| TOC |
|
This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as “work in progress.”
The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html.
This Internet-Draft will expire on April 23, 2010.
Copyright (c) 2009 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents in effect on the date of publication of this document (http://trustee.ietf.org/license-info). Please review these documents carefully, as they describe your rights and restrictions with respect to this document.
Over the last two decades many research efforts have been devoted to developing solutions for mobility support over the global Internet and resulted in a variety of proposed solutions. We conducted a survey of the previous efforts to deepen our understanding on the overall solution space of mobility support. This draft reports on our finding and identifies remaining issues in providing ubiquitous and efficient global scale mobility support.
1.
Introduction
2.
Terminology
3.
Basic Components in Mobility Support Protocols
4.
Existing Mobility Support Protocols
4.1.
Columbia Protocol
4.2.
Sony Protocol
4.3.
LSR Protocol
4.4.
Mobile IP
4.5.
MSM-IP
4.6.
Cellular IP, HAWAII and TIMIP
4.7.
Hierarchical Mobile IP
4.8.
NEMO
4.9.
E2E and M-SCTP
4.10.
Host Identity Protocol
4.11.
Connexion and WINMO
4.12.
ILNP
4.13.
Global HAHA
4.14.
Proxy Mobile IP
4.15.
Mobile Me
4.16.
LISP-Mobility
5.
Different Directions towards Mobility Support
5.1.
Routing-based Approach v.s. Mapping-based Approach
5.2.
IP Layer Indirection v.s. Mobility Exposure
5.3.
Operator-Controlled Approach v.s. User-controlled
5.4.
Local and Global Scale Mobility
6.
Discussions
6.1.
Backward Compatibility
6.2.
Undisrupted TCP Connection
6.3.
Interconnecting Heterogeneous Mobility Support Systems
6.4.
Flat-id Based Routing
7.
References
§
Authors' Addresses
| TOC |
This draft reports our findings from a historical survey of the Internet mobility research and standardization efforts since the early '90s. Our survey was motivated by two factors. First, supporting mobility over the Internet has been an active research area and has produced a variety of solutions; some specific solutions have become the Internet standards. Yet new issues and new solutions continue to arise, making one wonder how much more we are yet to discover about the problem space as well as the solution space. The second motivation is the rapid growth in Internet access via mobile devices in recent years, which will inevitably lead to new Internet application development in coming years and further underscore the importance of Internet mobility support. We believe that a historical review of all the proposed solutions not only can help us better understand their commonalities and differences, but can also help shed insight on future efforts.
In the rest of this document, we provide an overview of the mobility support solutions from the early works to the most recent proposals. In the process we also discuss the essential components in mobility support, analyze the design space and try to initiate an open discussion by sharing our understanding about the general direction for future mobility support.
| TOC |
This document uses a number of terms to refer to the entities or functions that are required in mobility support. Readers are also encouraged to scan [RFC3753] (Manner, J. and M. Kojo, “Mobility Related Terminology,” .) before reading this document.
Identifier
A stable piece of information that can be used to identify a mobile node. Anything could be used as an identifier as long as it remains unchanged when the mobile node roams around.
Locator
The IP address that indicates the mobile node's current location. It could be the IP address of the mobile node itself, or the IP address of the network entity that is currently serving the mobile node.
Mapping
In this document, mapping specifically means the mapping between a mobile's identifier and it's Locator.
Rendezvous Point
The place where the mapping is held. Some other functions such as data forwarding may also be placed on the rendevzvous point.
Global Mobility Management
The mobility management in a global scale. It keeps the mobile's reachability during the mobile's long distance moving, either geographically or topologically.
Local Mobility Management
The mobility management within a topologically local domain. It tries to keep the mobile's local movements transparent to the network entity that manages the mobile's mobility in a global scale. It also tries to improve the handoff performance.
Operator Controlled Mobility Management
The mobile node does not get involved in mobility management. Instead, the network entities, which are controlled by the network operators, do all the mobility related signalling job on behalf of the mobile node.
User Controlled Mobility Management
The mobile node participates in the mobility management. Typically, the mobile updates the mapping after it changes locations and refresh the mapping at a user-defined frequency.
| TOC |
Mobility support can be provided through through multiple different ways. The basic question is how to make data reach a moving receiver (a mobile in short; here we do not distinguish between mobile nodes and mobile subnets). Whoever sending data to a mobile must be able to identity the receiver via a piece of stable information, which "stable" means that the information does not change as the mobile moves. However if the sender's knowledge about the mobile does not change while the mobile moves, some means must exist to bind that unchanged identifier of the mobile with its dynamically changing locations.
The above intuitive reasoning leads to the following observation--mobility support essentially involves three basic components:
In the next section we show that different mobility support designs are merely different ways to choose mobile identifiers and different approaches to provide mapping between the identifiers and the mobiles' current IP addresses.
| TOC |
In this section, we review the existing mobility support protocols roughly in time order (There are exceptions: for the sake of convenience, we group closely related protocols together). We briefly describe their design and point out how they implement the three basic components defined in last section.
Figure 1 shows a list of the protocol names and the time when they were first proposed.
+----------------+-----+---------------+-----+
| Protocol Name |Year | Protocol Name |Year |
+----------------+-----+---------------+-----+
| Columbia |1991 | TIMIP |2001 |
+----------------+-----+---------------+-----+
| Sony |1991 | M-SCTP |2002 |
+----------------+-----+---------------+-----+
| LSR |1993 | HIP |2003 |
+----------------+-----+---------------+-----+
| Mobile IP |1996 | Connexion |2004 |
+----------------+-----+---------------+-----+
| MSM-IP |1997 | ILNP |2005 |
+----------------+-----+---------------+-----+
| Cellular IP |1998 | Global HAHA |2006 |
+----------------+-----+---------------+-----+
| HMIP |1998 | PMIP |2006 |
+----------------+-----+---------------+-----+
| HAWAII |1999 | Mobile Me |2007 |
+----------------+-----+---------------+-----+
| NEMO |2000 | WINMO |2008 |
+----------------+-----+---------------+-----+
| E2E |2000 | LISP-Mobility |2009 |
+----------------+-----+---------------+-----+
| Figure 1 |
| TOC |
This protocol was originally designed to provide mobility support in a campus. A router named Mobile Support Station (MSS) is set up in each wireless cell, which is the default access router for all mobile nodes in that cell. Mobile node obtains an IP address from a special block of IP addresses and keeps it regardless of it's current location. Each MSS keeps a list of mobile nodes that are currently in its cell. When a correspondent node sends a packet to a mobile node, the packet is first routed to the closest MSS near correspondent node. This MSS will deliver the packet directly to the mobile node if the mobile node happens to be in its cell, or forward the packet to the MSS that serves the mobile node. In the latter case, a MSS sends a query message to all other MSSs in the campus in order to find out current location of the mobile node. Each MSS in turn checks its tracking list of mobile nodes, and sends a reply if the target mobile node is on the list.
Here, the three basic components are:
The illustration of Columbia Approach is shown in Figure 2.
+---------+
| |
.------>| MAS |
| | |
| +---------+
| query
|
+--------+ query +--------+
| | -------------->| |
| MAS | <------------- | MAS |
| | reply | |
+--------+ ==============>+--------+
/\ data ||
|| ||
|| \/
+--------+ +---------+
| | | |
| CN | | MN |
| | | |
+--------+ +---------+
===>: data packets
--->: signaling packets
| Figure 2 |
| TOC |
In this protocol, the IP header is modified so that each mobile node has two IP addresses: a Virtual IP address and a conventional IP address. Virtual IP address is the IP address that a mobile node obtains from its home network, whereas a conventional IP address is the IP address a mobile node obtains from its access router. Every time the mobile node changes its location, it sends a special packet containing the mapping between its Virtual IP address and conventional IP address to its home network, which in turn stores the mapping. Routers in the middle as well as the correspondent hosts can also cache a set of mappings for mobile nodes.
To deliver data to the mobile node, the correspondent node sets the destination IP address the same as mobile node's Virtual IP address. If a router along the way from correspondent node to mobile node's home network happens to have a valid mapping for mobile node, it modifies the destination IP address field and forwards the packets to the new destination; otherwise, the packet will flow to mobile node's home network, which then delivers the packets to mobile node according to the mapping. Note that the destination IP address field can be modified again if the packet encounters a router with newer mapping.
The three basic components are easy to find:
Figure 3 shows how sony protocol works.
,---. +-------+
/ \ | CN |
( Router)<====| |
+---------+ //\ / | |
| | // `---' +-------+
| | ,---. //
| | / \ //
| Home |<--- Router)//
| Network | \ / \\
| | `---' \ \\
| | \ \\,---. +-------+
| | \ / \=======>| |
| | ( Router)<------+ MN |
| | \ / | |
| | `---' +-------+
+---------+
===>: data packets
--->: location update message
| Figure 3 |
| TOC |
Each mobile node has a designated router that manages its mobility, called Mobile Router. Mobile Router assigns an IP address for each mobile node it manages and announce reachability to those IP addresses. Another network entity required is Mobile Access Station (MAS), which, as indicated by its name, gives a mobile node access to the Internet regardless of the mobile node's IP address. The mobile node always reports the current serving MAS to its Mobile Router.
If the correspondent node and the mobile node are attached to the same MAS, then it's easy: the MAS can just forwards packets between the two; otherwise, the packet sent by the correspondent node would be routed to the Mobile Router according to rules of the IP routing. The Mobile Router exams the mappings, finds the serving MAS of the mobile node, and inserts the loose source routing (LSR) option into the IP header of the packet with the IP address of the MAS on it. This way, the packet is redirect to the MAS and then to the mobile node. After that, the mobile node reverses the LSR and replies to the correspondent node; the correspondent node, similarly, sends the following packets along an optimal path by reversing the LSR.
Let's identify the three basic components for this protocol:
Figure 4 shows the basic operation of LSR protocol.
+---------+
| |
___________________| CN |
| | |
| +---------+
V /\
+-------+ ||
|Mobile | ||
|Router | ||
| | || Reversing LSR
+---+---+ ||
| \/
| +---------+ +----------+
| LSR Inserted | |<====>| |
+------------------->| MAS | | MN |
| |----->| |
+---------+ +----------+
-->: first data packet
==>: following data packets
| Figure 4 |
| TOC |
IETF begun its effort in mobility support area following the path of the above three protocols. In 1996, the first version of Mobile IP was finished. Later, IETF further made Mobile IPv4 [RFC3220] (Perkins, C., “IP Mobility Support for IPv4,” 2002.) and Mobile IPv6 [RFC3775] (Johnson, D., Perkins, C., and J. Arkko, “IP Mobility Support in IPv6,” 2004.) standards in 2002 and 2004 respectively.
Let's take Mobile IPv6 as an example. Each mobile node has a Home Agent, from which it acquires its Home Address (HoA). Unlike in LSR protocol, however, the mobile node also obtains a Care-of Address (CoA) from its access router to the Internet. Whenever the mobile node changes its point of attachment and gets a new CoA, it sends a Binding Update message to notify the Home Agent about the new CoA.
The correspondent node simply sets the destination field in the IP header to the HoA of the mobile node, and the packets are routed to the Home Agent since Home Agent is announcing the reachability to the HoA. Home Agent then encapsulates the packets to mobile node's CoA according to the mapping. If the correspondent node supports Route Optimization, then it can also keep a mapping between the mobile node's HoA and CoA; as specified by the protocol, the mobile node would update the mapping in correspondent node as well. Thus the correspondent node can tunnel packets to the mobile node directly, bypassing the Home Agent and resulting in a shorter data path.
The three basic components in Mobile IP are:
Figure 5 illustrates the data path of Mobile IPv6 without Route Optimization.
+---+-----+
|HoA|DATA |
+---+-----+ +-------+
+----------------------| CN |
| +------------------->| |
| | +-------+
| |
V |
+--------+
| Home | Mapping: HoA <=> CoA
| Agent |
| |
+--------+
|| /\
|| || +-------+
|| +====================| |
|| | MN |
+=======================>| |
+-----+---+---+ +-------+
|DATA |HoA|CoA|
+-----+---+---+
==>: Tunnel
-->: regular IP
| Figure 5 |
| TOC |
MSM-IP stands for Mobility Support using Multicast in IP. As one can learn from its name, it leverages the IP multicast routing to support the mobility. In IP multicast, a host can join a group regardless of to which network it attaches, and all the packets sent to the group can be received after that. Thus, as claimed by the designers of MSM-IP, mobility is naturally supported if IP multicast problem is solved. Note that their purpose is not to show the feasibility of supporting mobility with current IP multicast architecture, but rather the possibility of reusing IP multicast, if it is well supported in the future, to solve mobility support problem.
MSM-IP [MSM‑IP] (Mysore, J. and V. Bharghavan, “A New Multicast-based Architecture for Internet Host Mobility,” 1997.) assigns each mobile node a unique multicast IP address. When the mobile node moves into a new network, it initiates a join to its own address, which makes the multicast router in that subnet join the multicast distribution tree. Whoever wants to communicate with the mobile node can just send the data to the mobile's multicast IP address.
In order to make this approach deployable in wide-area networks, a hierarchical location management service is also proposed. In the location server for each mobile node, the IP address of the multicast router that serves mobile node is stored. Thus, a new multicast router can always query the location server to find out how to join the multicast distribution tree.
Also note that, due to the nature of multicast routing, the mobile node can have the new multicast router join the group to cache packets in advance before it detaches the old one, resulting in smoother handoff.
The three basic components here are:
| TOC |
This is a group of protocols that share the common idea of seting up host route for the mobile node in the local domain. They are all intended to work with Mobile IP as a local mobility management protocol. We introduce them together so that it is much clearer to show the differences by comparison.
Cellular IP [CIP] (Valko, A., “Cellular IP: A New Approach to Internet Host Mobility,” 1999.) handles the local mobility in a network consists of Cellular IP nodes, which are essentially special routers that integrates location tracking service with routing. The mobile node registers the local gateway node's IP address to Home Agent as the regional CoA, and retains its assigned IP address when it roams within the Cellular IP network. The network monitors the packets originated from mobile node and maintains a distributed, hop-by-hop reverse path which can be used to route packets back to the mobile node. It utilizes paging technique from cellular network to track the location of mobile node efficiently, by allowing mobile node to update its location (via the way of sending dummy packets) with a relatively long period. When a packet from correspondent node arrives at the gateway node, controlled flooding query is initiated if the gateway node has no idea about the exact location of mobile node. However, a much more precise reverse path is maintained later when the communication between mobile node and correspondent node is going on.
Similarly, HAWAII [HAWAII] (Ramjee, R., Varadhan, K., and L. Salgarelli, “HAWAII: A Domain-based Approach for Supporting Mobility in Wide-are Wireless Networks,” 2002.) also aims to provide efficient local mobility support with two main differences from Cellular IP. First, it dynamically setup routes to mobile node by installing host-based forwarding entry in specific routers, usually the ones located in the shortest path between the old and new serving base stations. Second, it uses IP multicast to page idle mobile nodes when they have incoming packets arrived at the border router and there are no routing entries available for these addresses.
TIMIP [TIMIP] (Grilo, A., Estrela, P., and M. Nunes, “Terminal Independent Mobility For IP,” 2001.), which stands for Terminal Independent Mobile IP, integrated Cellular IP and HAWAII together. On one hand, it refreshes the routing paths with dummy packets if there is no traffic for the mobile node. On the other hand, handoff within a domain results in the changes of routing tables in the routers. Besides, the IP layer is coupled with layer 2 handoff mechanisms and special nodes can work as Mobile IP proxies for legacy nodes that do not support Mobile IP. Thus, as long as it roams within the domain, a legacy node has the same degree of mobility support as a Mobile IP capable node.
Given the similarity of the three protocols, the implementation of the three basic components in them are listed as below:
| TOC |
This [RFC4140] (Soliman, H., Castelluccia, C., Malki, K., and L. Bellier, “Hierarchical Mobile IPv6 Mobility Management (HMIPv6),” 2005.) is a simple extension to Mobile IP introduced in 1998. It aims to work with Mobile IP as a local mobility support protocol. A level of hierarchy is added to Mobile IP in the following way. A Mobility Anchor Point (MAP) is responsible for handling the mobility of mobile node in a local region. Simply speaking, MAP is the local Home Agent for the mobile node. The mobile node, if it supports Hierarchical Mobile IP (HMIP), obtains a Regional CoA and register it to Home Agent as its current CoA; at the same time, it also obtains a Local CoA from the subnet it attaches to. When roaming with the region, mobile node only needs to keep the mapping between Regional CoA and Local CoA updated in MAP. In this way, the handoff performance is usually better due to the short round-trip time, and it also enhance the scalability of Mobile IP.
The basic components here are:
| TOC |
It is not unusual for a group of hosts to move together. Consider vehicles such as ships, trains, airplanes which contain networks with a large number of nodes that need global mobility support. Mobile IP is not efficient when handling such mobility scenarios. Consequently, NEMO [RFC3963] (Devarapalli, V., Wakikawa, R., Peterson, A., and P. Thubert, “Network Mobility (NEMO) Basic Support Protocol,” 2005.), as a backward compatible extension to Mobile IP, was introduced in 2000 to handle the network mobility support problem.
A new entity call Mobile Router (note that this is different from the "Mobile Router" in LSR protocol) is introduced. Every mobile network has at least one Mobile Router via which it can be accessed. Mobile Router is a special mobile node as in original Mobile IP, with more powerful functions. After establishing bidirectional tunnel with Home Agent, the Mobile Router will distribute the network's pefixes (namely Mobile Prefixes) through the tunnel to Home Agent, but never leaks them to the infrastructure at its point of attachment (i.e. the access router never knows that it can reach the Mobile Prefixes via the Mobile Router). The Home Agent in turn announces the reachability to these Mobile Prefixes. Packets to and from mobile network always flow through the bidirectional tunnel to their destinations.
Note that, mobility is transparent to the nodes in the moving network.
Hence, in this protocol, the basic components are:
| TOC |
E2E (End-to-End communication) gets the name from its end-to-end architecture, and is the first formal proposal that utilizes existing DNS infrastructure to track mobile node's current location. It works in a conceptually simple way: Dynamic DNS update is used to refresh the current IP address of the mobile node in DNS server; to keep the ongoing TCP connection unaffected by mobility, a TCP Migrate option is introduced, which enables both ends to replace the IP addresses and ports in TCP 4-tuple on the fly. Migration security is protected based on token and sequence number.
Inspired by E2E, M-SCTP [M‑SCTP] (Xing, W., Karl, H., and A. Wolisz, “M-SCTP: Design and Prototypical Implementaion of An End-to-End Mobility Concept,” 2002.) was proposed in 2002. Similarly, it uses Dynamic DNS to track the mobile nodes and allows both ends to add/delete IP addresses use in SCTP in order to keep the SCTP association during the move.
The three components hence are:
| TOC |
Host Identify Protocol (HIP) [RFC5201] (Nikander, P., Moskowitz, R., Jokela, P., and T. Henderson, “Host Identity Protocol,” 2008.) introduces a new name space consisting of cryptographic keys, and places a Host Identity layer between transport and network layers. As a result, the Host Identities, which are essentially public keys, are used to identify the mobile nodes; while at the same time, IP addresses are only used for routing purpose. For the convenience of reusing existing code, Host Identity Tag, a 128-bit hash value of the Host Identity, is used in the upper layer connections.
Dynamic DNS again can serve as the rendezvous point and holds the mappings for HIP. Nevertheless, HIP has also specified its own static infrastructure Rendezvous Servers, in expectation of better rendezvous service. Each mobile node has a designated Rendezvous Server (RVS), which tracks the current location of mobile node. When a correspondent node wants to communicate with mobile node, it queries DNS with mobile node's HIT to obtain the IP address of mobile node's RVS, and sends out the first packet (I1) of Base Exchange, which is a four-packet handshake with the purpose of establishing an IPSec Encapsulated Security Payload and Security Association pair. After receiving this packet, RVS will relay it to mobile node. Then mobile node and correspondent node can finish Base Exchange and start communication directly. If mobile node moves to a new address, it simples notifies correspondent node by sending HIP UPDATE with LOCATOR parameter indicating the new IP address.
The basic components here are:
| TOC |
Connexion [Boeing] (Andrew, L., “A Border Gateway Protocol 4 (BGP-4),” 2006.) is a service provided by Boeing that uses BGP to support mobility. Every mobile network is assigned a /24 prefix. When the plane moves between ground stations, which are the access points to Internet, it withdraws its prefix from the BGP table in the older ground station, and announce the prefix via the new ground station. Thus, the location change of the plane is effectively propagated to the rest of the world. However, if the number of moving networks soars, an avalanche of BGP updates will be generated to the whole Internet, resulting in severe global routing instability.
WINMO [WINMO] (Hu, X., Li, L., Mao, Z., and Y. Yang, “Wide-Area IP Network Mobility,” 2008.) (which stands for Wide-Area IP Network Mobility) was later designed in 2008 to address the routing instability problem of Connexion.
Like Connexion, WINMO also assigns each mobile network a stable prefix, however, WINMO made the BGP updates overhead for mobile networks orders of magnitude lower than that of Connexion by the following two approaches. First, WINMO uses various heuristics to reduce the propagation scope of routing updates caused by mobile movements. Consequently, not every router may know all the mobiles' current locations. Resolving this issue led to the second, and more fundamental approach taken by WINMO: adoption of the basic idea from Mobile IP. WINMO assumes that each mobile subnet is assigned a prefix out of a small set of well defined Mobile Prefixes. These Mobile Prefixes are announced by a small set of Aggregation Routers, which keep track of the mobile networks. Thus one may view WINMO's Aggregation Routers as playing the role of Home Agents in Mobile IP.
To prevent frequent iBGP routing changes due to the movement of mobile networks within an AS, WINMO also adopts the Mobile IP way. It introduced a Home Agent for the Mobile Prefixes within an AS: only a designated BGP speaking router (DBR) acts as the origin of Mobile Prefixes; mobile networks always update their addresses of attachment points to DBR, which resembles the binding updates in Mobile IP. Thus, packets destined to mobile networks will be routed to DBR after it enters the border of AS, and DBR will be responsible to tunnel them to the right receivers. A BGP community name Packet State is also created to eliminate the triangle routing problem caused by DBR, in a way similar to Route Optimization, by leaking CoA to corresponding node.
The basic components in Connexion are:
The basic components in WINMO are:
| TOC |
ILNP [ILNP] (Atkinson, R., “An Overive of the Identifier-Locator Network Protocol,” 2005.) stands for Identifier/Locator Network Protocol. It splits IPv6 address into two parts: high-order 64 bits as a Locator field and low-order 64 bits as an Identifier field. Locator is used for IP packet delivery, while Identifier is used by all upper layer protocols, just as HIT in HIP. During the movement, the mobile node updates the binding entry between Locator and Identifier via Dynamic DNS Update; it also updates Locator to correspondent nodes with a new ICMP Locator Update message.
The basic components in ILNP are clear:
| TOC |
Global HAHA [HAHA] (Wakikawa, R., Valadon, G., and J. Murai, “Migrating Home Agents Towards Internet-scale Mobility Deployment,” 2006.), first proposed in 2006, aims to eliminate the triangle routing problem in Mobile IP and NEMO by distributing Home Agents in a wide area. A set of Home Agents joins an IP anycast group and forms an overlay network. The same home prefixes are announced by all the Home Agents at different locations. Each mobile node can register with any Home Agent that is closest to it. The Home Agent that accepts the binding request of mobile node, namely the primary Home Agent, is responsible to notify all other Home Agents in the overlay network, so that the binding information databases in all Home Agents are always synchronized.
Packets from correspondent node are delivered to the nearest Home Agent by anycast routing. This Home Agent then forwards these packets through the overlay network to the target Home Agent that is currently serving mobile node, which will finally deliver the packets to mobile node after striping the tunneling headers. In the reverse direction, this approach works exactly the same as Mobile IP. If the distribution of Home Agents are carefully designed, the triangle routing problem is solved naturally without Route Optimization.
Thus, the basic components in Global HAHA are:
The data flow in Global HAHA is shown in Figure 6.
+------+ +------+ +-----+
| HA |_____________| HA | | |
| | | | | CN |
+--+---+ .'+++----+ +-----+
| .' || /\
| .' || ||
| .-' || ||
| .' || ||
+--+---+.' || ||
| |<==============+ ||
| HA |==============================+
+-++---+
|| /\
\/ ||
+---++-+ ===>: data flow
| | ----: HA overlay network
| MN |
+------+
| Figure 6 |
| TOC |
Proxy Mobile IP [RFC5213] (Gundavelli, S., Leung, K., Devarapalli, V., Chowdhury, K., and B. Patil, “Proxy Mobile IPv6,” 2008.) was proposed in 2006. The two new types of network nodes, Local Mobility Anchor (LMA) and Mobile Access Gateway (MAG), together handle the local mobility without interference from mobile node. LMA serves as a local Home Agent and assigns a local Home Network Prefix for each mobile node. MAGs monitor the attaching and leaving events of mobile node, and performs Proxy Binding Update to LMA on behalf of mobile node during handoff. After the success of binding, LMA updates mobile node's Proxy-CoA with the IP address of the MAG that is currently serving mobile node. The MAG then emulates mobile node's local Home Link by advertising mobile node's local Home Network Prefix in Router Advertisement. When roaming in the PMIP domain, mobile node always obtains its local Home Prefix, and believes that its on local Home Link.
Here, the basic components are:
| TOC |
Mobile Me [BTMM] (Nakamoto, R., Zhu, Z., Lau, D., and B. Allen, “Understanding Apple's Back to My Mac Service,” 2009.) is a pragmatic approach to support mobility and has been deployed as real-world commercial service since 2007 (with Mac OS leopard release). In this approach, Apple provides DNS server for all Mobile Me users. Each mobile node performs secure DNS update to dynamically refresh its current location. In the DNS database, an AAAA RR represents mobile node's identifier, which is an IPv6 address, and a SRV [RFC2782] (Gulbrandsen, A., Vixie, P., and L. Esibov, “A DNS RR for Specifying the Location of Services (DNS SRV),” 2000.) RR records mobile node's current location with IPv4 address, and port number in case of NAT traversal. Every node establishes long-lived query (llq) with DNS, which means that the DNS will automatically notify the node if the answer to the query changes. In this way, the node only need to refresh the llq after a certain period, avoiding the frequent queries for the up-to-date location of the other communicating end host. A node uses the IPv6 address in all applications. It then uses UDP/IPv4 encapsulation to deliver the IPv6 packets according to the DNS reply. Note that the IPv6 address is only for identification purpose and not used in routing. In fact, it could be any form of identifier (e.g. HIT in HIP, domain name, etc.); IPv6 address was chosen simply because the designer do not want to change the IP header and cause problems of reusing existing code.
Mobile ME is perhaps the first large-scale commercial host mobility support in Internet with millions of subscribers as of today. It is simple and requires little budget to deploy. However, it does not fully support mobility yet.
We can also list the basic components for Mobile Me:
Figure 7 shows an example communication scenario of Mobile Me.
+----------+ dynamic DNS update
llq | |<------------------+
+-------->| DNS | |
| | | <------------+ |
| +----------+ llq | |
| | |
| | |
V V |
+-------+ ,---. +-----+-+
| | UDP/IPv4 / \ | |
| Mac1 |<=============>( NAT )<---->| Mac2 |
| | tunnel \ / | |
+-------+ `---' +-------+
| Figure 7 |
| TOC |
LISP-Mobility[LISP‑Mobility] (Farinacci, D., Fuller, V., Lewis, D., and D. Meyer, “LISP Mobility Architecture,” 2009.) is a relatively new idea, in which the designer hopes to support mobility via LISP[LISP] (Farinacci, D., Fuller, V., Lewis, D., and D. Meyer, “Locator/ID Separation Protocol (LISP),” 2009.) functionality. Conceptually, however, it is very similar to some protocols we have mentioned so far, such as ILNP and Mobile IP. Light-weight Ingress Tunnel Router and Egress Tunnel Router are implemented on each mobile node, encapsulating/decapsulating the packets to and from the mobile node. Each mobile node is assigned a static Endpoint ID, which is used in upper layer connections, as well as a pre-configed Map-Server. When a mobile node roams into a network and obtains a new Routing Locator, it updates its Routing Locator set in the Map-Server and also in the Ingress Tunnel Routers or Proxy Tunnel Routers of the correspondent nodes. Thus the correspondent node can always learn the location of the mobile node via the resolution of its Endpoint ID, and the data would always travel through the shortest path. Note that both Endpoint IDs and Routing Locators are essentially IP addresses.
It is trivial to list the basic components in this protocol:
| TOC |
What would be the best way to provide global scale ubiquitous mobility support for an essentially unlimited number of mobile devices with unknown future applications? No one has the crystal ball that clearly shows the future. Nevertheless, we identified several different directions towards mobility support by studying various existing protocols. We believe a thorough understanding of these directions will help us make wise decisions when designing mobility support system for the future.
| TOC |
All existing mobility support designs can be broadly classified into two basic approaches.
The first one is to support mobility through dynamic routing, in which case there is not explicit "mapping" component as we discussed in last section. In such designs, a mobile keeps its IP address regardless of its location changes, thus the IP address can be used both to identify the mobile and to deliver packets to it. As a result, these designs do not require explicit mapping function. Rather, the routing system must continuously keep track of mobile's movements and reflect their current positions in the network on the routing table, so that at any given moment packets carrying the (stable) receiver's IP address can be delivered to the right place.
Supporting mobility through dynamic routing is conceptually simple as it does not require a mapping function; it can also provide robust and efficient routing, assuming that the routing system can keep up with the mobile movements. However, because the whole network must be informed of every movement by every mobile, this approach is feasible only in small scale networks with a small number of mobiles; it does not scale well in large networks or for large number of mobiles.
The second approach to mobility support is to provide a mapping between a mobile's stable identifier and its dynamically changing IP address. Instead of notifying the world on every movement, a mobile only needs to update a single binding location about its location changes. In this approach, if one level of indirection at IP layer is used, as in the case of Mobile IP, it has a potential side effect of introducing triangle routing; otherwise, if the two end nodes need to be aware of each other's movement, which means that both ends have to support the same mobility protocol.
Yet there is the third case in which the protocols combine the above approaches, in the hope of keeping the pros and eliminating some cons of the two. WINMO is a typically protocol in this case.
In Figure 8 we show the classification of the existing protocols according to the above analysis.
_.--------.
,-'' `--.
,-' `-.
,' +----+ `.
,-------. / +----+ |ILNP| +----+ \
,+--------+ `-,' |Sony| +----+ |HMIP| `.
,' |Columbia| / `. +----+ +----+ \
,' +--------+ / `. +---+ +------+ \
/ ; \ +----+ |E2E| |M-SCTP| :
/ +---------+ ;+-----+ \ |NEMO| +---+ +------+ :
; |Connexion| ; |WINMO| : +----+ +---------+ :
; +---------+ | +-----+ : +---+ |Mobile Me| |
; +-----+ | : +---+ |HIP| +---------+ |
| |TIMIP| | +------+ | |LSR| +---+ |
: +-----+ : |MSM-IP| ; +---+ +---------+ ;
: +------+ :+------+ ; +----+ |Mobile IP| ;
: |HAWAII| : ; |PMIP| +---------+ ;
\ +------+ \ / +----+ +-----------+ /
\ +-----------+ \ / |Global HAHA| /
`. |Cellular IP| `. ,' +-----------+ ,'
`+-----------+ ,' +-------------+ /
`-. ,-' `. |LISP-Mobility| ,'
`-------' '-+-------------+ ,-'
`--. _.-'
`--------''
Routing-based Mapping-based
| Figure 8 |
| TOC |
One critical choice in mapping-based systems is the choice of rendezvous point, i.e. where to locate the mapping function and whether or not the rendezvous point provides other functions except storing mapping information.
Several mobility support designs provide the mapping function at IP layer, which both the identifier and the current location of a mobile are represented by IP addresses. In this approach, the IP address which is used as the mobile's identifier points to the rendezvous point that keeps track of the mobile's current location. Such designs offer an advantage of hiding the mobility from correspondent nodes through one level of indirection. When a correspondent node sends packets to an IP address which is a mobile's identifier, the packets will be delivered to the location where the mapping information of the mobile is kept (e.g. the Home Agent in Mobile IP), and later they will be forwarded to the mobile's current location via either encapsulation or destination address translation.
Although this one level of indirection at IP layer makes mobility transparent, it has a potential side effect of introducing triangle routing: the path taken by the packets via the rendezvous point can be much longer than the direct path between the correspondent and the mobile's current location. Besides, as increasing number of mobile devices are connected to Internet (why hide mobility to them), some mobility solutions have opted to expose mobility to both ends and let them communicate directly. One common approach taken by these protocols is to use DNS for the mapping function to keep track of mobiles' current locations. Mobiles use dynamic DNS updates to keep their DNS servers updated with their current locations. This approach re-utilizes the DNS infrastructure, which is ubiquitous and quite reliable, and makes the mobility support protocol simple and easy to deploy.
Figure 9 shows the two categories of mobility protocols according to their choices of rendezvous point. Note, however, protocols like Mobile IP can also expose mobility to the other end if Route Optimization is supported.
,-------.
,-' `-.
,' +---------+ `. ,-----.
/ |Mobile IP| \ ,-' `-.
,' +---------+ `. / +----+ \
; : ,' |ILNP| `.
; +----+ : ; +----+ :
/ |HMIP| +-----------+ \ ; +------+ :
; +----+ |Global HAHA| : / +---+ |M-SCTP| \
| +-----------+ | ; |E2E| +------+ :
| +---+ | | +---+ +---------+ |
| |LSR| +----+ | | |Mobile Me| |
: +---+ +----+ |NEMO| ; | +---+ +---------+ |
\ |Sony| +----+ / : |HIP| ;
: +----+ ; \ +---+ /
: ; : +-------------+ ;
`. +----+ ,' : |LISP-Mobility| ;
\ |PMIP| / `. +-------------+'
`. +----+ ,' \ /
'-. ,-' `-. ,-'
`-------' `-----'
IP Layer Indirection Mobility Exposure
| Figure 9 |
| TOC |
At the time when this draft was written, the largest global mobility support today is provided by cellular networks, using a service model that bundles together the device control, network access control and mobility support. The tremendous success of cellular market speaks loudly that the current cellular service model is a viable one, and is likely to continue for foreseeable future. As a result, there is a strong advocate in IETF that we continue the cellular way of handling mobility, i.e. the mobile do not necessarily need to participate in the mobility related signaling; instead, the network entities deployed by the operators will take care of any signaling process of mobility support. A typical example is Proxy Mobile IP, in which LMA work together with MAGs, assuring that the mobile always obtains its Home Prefixes as long as it roams within the domain.
The main reason for this approach is perhaps backward compatibility. By not requiring the participation of mobiles in control signaling process, it avoids any changes to the mobile nodes, which essentially means that the mobile nodes can stay simple and all the legacy nodes can obtain the same level of mobility services as the most fancy mobile devices. According the the claim of 3G vendors and operators, this is a key aspect as they learn from their deployment experience.
On the contrary, most mobility support protocols so far focus on mobility support only. And the mobile nodes typically need to update their locations by themselves to the rendezvous points chosen by the user. In these protocols, only the node implementing them can benefit from mobility support. However, the users get more flexibility and freedom, e.g. they can choose whatever mobility services available as long as their software support that protocol, and they can also tune the parameters to get the services that are most suitable to them.
Which one is better? No one knows. We expect them to co-exist as they do today. However, as the technology advances and the hand-hold devices becomes more and more powerful, the latter approach seems to be a much simpler way to move forward with.
| TOC |
The works done on mobility management can also be divided according to their scale into two categories: local mobility management and global mobility management.
Global mobility management is typically supposed to support mobility of unlimited number of nodes in a geographically as well as topologically large area. Consequentially, it pays a lot of attentions to the scalability issues. For the availability concern, it also tries to avoid failure of single point.
Local mobility management on the other hand is designed to work together with global mobility management, and thus focuses more on performance issues, such as shorten handoff delay, reduce handoff loss, short local data path and etc. Since it is typically used in a small scale with no-so-large number of mobile nodes, sometimes the designers can use some fine-tune mechanisms that are not scale with large network (such as host route) to improvement performance. As a side effect of local mobility management, the number of location updates sent by mobile nodes to their global rendezvous points is substantially reduced. Thus, the existence of local mobility management also contribute to the scalability of global mobility management.
One problem of the local mobility management is that it often requires many infrastructure support, such as MAGs in PMIP, or MAPs in HMIP. These kind of local devices are essentially required in all small domains, which can be a huge investment.
Neverthe less, the mobility managements in two scale make it possible for designers to design protocols that fit into specific user requirements; it also enables the gradual deployment of local enhancement while not losing the ability of global roaming. The co-existence of the two seems to be a right choice in the foreseeable future.
Figure 10 shows the classification of the studied protocols according to their serving scale.
_.------------.
,-------. _.-'' `---.
,-+--------+`-. ,-'' +---+ +---------+ `--.
,' |Columbia| `,' |HIP| |Mobile IP| `.
,' +--------+ ,' `. +---+ +---------+ `.
/ +----+ / \ \
/ |PMIP| ,'+-----+\ +---+ +------+ +----+ `.
; +----+ ; |WINMO| : |E2E| |MSM-IP| |NEMO| :
;+----+ ; +-----+ : +---+ +------+ +----+ :
; |HMIP| +-----+ ; : :
| +----+ |TIMIP| +-----------+| +------+ +---+ |
: +-----+ |Global HAHA|; |M-SCTP| |LSR| +----+ ;
: +-----------++-----------+ +------+ +---+ |ILNP| ;
: |Cellular IP| : ; +---------+ +----+ ;
\ +-----------+ `. / +----+ |Mobile Me| ,'
\ +------+ \ / |Sony| +---------+ /
`. |HAWAII| `. ,' +----+ +---------+ ,'
`. +------+ ,'. |Connexion| ,'
`-. ,-' `--. +---------+ _.-'
`-------' `---. _.-''
`------------''
Local Mobility Management Global Mobility Management
| Figure 10 |
| TOC |
In last section we discussed the different directions towards mobility support. We now turn our attention to identify both new opportunities and remaining open issues in providing global scale mobility support for unlimited number of online mobility devices. We are not trying to identify the solutions to these issues, but rather, the goal is to share our opinions and to initiate an open discussion.
| TOC |
The Internet has been running for more than two decades, and the scale of the Internet gets so large that it is impossible to upgrade the whole system over night. As a result, it is not possible for a mobility support system designer to overlook this problem: how important the backward compatibility is?
As one can expect, different designers have different opinions.
Some assume that the other end is by a large chance also mobility capable (as of today, more people are accessing the Internet via mobile devices than a desktop), and thus do not provide backward compatibility at all; but as a tradeoff, the system design becomes much simpler and the data path is always the shortest one. The examples of protocols fall into this categories could be HIP, Mobile Me and etc.
Some take a more conservative approach. They do not want to lose the ability of communicating with legacy nodes during the movement, and they also do not want the rest of the world (meaning, the static nodes) to be bothered by the mobility of mobile nodes. Thus they add a level of indirection at IP layer, and hide the mobility of the mobile nodes. The mobile node thus can benefit from mobility support even when communicating with legacy node. And when the other end also supports mobility, it is also possible to achieve the shortest data path, but with additional complexity. The Mobile IP related protocols fall into this category.
Others want to cater to the inertness of the Internet (and the users) and keep everything status quo, at least from the users' point of view. And thus they take backward compatibility more serious and hide the mobility completely, even for the mobile node itself. Proxy Mobile IP represents this approach.
We all know that backward compatibility is important in system design. But how important is that? How much effort should we make for this issue? At least for now, the answer is not clear yet.
| TOC |
TCP is the most widely used transport layer protocol in the Internet, and the majority of the data traffic today is TCP traffic. In order for the users to benefit from the mobility support, it is of vital importance to keep the established TCP connections undisrupted during the mobile's movement.
Basically there are three approaches to do this: 1) make the mobile node keep its IP address regardless of its point of attachment to the Internet; 2) add a level of indirection and use a static IP address in the TCP pseudo header; 3) modify the TCP protocol.
The routing-based protocols typically make the mobile node keep a stable IP address and thus the first approach is the right one for them. Proxy Mobile IP, however, also takes the first approach. No overhead or complexity is introduced in order to maintain TCP connections in this approach.
A large number of protocols take the second approach. This approach could be achieved via a forwarding agent, as in the case of Mobile IP; or the two end nodes can directly set up a tunnel and use a stable IP address in TCP pseudo headers, as in the case of Mobile Me. In order to maintain the TCP connections undisrupted via this approach, an extra IP header is appended to every data packets, and the possible forwarding agent also introduce problems such as single point of failure, triangle routing, bottleneck of bandwidth and processing power, etc.
A few protocols also take the third approach. In this approach, the TCP protocol itself is modified, either to 1) put an identifier rather than an IP address as connection id or to 2) allow the TCP connections to change the IP address of both ends. HIP, ILNP uses designated ID in TCP layer (they can be in the format of IP address though), while E2E allows using dynamically changing IP addresses. This approach faces the difficulty of changing the TCP protocol itself, which will introduce serious backward compatibility problem.
What is the best choice for the future? Although it seems that most people favor the second approach, whether or not it is the right direction is still uncertain.
| TOC |
As our survey suggests, multiple solutions of mobility support are already running today, and it is almost for sure that the mobility support systems in the future are going to be heterogeneous. However, as of today, the inter-operation between different protocols is still problematic. For example, when a mobile node supporting Mobile IP only wants to communicate with another mobile with only HIP support, both of them can not benefit from mobility support.
This situation reminds us the days before IP were adopted. In that time, the hosts in different networks are not able to communicate with each other. It is the IP that merged the networks and created the Internet, where each host can freely communicate with any other host. Is it necessary to introduce something like IP to the mobility support in the future? Is it possible to design an architecture, so that it glues all the mobility support systems together? We believe the answer to both questions is yes.
The basic idea for the solution is simple, as the famous quote says: "Every problem in Computer Science can be solved by adding a level of indirection. However, the devil is in the details and we still need to figure that out.
| TOC |
Up until now, all our discussions are based on an implication that the underlying global routing systems is not fundamentally changed, that is, the location of the mobile node is always indicated by an IP address. However, recently there are a lot of works that challenge this "classic" idea. Flat-id based routing schemes [VRR] (Caesar, M., Castro, M., Nightingale, E., O'Shea, G., and A. Rowstron, “Virtual Ring Routing: Network Routing Inspired by DHTs,” 2006.) [ROFL] (Caesar, M., Condie, T., Kannan, J., Lakshminarayanan, K., Stoica, I., and S. Shenker, “ROFL: Routing on Flat Labels,” 2006.)[VIR] (Lu, G., Jain, S., Chen, S., and Z. Zhang, “Virtual Id Routing,” 2008.) are especially popular.
With flat-id based routing, mobility as well as multihoming are naturally incorporated. And most of these protocols are mathematically elegant. However, although the notion of flat-id space has been widely adopted in peer-to-peer network, it seems that the Internet is not going to undertake such revolutionary changes (from hierarchical IP Address based routing to flat-id based routing) in the foreseeable future due to the inertness of the extremely huge system and also the backward compatibility problem.
Nevertheless, whether or not flat-id based routing will be a good solution to the mobility support problem, or perhaps more generally, the routing problem is not determined yet.
| TOC |
| [BTMM] | Nakamoto, R., Zhu, Z., Lau, D., and B. Allen, “Understanding Apple's Back to My Mac Service,” UCLA CS217 Project, 2009. |
| [Boeing] | Andrew, L., “A Border Gateway Protocol 4 (BGP-4),” Boeing White Paper, 2006. |
| [CIP] | Valko, A., “Cellular IP: A New Approach to Internet Host Mobility,” ACM SIGCOMM, 1999. |
| [E2E] | Snoeren, A. and H. Balakrishnan, “An End-to-End Approach to Host Mobility,” ACM Mobicom, 2000. |
| [HAHA] | Wakikawa, R., Valadon, G., and J. Murai, “Migrating Home Agents Towards Internet-scale Mobility Deployment,” ACM CoNEXT, 2006. |
| [HAWAII] | Ramjee, R., Varadhan, K., and L. Salgarelli, “HAWAII: A Domain-based Approach for Supporting Mobility in Wide-are Wireless Networks,” IEEE/ACM Transcations on Networking, 2002. |
| [I-TCP] | Bakre, A. and B. Badrinath, “I-TCP: Indirect TCP for Mobile Hosts,” Proceedings of the 15th International Conference on Distributed Computing System, 1995. |
| [ILNP] | Atkinson, R., “An Overive of the Identifier-Locator Network Protocol,” Research Note, 2005. |
| [LISP] | Farinacci, D., Fuller, V., Lewis, D., and D. Meyer, “Locator/ID Separation Protocol (LISP),” draft-farinacci-lisp-12.txt (work in progress), 2009. |
| [LISP-Mobility] | Farinacci, D., Fuller, V., Lewis, D., and D. Meyer, “LISP Mobility Architecture,” draft-meyer-lisp-mn-00.txt (work in progress), 2009. |
| [LSR] | Bhagwat, P. and C. Perkins, “A Mobile Networking System Based on Internet Protocol (IP),” Mobile and Location-Independent Computing Symposium, 1993. |
| [M-SCTP] | Xing, W., Karl, H., and A. Wolisz, “M-SCTP: Design and Prototypical Implementaion of An End-to-End Mobility Concept,” 5th Intl. Workshop on the Internet Challenge, 2002. |
| [MSA] | Ioannidis, J., Duchamp, D., and G. Maguire, “IP-based Protocols for Mobile Internetworking,” ACM SIGCOMM CCR, 1991. |
| [MSM-IP] | Mysore, J. and V. Bharghavan, “A New Multicast-based Architecture for Internet Host Mobility,” ACM Mobicom, 1997. |
| [MSOCKS] | Ferguson, P. and D. Senie, “MSOCKS: An Architecture for Transport Layer Mobility,” IEEE INFOCOM, 1998. |
| [RFC2002] | Perkins, C., “IP Mobility Support,” RFC 2002, October 1996. |
| [RFC2782] | Gulbrandsen, A., Vixie, P., and L. Esibov, “A DNS RR for Specifying the Location of Services (DNS SRV),” RFC 2782, 2000. |
| [RFC3007] | Willington, B., “Secure Domain Name System (DNS) Dynamic Update,” RFC 3007, 2000. |
| [RFC3220] | Perkins, C., “IP Mobility Support for IPv4,” RFC 3220, 2002. |
| [RFC3753] | Manner, J. and M. Kojo, “Mobility Related Terminology.” |
| [RFC3775] | Johnson, D., Perkins, C., and J. Arkko, “IP Mobility Support in IPv6,” RFC 3775, 2004. |
| [RFC3963] | Devarapalli, V., Wakikawa, R., Peterson, A., and P. Thubert, “Network Mobility (NEMO) Basic Support Protocol,” RFC 3963, 2005. |
| [RFC4140] | Soliman, H., Castelluccia, C., Malki, K., and L. Bellier, “Hierarchical Mobile IPv6 Mobility Management (HMIPv6),” RFC 4140, 2005. |
| [RFC5201] | Nikander, P., Moskowitz, R., Jokela, P., and T. Henderson, “Host Identity Protocol,” RFC 5201, 2008. |
| [RFC5213] | Gundavelli, S., Leung, K., Devarapalli, V., Chowdhury, K., and B. Patil, “Proxy Mobile IPv6,” RFC 5213, 2008. |
| [ROFL] | Caesar, M., Condie, T., Kannan, J., Lakshminarayanan, K., Stoica, I., and S. Shenker, “ROFL: Routing on Flat Labels,” ACM Sigcomm, 2006. |
| [Sony] | Teraoka, F., Yokote, Y., and M. Tokro, “A Network Architecture Providing Host Migration Transparency,” ACM SIGCOMM CCR, 1991. |
| [TIMIP] | Grilo, A., Estrela, P., and M. Nunes, “Terminal Independent Mobility For IP,” IEEE Communications Magazine, 2001. |
| [VIR] | Lu, G., Jain, S., Chen, S., and Z. Zhang, “Virtual Id Routing,” ACM MobiArch, 2008. |
| [VRR] | Caesar, M., Castro, M., Nightingale, E., O'Shea, G., and A. Rowstron, “Virtual Ring Routing: Network Routing Inspired by DHTs,” ACM Sigcomm, 2006. |
| [WINMO] | Hu, X., Li, L., Mao, Z., and Y. Yang, “Wide-Area IP Network Mobility,” IEEE INFOCOM, 2008. |
| TOC |
| Zhenkai Zhu | |
| UCLA | |
| 4805 Boelter Hall, UCLA | |
| Los Angeles, CA 90095 | |
| US | |
| Phone: | +1 310 993 7128 |
| Email: | zhenkai@cs.ucla.edu |
| Ryuji Wakikawa | |
| TOYOTA ITC | |
| 465 Bernardo Avenue | |
| Mountain View, CA 94043 | |
| US | |
| Email: | ryuji@jp.toyota-itc.com |
| Lixia Zhang | |
| UCLA | |
| 3713 Boelter Hall, UCLA | |
| Los Angeles, CA 90095 | |
| US | |
| Phone: | +1 310 825 2695 |
| Email: | lixia@cs.ucla.edu |