Requirements for Localised IP Mobility Management T. Pagtzisy, C. E. Williamsz , C. Perkins , A. E Yeginx and P. T. Kirsteiny y Dept. of Computer Science
University College London, UK Email: ft.pagtzis,
[email protected] z MCSR Laboratories, USA Email:
[email protected] x DoCoMo USA Labs Email:
[email protected] Nokia Research Center, USA Email:
[email protected] Abstract— Standard IP mobility management (IP-MM) commonly known as Mobile IP, presents a set of technical challenges in performance and scalability, as the mobile host (MN) varies the handoff rate, due to its mobility pattern, between successive wireless points of attachment; these manifest themselves primarily as increased signalling between the MN and its peers. In addition, they encompass sources of latency external to the mobility protocol; a significant component of this latency is induced by signalling round trip time between the communicating entities while in transit. Recently, Mobile IP has been extended by certain micromobility protocol mechanisms, aiming to alleviate the above performance limitations; they are identified as hierarchical/regional or more generically localised IP mobility management. This paper presents an investigation on requirements for IP localised mobility management (IP-LMM). Based on generic principles derived from the effect of localising mobility control signalling it explores the requirement space essential for robust micro-mobility protocol extensions to base IP mobility models.
Nodes (CNs); they manifest themselves as increased latencies in the signalling path, magnifying the bursty, ’packetswitched’ character of IP, where applications experience nondeterministic delays and congestion. It is well-established that real-time applications impose stringent delay boundaries in IP traffic delivery [2]–[6]. IP mobility-related latency beyond these boundaries, will cause real-time traffic to experience noticeable degradation in quality, while the MN transits within the same or over different internet (ISPs) or content (CSPs) service providers; these are manifested in the form of provisioning network domains. 01 1 0 1 1 0 0 11 00
0 1 0 1 111 00 11 HA000 0 1 0 1 00 11 0 1 0 1
: Access Router : Access Network Router (ANR)
BUs
CN (e.g. HTTP server)
00 11 1 0 0 1 BUs
BU−ACKs BUs
Internet
Domain A
BU−ACKs
BU−ACKs
I. I NTRODUCTION Wireless Internet access is expected to surpass traditional wire-line1 access practices in the near future. The appeal of cellular (PCS) communications has resulted in a significant growth of disparate wireless infrastructures. The latter, under the all-encompassing abstraction of next generation Internet protocols (IPv6), is expected to establish a unified IP wireless access network infrastructure, enabling attainable ubiquitous IP-mobile communication services in a manner similar to the use of public utilities. The above paradigm shift in network access practices instigates expectations for IP-mobile service quality similar to wire-line access. Such vision, however, unveils a number of technical challenges in terms of performance and scalability for base IP mobility management protocol standards [1]; these stem from IP mobility management (IP-MM) signalling between the mobile node (MN) and its communication peers, namely the Home Agent (HA)2 and the Corresponding
Domain C
Home Domain Domain B
BU
MN
Fig. 1. Multiple mobility binding signals per handoff from MN towards its peers in base IP-MM
IP-mobility latency arises when the MN signals its peers with valid IP mobility bindings3 ; as the round-trip time (RTT) between the MN and its peers exceeds real-time delay boundaries4 , the MM signalling component unavoidably injects latency in the communicated traffic of the MN, during its handoff. This introduces the possibility for packet loss or at worst loss of session during communications.
1 such
as Ethernet or DSL the sake of clarity and robustness terms identified in this paper and their abbreviations will be used interchangeably 2 For
3 through 4 on
a cross-domain mobility Binding Update average above 150-200 msec for one way delay
An increase in the transition rate of the MN, between different network domains or links with small coverage area radius, exacerbates the above latency, experienced by the application; while the mobility bindings of the MN must be effected with the same rate at its peers, their rate of effect is tracked by the RTT between the MN and each peer. This is illustrated in figure 1 where the MN roves during a web-browsing session with a peer HTTP server. These peer entities will experience increased signalling per MN handoff transition, the effect of which is dependent on the RTT between the MN and the HTTP server. As the rate of transition increases, long RTTs incur further transient nesting in tunnels; this is due to IP encapsulation effected by the forwarding mechanism of the HA on the last visited link. Frequent tunnel establishment can introduce additional delays during the MN handoff, causing additional packet loss from delayed packet delivery to real-time applications. To alleviate the above mobility management issues, a number of Localised Mobility Management (LMM) protocols have been proposed in the Mobile IP WG for next generation Internet (IPv6) in the greater context of micro-mobility [7]. These LMM proposals [8], [9] currently address only partly issues dealing performance, scalability or reliability. A clear set of requirements is, thus, essential to approach robustness in the design of any localised mobility solution for future IPv6 wireless access networks. This paper is organised as follows: Section II presents the principles behind localised mobility management schemes. Section III identifies requirements essential for acceptable signalling performance; section IV presents requirements relevant to issues of LMM scalability. Section V addresses interoperability considerations for LMM mechanisms; section VI identifies security issues that must be addressed by LMM services. We conclude with a summary for the established requirements set for LMM in Section VII.
The distinction between types of transitions by an LMM scheme, extends standard5 IP-MM by identifying two separate classes of control signalling messages with respect to the locality of the notion of an administrative domain: localised or regional mobility signals (intra-domain): bound within a single administrative domain and primarily sent towards some LMA. Localised mobility signals are generally referred as Regional Binding Updates (RBU). global mobility signals (inter-domain): communicated across different administrative domains with ultimate destination the true peer entities, namely the HA and CNs of the MN. Global mobility signals are commonly known as Binding Updates (BUs). The separation of IP-MM signalling, implements effectively a level of indirection in two respects: signalling and addressing of the MN. This design approach is supported by Kirby [10] identifying that 69% of the mobility of a user remains local with respect to a reference location. For signalling, the indirection allows the end-to-end message overhead experienced by peers to be reduced for the time period that the MN remains mobile within some visiting domain. As illustrated in figure 2, the MN informs its peers of its change of domain-wide CoA when it moves from H-CA to V-CA by means of a global BU while the LMA is informed accordingly through an R-BU signal. For the remaining time that the MN transits through the particular network domain no global BU signals are sent towards the peers; instead an RBU signal is sent to the edge LMA.
Backbone RTT component (variable)
0 1 0 1 0 1 0 1 11 00 0 1 0 1 000 111 00 11 0 1 0 1 00 11 0 1 0 1 00 11 0 1 0 1
HA CN
Internet edge LMA
00 11 00 11 000 111 11111 00000 000 00000111 11111
Scope of visibility for RCoA
domain edge
II. P RINCIPLES
OF
L OCALISED M OBILITY M ANAGEMENT
A localised mobility management scheme, strives to minimise excessive IP-MM signalling (BUs) towards its peers, caused by frequent change of care-of address (CoA). This is achieved by establishing an entity instance of the home domain, similar to the Home Agent, into a visited network domain hosting the MN; such entity is generally identified as Localised Mobility Management Agent (LMA). By ensconcing the LMA closer to the MN as shown in figure 2, MN transitions are now managed according to the topological locality of the visited domain as opposed to the total number of hops between the MN and its Home Agent. Network transitions for an MN can now be characterised as:
intra-domain transitions. The MN transits within a set of subnets/sub-domains that define the topology of a single administrative domain. Standard IP-MM does not distinguish between intra and inter-domain transitions. inter-domain transitions. The MN transits between different administrative domains.
Scope of visibility for LCoA
base MIPv6
LMM
: Coverage Area (CA) : Access Network Router (ANR) : Access Router (AR)
V−CA
mobility management signaling RTT (BU)
MN
: Regional BU (RBU) : RBU−ACK : Global BU (BU)
H−CA
Home Domain
: Global BU−ACK
Visited Domain
Fig. 2. Domain differentiation in mobility binding signalling for signalling reductions in IP-LMM
The addressing indirection effects at the MN a mobilityglobal or regional care-of address (RCoA) for its entire mobility pattern within the visited domain, and a mobilitylocal or on-link care-of address (LCoA) to the MN. While the RCoA address is globally visible to peers and thus, locates the MN with domain-wide accuracy, the LCoA address is 5 the terms standard or base IP-MM refer to the Mobile IPv6 protocol standard [1] and will be used interchangeably
visible explicitly only within the visited domain; this allows for location privacy potentials for the MN. Localising the mobility bindings carries further associated benefits on the communicated packet flows between the MN and its peers. These are manifested primarily as reduction of signalling latencies by localising IP-MM for the MN. In this light, an LMM protocol may contribute towards sustaining stringent packet delivery constraints for real-time delivery of IP traffic such as interactive multimedia; nevertheless, existence of an LMM mechanism by itself does not warrant realtime performance in such classes of applications. Real-time performance depends on other factors such as IP connectivity latencies [11], [12] with respect to the LCoA; such factors are typically outside the scope of core LMM mechanisms. Investigating the protocols internals of the generic LMM function, in contrast to standard IP-Mobility is important in identifying solid requirements or boundaries that standardise a well-founded common denominator in LMM protocol design. It is with such foundations that we can seek to ensure that the resulting LMM solution will best preserve the fundamental philosophies and architectural principles of the Internet currently in practice. In the following sections we present a requirements set that encompasses essential considerations for the design of an LMM protocol mechanism; the requirements derived, are relevant to both IPv4 and IPv6 mobility-enabled network infrastructures. The set of requirements identified in this paper, is currently going through the review process for standardisation from the Mobile IP WG of the IETF; it is available as a working group item in [13]. III. P ERFORMANCE
AND
S IGNALLING
Reduction of control signalling towards the peers of the MN has been the fundamental design consideration for LMM extensions in supporting scalable signalling over IP-mobility. However, the signalling efficiency of an LMM mechanism is effectively dependent on the degree of localisation of IP mobility signalling against global mobility signals; such factor translates to the frequency of inter-domain transitions against the total number of handoffs performed by the MN. This is verified formally by previous work in [14]; it has shown that the signalling performance of the LMM function is superior to that of standard IP-mobility, iff the domain crossing in the travel path of the MN exceeds a lower average threshold value bn l of the form:
bl n
>
h
n
( + 1)
r k
;
^
l h
n ; n ; r; k
2
(1)
l is the average number of intra-domain handoffs where n identifying the crossing of the MN within a domain, nh the total number of handoffs within the travel path of the MN, the average number r the number of domains crossed and k of corresponding nodes (CNs) communicating with the MN. From 1 it becomes clear that the signalling efficiency of the LMM function is dependent on the rate of arrival of new CNs (k) at the MN and the ratio of inter-domain handoffs over the total number of handoffs during the travel path of the MN. In
fact we can provide an upper bound on the number of interdomain handoffs, as a function of the average number of CNs communicating with the MN during its movement between domains, by restructuring equation (4) of [14] and 2 as follows:
h =r +
n
C
Xr l=0
l
(2)
n
lmm =(k + 1)nh + [nh (k + 1)nh
(k
(k + 1)
(k + 1)(nh
Xr
(k + 1)r + nh
l=0
r X + 1)
Xr l=0
l=0
l℄
n
l + nh
n
l ) + nh
n
(3)
by using the equality (3) the upper bound of inter-domain handoffs that sustains signalling efficiency of the LMM function becomes: C
lmm
(k + 1)r + nh kr + r
de r
< C
