Assessment and optimization of schemes for tracking and routing to mobile users in packet-based networks

Abstract

Although a lot of progress has been made in recent years, supporting mobility in the Internet is still a difficult challenge. The Internet today consists of multiple networks, interconnected by IP routers. In each network multiple subnetworks may coexist, also connected via routers. One subnetwork may connect multiple LAN segments by means of MAC switches. For scalability purposes, IP addresses are hierarchical, which allows routing decisions to be performed using address aggregates. Due to this addressing structure, the IP address of a mobile user needs to change dynamically as the user moves from one subnetwork to another (or one network to another) to match addressing at the new subnet (or the new network). Unlike IP addresses, MAC addresses are flat, as the scale of the problem is smaller. Switches inside a subnetwork track individual users and forward packets to them.

If a source sends packets to a destination that is moving and changing IP addresses, the contents of the packet may refer to a destination address that is stale, and the packet may be discarded, or redirected by some special-purpose entity in the network known as an agent. The source may be informed of the change in the IP address of the destination as well, so that future packets can be sent directly to its new location. As users move inside a subnetwork, changes need to take place at one or more switches tracking that user, in an attempt to maintain connectivity to the user at all times.

Tracking updates may take a long time to arrive at the agents, end-hosts and switches, which can result in a temporary loss of connectivity to the user. This becomes particularly noticeable when users are engaged in streaming multimedia applications, and may even result in the abortion of TCP sessions. Furthermore, current technologies significantly limit the number of individual addresses that can be tracked by a device built at a reasonable cost. Therefore, any solution to support mobility must deal with inherent delays caused by distances between the moving user and entities in the network that track and assist the user, as well as limitations of the current technology in terms of cost and performance.

In this paper we assess, design and optimize schemes to support mobility in the Internet. These schemes exploit techniques called address-lookahead, packet n-casting, transparent learning and light-weight explicit registration. Via numerical simulation, we demonstrate considerable improvements in the user-perceived quality of applications, at no significant increase in cost. For example, 40% more voice calls can experience a high level of quality, at the cost of only a 3% increase in the bandwidth consumption in the Internet.

Introduction

In the broadest sense, the term mobile networking refers to a networking technology that allows users to maintain uninterrupted end-to-end connectivity while moving (at any speed) over a geographical area (building, campus, metropolis or even the entire world). Such technology is often associated with wireless communication.

Mobile networking already exists today for voice communication services offered with cellular telephony, however, it relies on circuit-switching, which, in general, is appropriate only for handling voice traffic. With the growing interest in providing data communication services and access to the Internet to wireless mobile users, there is a need to seek a packet-based mobile networking solution to support any communication service. Indeed, packet switching and the Internet protocol are quickly becoming the networking standards towards which all networks are converging, including those used by the traditional telecommunication sector, as less appropriate circuit-switched technologies are abandoned. Interest in having mobile users access the Internet is evidenced by the increase in cellular phones with built-in Internet access, PDAs and laptops with wireless interfaces and car navigation systems with Internet connectivity.

Although a lot of progress has been made towards supporting mobility in the Internet, this problem still poses difficult challenges. To understand these challenges, let us first review some basics of Internet networking.

The Internet today consists of multiple networks, interconnected by IP routers. In each network multiple subnetworks may coexist, also connected via routers. One subnetwork may connect multiple LAN segments by means of MAC switches. For scalability purposes, IP addresses are hierarchical, which allows routing decisions to be performed using address aggregates. Due to this addressing structure, the IP address of a mobile user needs to change dynamically as the user moves from one subnetwork to another (or one network to another) to match addressing at the new subnet (or the new network). This type of movement is generally referred to as macro-mobility. Unlike IP addresses, MAC addresses are flat, as the scale of the problem is smaller. Switches inside a subnetwork track individual users in order to route packets to the appropriate LAN segment for delivery. User tracking takes place using the transparent learning protocol, which allows switches to learn about the location of users implicitly, by listening to the traffic the users generate. Movement inside such a network with a flat address space is usually referred to a micro-mobility.

In the rest of this section, we will describe the challenges and approaches for supporting macro-mobility and micro-mobility respectively. We will then describe one final approach to supporting macro-mobility that relies on a network infrastructure built using the techniques for micro-mobility support proposed in this paper.

To handle macro-mobility, one or more entities in the Internet are required to track changes in the IP address of a mobile user, so that packets can be properly forwarded to the new address. For example, Mobile IP exploits IP agents that redirect any traffic sent to the old IP address of a mobile user to its new address [12]. Binding updates may be sent to each corresponding host, CH, of a mobile, to inform them of the new address [12], [20]. Future packets sent by these hosts will be directed to the new address of the mobile user.

Unfortunately, the acquisition of a new IP address can take a very long time (on the order of hundreds of milliseconds [21]). Moreover, updating the tracking entities can take time, due to propagation delays in the Internet. Consequently a mobile can experience blackouts at times of movement, during which no packets can be received. This can result in packet loss, unless packet buffering is performed in the Internet to assist the user. However, this technique can lead to increased packet delay, which may result in loss of interactivity, affecting the overall user-perceived quality of interactive applications. Handover delay can even affect reliable applications such as those employing the TCP–IP protocol. Blackouts may lead to packet re-transmissions, however, there is a maximum number of re-transmissions allowed, after which the TCP sessions must be aborted.

Given these delay overheads, it becomes important that the network act in advance of user movement (instead of reacting to user movement), to enable the mobile host to acquire the new IP address ahead of time. Indeed, various proposals have been made in the context of Mobile IP to allow a mobile host to perform address lookahead while still at the old subnet, by exploiting hints from the lower layers [10], [11]. Moreover, packet n-casting may be performed at the agents to allow traffic to be received at any of the IP addresses associated with the user at a given time [13].

Apart from Mobile IP, a number of solutions exist to support macro-mobility [20], [23], but these also suffer from the same delay impairments caused by the address acquisition process and propagation delays in the Internet. Hence in our work, we propose that address lookahead and packet n-casting be also exploited in the context of these schemes for enhanced user-perceived quality of Internet applications.

Thus, in the first part of this paper we assess the contribution of techniques known as address lookahead and packet n-casting to optimizing the performance of various schemes for supporting macro-mobility in the Internet. We show that the reliability of TCP/IP applications can increase to almost 100%, while up to 40% more voice calls experience a high level of quality.

Over the past decade, we have witnessed tremendous developments in LAN technolog ies, such as increases in switch processing by a few orders of magnitude, and increases in link bandwidth and distances covered (owing to the fiber optics technology). These advances resulted in an increase in the size of LANs, and more recently, the deployment of such technologies in metropolitan areas (MANs). When users roam inside MANs of this type, they may depart one LAN segment and join another. Tracking information at one or multiple switches in the subnetwork can become stale due to this movement. Hence, a packet that is received at the switch for the mobile user will be sent to the wrong LAN segment, where it will be discarded. This can result in a temporary loss of connectivity to the user, which becomes particularly noticeable when users are engaged in streaming multimedia applications.

To prevent packets form being mis-routed or even lost, packet flooding may be employed, however at the expense of increased bandwidth consumption in the network. Another choice is to employ lightweight control-based mechanisms for user tracking, however this can lead to larger databases at the switches. Scalability problems can arise as current technologies significantly limit the number of individual addresses that can be tracked by a switch built at a reasonable cost.

Thus, in the second part of this paper, we propose to devise optimal schemes for tracking users in networks that handle user micro-mobility. We show that the transparent learning protocol, currently used in switched LANs, cannot be used efficiently to track fast moving users. Additional tracking mechanisms are necessary, such as based on explicit control messaging. Furthermore, we show that, when implementing these mechanisms, the selection of parameters can drastically affect performance, and thus must be done carefully to take into account the design constraints, as well as the application and user mobility characteristics.

We have seen how macro-mobility handoffs can be slow because they involve detecting that a new subnet is reached, acquiring a new address at that subnet and propagating this information to some entities in the network. To mitigate these problems we use anticipation techniques at the lower layers that permit address lookahead and packet n-casting to take place at the higher layers. Unfortunately, in some networks anticipation cannot be implemented. For example, in wireless LANs, handoffs are hard and forward, which means that connectivity with the existing wireless cell must be discontinued prior to establishing connectivity at the new cell. For these networks, we need to investigate a different method to optimize support for macro-mobility.

To this end, in the third part of this paper, we investigate the idea that a mobile user maintain a fixed IP address while roaming inside a large region that spans many IP subnets from before. This is equivalent to building a large network to support micro-mobility, within which switches track users through their IP or MAC address. Using numerical simulation, we show that this technique can result in the elimination of voice calls with poor quality.

To make building such a network feasible, one needs to address the scalability aspect of extended LANs and their potential deployment in large metropolitan areas to serve large populations of mobile users. In our prior work [2] we show that the scalability of such infrastructures can be obtained by partitioning the switches in the infrastructure into subsets that play different roles with respect to mobile hosts and the hosts communicating with them. More specifically, we limit the set of switches in the infrastructure that track a mobile user. To reach such a mobile host, packets originating at a communicating host are broadcast over a set of switches which intersect with the tracking subset. Using numerical simulation, we show that by using network partitioning, the traffic load at the switches can be reduced by more than 50%.

In conclusion, in this paper we address the problem of supporting micro-mobility and macro-mobility in the Internet, with the goal of improving the user-perceiv ed quality of Internet applications for mobile users. Specifically, we assess, design and optimize various schemes for tracking mobile users that account for the state of technology, and are optimized for a variety of user mobility rates and application characteristics.