a large network such as the public Internet.
Mobile IP must then provide for a few new mechanisms in the traditional IP network if it is to support mobile computing. Two new fundamental concepts for a Mobile IP network are the "home agent" and a "foreign agent".
A home agent may be nothing more than a forwarding mechanism, keeping track of where the mobile computer currently resides in the network. The mobile computer may connect to the home network for the majority of connect time over a long period, or it may not. In either case, people who are looking for the mobile computer would probably start there. There must be something at the home base, which can intercept and forward requests. Typically this would be a router, firewall or gateway type of device. This is the role of the home agent.
The foreign agent is assigned to the mobile computer when it is away from its home base. The foreign agent maintains a "care-of" address, which is used by the home agent. Packets destined for the mobile computer's home address are intercepted by the home agent and then "tunneled" to the foreign agent. The foreign agent then forwards the packets to the mobile computer for which it is providing services. The care-of address is considered temporary and changes in a mobile computer's location should be expected. A care-of address can be a temporary IP address assigned to the mobile computer on the new network or it may be assigned to a distinctly separate foreign agent, who is arbitrating the mobile computer's access to the rest of the network.
If the mobile computer receives its own care-of address, all packets will be sent via the home base directly to this new address. If a separate foreign agent is used, the home base will forward packets to the agent, who is then responsible for forwarding packets on to the mobile computer.
In both scenarios, the home agent encapsulates the original IP packet from the sender into a new IP packet, which has the IP destination address of either the foreign agent or the actual mobile computer acting as its own foreign agent.
When the mobile computer responds to the sender, it does so using traditional IP routing instead of tunneling back towards its home agent. A mobile computer responds directly to its peer in order to minimize delay and avoid the overhead of going through the home base on the return path. The diagram below should help make the process clear.
Figure 1 assumes that the mobile computer does not receive a new IP address on the visiting network. In fact, it could do so, which would allow the home base to forward encapsulated packets directly to the mobile computer. This would allow network managers to support IP mobility on their networks that do not yet contain foreign agents, but do allow for the dynamic allocation of IP addresses to visiting hosts using a mechanism such as DHCP. The drawback with this scenario is the added complexity that is required in the mobile computer as well as a "double-consumption" of available IP address space.
Although we are leaving out many details of the Mobile IP specification, we now have a groundwork for which to discuss the applicability of using mobile IP as a mechanism with which to deliver new services such as data, voice and video in a mobile computing environment.
What happens when you take traditional real-time and multimedia applications and adapt them to a Mobile IP environment? Is there going to be enough bandwidth in the mobile networks that are traditionally more limited than their stationary counterparts? What effect does a roaming mobile computer have on multimedia and real-time applications? Is bandwidth going to be available for each network a mobile computer roams to? Are class or quality of service mechanisms required? How would they be implemented?
Let us examine some of the challenges with these mobile applications and see how the Mobile IP specifications are being used to address them. Details of practical applications using Mobile IP are discussed in other sections of this document.
Perhaps the greatest concern for Mobile IP applications that comes to mind is that of performance. When talking about performance, we typically talk about response-time, throughput and bandwidth. Today's mobile networks are often much more constrained than their stationary companions, therefore we may need to make some adjustments either to the applications, the network and/or user expectations. As we all know, these factors are always demanding more from technology. The call for acceptable performance on a mobile network then, should be at least tolerable compared to stationary networks with regards to the application being used.
Let's assume for a moment, however, that a Mobile IP network is not constrained with regards to bandwidth and throughput. Let's imagine that a Mobile IP home agent is located in Chicago, IL with the Mobile IP computer located in the same city, but on a different IP network a few miles away.
To get a rough idea of response time, we simulated performance with a series of IP PING tests between DePaul University, Northwestern University and University of Chicago. We observed maximum response times in the range of 55 to 65 milliseconds with good response times of around 25 to 35 milliseconds. Approximately ten router hops were traversed along each path.
We'll pretend we have a Mobile IP computer on the DePaul network and its agent is located at Northwestern. Performance depends on where the Mobile IP computer's peer is located. If we assume the peer is at Northwestern, the communication between the home agent and the peer is relatively instantaneous. (see figure 2) This is because the peer communicates first with the home agent at Northwestern, where the transmit time will be very good. The home agent then communicates with the Mobile IP computer at DePaul. Only the minimal overhead of first sending to the home agent coupled with the necessary processing time to encapsulate and forward packets will degrade performance.
If we move that peer from Northwestern to the DePaul network, where the mobile computer is located, and leave the home agent at Northwestern, what differences can we see in performance? (see figure 3) Packets from the peer now have to first travel all the way back to Northwestern to the home agent, then be forwarded back to the Mobile IP computer at DePaul. However, response packets from the mobile computer to the peer are sent directly to the peer with minimal delay over the local DePaul network.
If the bulk of the conversation is symmetrical, either scenario will perform equally well. However, if the majority of data is sent from the peer to the mobile computer, as in a file download, response time will be almost double what it would have been if the peer was located on the Northwestern network.
As a final example, we now move the peer to the University of Chicago network. (see figure 4) In this case, the home agent, mobile computer and peer are all on different networks. The "triangular routing" effect presented in this example exemplifies the performance concerns one may encounter in a Mobile IP network. Imagine a conversation between two Mobile IP computers and these concerns are only exacerbated.
By extending the distances between the mobile computer, the home agent and the peer, you begin to see some potential performance issues. These issues can translate into serious problems for users of real-time or bandwidth intensive applications.
One may be able to simply bypass the transmission path of the peer to the home agent and go directly to where the mobile computer is located. This has been discussed extensively in the IETF. Based on our research 3, this performance issue was a major design concern for the Mobile IP working group of the IETF. However, other problems were more pressing, such as privacy, authentication and complexity, which had prevented it from entering into the initial specifications4. Recently, cellular network providers talking about a move into the Mobile IP arena with their networks have made this "routing optimization" issue one that is receiving more attention today.5
In the world of IP, maintaining mobility is not trivial. As Johnson and Maltz point out:
High bandwidth connectivity for mobile computers will likely be limited for some time to come. For this reason, operating efficiently at all layers in the protocol stack is of utmost importance. In addition to route optimization, there are a number of techniques that can help make IP overhead less of a burden. Administrative messages, protocol overhead, and processing time are all areas where IP can be tuned to improve performance.7
Administrative messages are those that are used to help manage the mobile computer, the home agent and the foreign agent in a Mobile IP network. These messages are used to register mobile computers with the home agent and foreign agent. For example, when a mobile computer travels to a distant network, it cannot just connect to the new network, it must first swap messages with a local foreign agent or somehow receive a new care-of IP address. Then the mobile computer must exchange information with the home agent. These messages are required not only for initialization, but may also be required for authorization purposes as well.
Sending arbitrary messages on LANs and traditional high-speed networks has historically been of little concern since bandwidth was plentiful and it was an easy thing to do. Broadcasting routing information periodically is a good example of arbitrary protocol overhead. It isn't the most efficient way to exchange routing information, but it's an easy method with minimal impact on network efficiency.
In Mobile IP, where bandwidth is almost certainly constrained, care is taken to reduce unnecessary transmission of bits. Compression, prediction routines, caching and packet sizes are carefully implemented to minimize data transmissions to and from mobile computers.
Processing time involves not only computation of information by communicating nodes in a Mobile IP network, but also the processing time that is required to maintain and track mobile computers. Take for example a mobile computer that roams from network to network rather rapidly. It may be difficult for the home agent, foreign agent and other communicating nodes to keep up with the moving mobile computer. A mobile computer that moves too quickly for the underlying network protocol and application layer protocols could be problematic.
Even if the routing and bandwidth problems are overcome, what about the roaming problem? The IP roaming problem is the affect of a mobile computer, using IP, moving to another IP network in the middle of a connection. This is currently a problem not so much for IP itself, but for higher layer protocols and applications.
When two IP hosts setup a connection using TCP and one of the hosts is a mobile computer using Mobile IP, what happens to the TCP session when the mobile computer moves to a new IP network? How will the existing TCP session be maintained? Will TCP session information remain intact?
There are mechanisms for nodes that communicate with mobile computers to learn of the new mobile computer IP address, but this only solves the IP addressing and routing problem. One of the goals of the Mobile IP solution is to avoid having hosts know whether they are communicating with a mobile IP computer or not. How do you inform a peer, in the middle of a UDP data stream, that you are now using a different IP address, and to send future messages to the new address.
The answer to this problem may be to avoid the question. Perhaps for a mobile computer to properly maintain connectivity at protocol layers higher in the stack than IP, you simply do not allow the IP address to change. This can and has been achieved using traditional wireless networks.8 By implementing a care-of agent at the border of a physical mobile computing network, the roaming mobile computer can always use the same IP address within the confines of the underlying mobile network. For example, if you wanted to use Mobile IP over a CDPD network, you simply use one CDPD network provider who can hand-off at the data-link layer, but maintain a consistent network layer address throughout their entire network. This may make IP networking for mobile network providers difficult, but it may be the easiest way to fix the TCP and application level problems with roaming.
However, Mobile IP gets around this problem by tunneling. Since a Mobile IP computer always maintains its permanent home IP address, all messages are sent to this address. Either at the home agent, or if route optimization is used at the peer, the applications are still communicating with one IP address and sending one stream of IP packets. These IP packets are encapsulated as necessary and provide for seamless connectivity of applications above the IP layer. Applications never know they are talking to different IP addresses as a Mobile IP computer roams.
It has been largely foreseen that mobile computing devices will become more pervasive, more useful and more powerful in the future. The power and usefulness will come from being able to extend and integrate the functionality of all types of communication such as web browsing, e-mail, phone calls, information retrieval, and perhaps even video transmission.
For Mobile IP computing to become as pervasive as stationary IP networks of the world, the ubiquitous protocol for the integration of voice, video and data must be developed. The most widely researched and developed is Mobile IP. However, the challenge of supporting multiple formats of data even in the Internet today is still troublesome at best. In order for this support to work in a mobile computing network, much work needs to be done. This final section will briefly describe some of the challenges that lie ahead.
Perhaps the underlying problem with using IP for mobile computing in the first place is in the way the architecture is fundamentally designed? In comparing the Internet and IP with traditional telephone networks, a primary difference is in where the intelligence is located. For instance, when is the last time you performed an upgrade on your telephone? Sure you may have bought a new phone, but for the most part, a phone is a phone is a phone. The same cannot be said of IP end stations. For changes in the network to occur, an IP network requires the end stations to change. Just the opposite is true in a telephony network. For example, when cellular phones were introduced, no changes were required in stationary, wired phones.9
Regardless of the problem, something needs to be fixed or retrofitted for mobile computing over IP to work. There are some interesting "mobile middleware" technologies that may provide a peek into the future solutions. Queuing, caching, recovery, minimized session overhead and re-connection strategies can all be used to help alleviate mobile IP problems.10 In fact, there is work in IPv6 and TCP committees which are developing techniques to more effectively support Mobile IP computing.
The Mobile IP community has not yet even started to talk about class or quality of service issues, which are just entering development and deployment in stationary IP networks today.
Perhaps the real driving force for the success of Mobile IP will be IPv6. IPv6 provides most of the necessary requirements that are needed for mobile computing. It has proven more difficult to modify existing IPv4 stacks to support Mobile IP than was originally thought. 11
Mobile computing with Mobile IP is a topic that is surely going to receive more attention over the coming millennium. With so little in the way of research and test networks today, it remains to be seen how far and how quickly Mobile IP can take us. With the popularity of IP based applications, there are sure to be plenty forward-thinkers and early investors willing to try and make it happen.
Goto John Kristoff's Home Page
Last updated: October 31, 1999