Host Mobility Using an Internet Indirection Infrastructure Shelley Zhuang Kevin Lai Ion Stoica Randy Katz Scott Shenker∗ University of California, Berkeley fshelleyz, laik, istoica, [email protected] Abstract mobile host to have seamless connectivity and continu- ous reachability, it must retain its identifier while chang- We propose the Robust Overlay Architecture for Mobil- ing its location. Previous mobility proposals decouple ity (ROAM) to provide seamless mobility for Internet this binding by introducing a fixed indirection point (e.g., hosts. ROAM is built on top of the Internet Indirection Mobile IP [1]), redirecting through the DNS (e.g., TCP Infrastructure (i3). With i3, instead of explicitly sending Migrate [2]), or using indirection at the link layer (e.g., a packet to a destination, each packet is associated with cellular mobility schemes). an identifier. This identifier defines an indirection point in i3, and is used by the receiver to obtain the packet. However, these proposals lack one or more of the follow- ing properties to fully realize the promise of ubiquitous ROAM takes advantage of end-host ability to control mobility: the placement of indirection points in i3 to provide ef- ficient routing, fast handoff, and preserve location pri- Efficient routing: packets should be routed on paths vacy for mobile hosts. In addition, ROAM allows end • with latency close to the shortest path provided by hosts to move simultaneously, and is as robust as the un- IP routing. derlying IP network to node failure. We have developed a user-level prototype system on Linux that provides • Efficient handoff : the loss of packets during handoff transparent mobility without modifying applications or should be minimized and avoided, if possible. the TCP/IP protocol stack. Simulation results show that ROAM's latency can be as low as 0.25-40% of Mobile • Fault tolerance: communication between mobile IP. Experimental results show that with soft handoff the hosts should not be more vulnerable to faults than TCP throughput decreases only by 6% when there are as communication between stationary hosts. many as 0.25 handoffs per second. • Location privacy: the host's topological location should not be revealed to other end-hosts. 1 Introduction • Simultaneous mobility: end hosts should be able to While the wired Internet reaches many homes and busi- move simultaneously without breaking an ongoing nesses, the wireless Internet has the potential to not session between them. just reach, but encompass all the spaces that people use to live, work, and travel. Wireless data services (e.g., • Personal/session mobility: a user should be able to 802.11b, GPRS, 3G cellular) will soon provide the po- redirect a new session or migrate an active one from tential for ubiquitous, though heterogeneous, coverage. one application or device to another one when a bet- To realize this potential, users will want both seamless ter choice becomes available [3, 4, 5]. connectivity (flows uninterrupted by mobility) and con- tinuous reachability (the ability of other hosts to contact • Link layer independence: users should be able to the user's host despite mobility). These services would seamlessly operate across heterogeneous link layer enable users to run applications such as IP telephony, in- technologies, not all of which support the same link stant messaging, and audio streaming while mobile. layer mobility scheme (e.g., GSM mobility). Unfortunately, the standard Internet cannot provide these In this paper, we propose (to the best of our knowledge) services. The fundamental problem is that the Internet the first solution to achieve all of these properties. Our uses IP addresses to combine the notion of unique host solution, called Robust Overlay Architecture for Mobil- identifier with location in the network topology. For a ity (ROAM), is built on top of the Internet Indirection ∗ICSI Center for Internet Research (ICIR), Berkeley, Infrastructure (i3) [6]. i3 is implemented as an overlay [email protected] network on top of IP, and provides a rendezvous-based communication abstraction. In i3, each packet is sent to bile IP (MIP). MIP in IPv4 [1] and IPv6 [8] uses an an identifier. To receive a packet, a receiver inserts a trig- explicit indirection point, called the Home Agent (HA), ger, which is an association between the packet's identi- to encapsulate and relay the Correspondent Host's (CH) fier and the receiver's address. The trigger is stored at an initial packet to the Mobile Host (MH). MIP provides i3 node (server). Each packet is routed through the over- the following options that determine how the following lay network until it reaches the i3 server which stores the packets are routed: 1) triangle routing, 2) bidirectional trigger. Once the matching trigger is found the packet is tunneling, and 3) route optimization. forwarded to the address specified by the trigger. Thus, As noted by Cheshire and Baker [9] no MIP routing op- the trigger plays the role of an indirection point that re- tion is clearly better than the others; instead, different op- lays packets from the sender to the receiver. tions are suitable for different circumstances. Options (1) ROAM addresses each of the properties described above. and (2) preserve location privacy, but routing can be in- For instance, since an i3 identifier can be bound to a efficient when the MH and CH are close relative to their host, session, or person (unlike Mobile IP, where an IP distance from the HA. With route optimization (an ex- address can only be bound to a host), personal/session tension in MIPv4 [10], but standard in MIPv6), the MH mobility applications can leverage the ROAM infrastruc- conveys its care-of IP address to the CH using a Binding ture for efficiency, fault tolerance, and privacy. Section 4 Update (BU). Routing is efficient because the ratio of the discusses in detail how ROAM achieves the above prop- latency of the optimized route to the latency of the short- erties. est IP path (or latency stretch) is 1.0. However, the CH At the architectural level, this paper makes two contri- must be modified to support MIPv4 with route optimiza- butions. First, it demonstrates the benefit of giving end- tion or IPv6. This also exposes the MH's current care-of hosts control on the placement of the indirection points. address (and therefore its location) to the CH, thus com- This allows, end-hosts to optimize the routing and hand- promising location privacy. In certain delay-sensitive or off efficiency. Second, it demonstrates the benefits of a real-time applications, the latency involved in handoffs mobility architecture based on a shared overlay network. can be above the threshold if the MH is far away from Such a solution leverages the robustness of the overlay the CH. networks. In general, the dependence in MIP on a fixed HA reduces In addition, we use a proxy based solution to transpar- fault tolerance. If the HA or its network fails or is over- ently support unmodified applications on an unmodified loaded, then the MH will be unreachable. Linux kernel. Using our prototype implementation, we To address routing anomalies and robustness issues as- show that our solution can perform rapid soft handoffs sociated with a fixed HA, researchers have proposed the with no noticeable disruption of TCP throughput. notion of dynamic home agents in MIPv4 [11]. How- The paper is organized as follows. Section 2 presents ever, the actual algorithm used to discover and allocate the related work, and Section 3 gives an overview of a nearby home agent is still under investigation. MIPv6 i3. Section 4 discusses the design of ROAM, and Sec- provides a dynamic home agent address discovery mech- tion 5 presents the ROAM support for legacy applica- anism [8] that allows a MH to dynamically discover the tions. Section 6 presents some implementation details. IP address of a HA on its home network. This scheme in- Section 7 presents simulation and experimental results. creases the robustness of MIPv6 as the HA is no longer a Finally, Section 8 discusses some open issues, and Sec- statically fixed entity, but it does not address routing in- tion 9 concludes the paper. efficiencies caused by routing through the HA when the MH is far away from its home network. 2 Related Work Recently, two mechanisms have been proposed to in- crease handoff performance in MIPv4 and MIPv6: (1) In this section we review the main mobility proposals. low latency handoff [12], and (2) fast handover [13]. The first mechanism attempts to send a BU in advance of an Several link layer technologies provide mobility at the actual link-layer handoff when the handoff is anticipated. link layer (e.g., as in Ricochet [7], 802.11b, or GSM). However, timing must be arranged such that the BU com- However, these solutions preclude mobility across link pletes before the actual handoff does, which may be hard layer technologies. In addition, hiding mobility at the to achieve in practice. Similar in concept to Regionalized link layer results in a reinvention of mobility support in Tunnel Management [14] and Hierarchical Mobility [15] each new wireless system; solving the mobility problem extensions in MIPv4 and MIPv6, the second mechanism at the network layer results in a reusable mobility infras- sets up a bi-directional tunnel between an anchor Foreign tructure for all link technologies. Agent (FA) that stays the same during rapid movements One proposal to achieve mobility in the Internet is Mo- and the current FA. This allows the MH to delay a formal i3's Application Programming Interface (API) BU to the HA which minimizes the impact on real-time sendP acket(p) send packet applications. However, this mechanism relies on the ex- insertT rigger(t) insert trigger istence of a FA in each network the MH visits.