| Internet-Draft | Anchorless mobility through hICN | July 2020 |
| Auge, et al. | Expires 9 January 2021 | [Page] |
This document first discusses the use of locators and identifiers in mobility management architectures, and their implication on various anchorless properties. A new architecture is then proposed that is purely based on identifiers, and more specifically names as defined in Hybrid-ICN (hICN). The document then focuses on two main cases: the end-point sends data (data producer) or the end-point receives data (data consumer). These two cases are taken into account entirely to provide anchorless mobility management in hICN.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
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."¶
This Internet-Draft will expire on 9 January 2021.¶
Copyright (c) 2020 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 (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶
New usages of the network and the rapid growth of the Mobile Internet calls to reconsider the way we deploy and operate IP networks, where mobility is not built into the design, but rather added as an afterthought. Notable examples are IETF Mobile IP and its variants [RFC5944] and [RFC2275] 3GPP GTP-based architecture [TS29.274], both based on tunnelling and encapsulation.¶
One identified difficulty in proposing mobility models for IP lies in the semantic overloading of IP addresses which are both host identifiers, and locators used for routing. Starting with LISP, the identifier/locator split paradigm has shown promising results in virtue of its scalability properties with respect to routing tables entries, and the possibilities it offers in terms of mobility. Several solutions have been proposed around this concept, namely ILNP, ILSR, and ILA. One common facet of these proposals is to embrace the current trend of a clear separation between the control and data planes, which both allows for distribution of the control infrastructure, as well as anchorless operations in the data plane, including facilitated local breakout.¶
The counterpart of these architectures is an increased dependency on the control plane which is responsible for binding the identifiers used by the application at the edge to the locators used to forward the traffic in the network. Device mobility will typically induce a change of IP address, which makes performance of flows in progress dependent on interactions with those control elements which have to remain globally consistent.¶
In this document, we first propose to clarify protocol descriptions by adopting a new terminology of control-plane and data-plane anchors to characterize anchorless operations, and show their tight coupling with the use of locators and identifiers. This definition serves to position Loc/ID-split architectures with respect to the traditional use of tunnels, before advocating to push this step further and perform mobility management purely based on identifiers. We introduce a mobility approach based on Hybrid ICN (hICN) as described in [I-D.muscariello-intarea-hicn], for which we perform an in-depth analysis of mobility considerations. We show how this proposal can help addressing further challenges faced by networks today, such as multihoming and multipath, while preserving the simplicity and end-to-end design of the current Internet.¶
The consideration of mobility in network design shares the challenges raised for routing for instance in [RFC6115], where the terminology for locators and identifiers is presented.¶
Because they are intrinsically bound to the topological location of a node, session established through locators require additional mechanisms to support mobility, including the presence of anchor points in the network. The notion of anchor and the resulting anchorless properties of mobility schemes are prone to different interpretation depending on the context. The following definitions attempt to reconcile those interpretations and propose a unifying terminology used throughout this document.¶
We distinguish network architectures based on their use of location independent identifiers (or names) and locators for forwarding.¶
Locator-based architectures¶
As mentioned earlier, IP architectures are typically operated based on locators corresponding to the IP addresses of the host interfaces in their respective network attachment, used also as session identifiers. This results in a complex mobility architecture built on top, involving traffic anchors and tunnels to preserve the identifier exposed to the transport layer: IP/IP or GRE tunnels in Mobile IP, and GTP tunnels in 3GPP architectures.¶
The limitations of locator-based schemes in terms of complexity, overhead and efficiency are well-recognized and led to other alternatives to be considered.¶
Locator-ID separation architectures¶
LISP [RFC6830] was the first proposal to distinguish between the usage of IP addresses as locators or identifiers by explicitly defining two namespaces, respectively used for endpoint identification and forwarding. A mapping service is further used to bind an identifiers to a given location, and updated after mobility. From there, several approaches have been defined, either host-based like SHIM6 [RFC5533] or HIP [RFC4423]), or network-based, like LISP [RFC6830], ILSR, ILNP [RFC6740] or ILA [I-D.herbert-intarea-ila] to cite a few.¶
An overview of these approaches and their use in mobility is presented in {?I-D.bogineni-dmm-optimized-mobile-user-plane}}.¶
The main challenge consists in maintaining a (distributed) mapping service at scale, including the synchronization of local caches required for scalability and efficiency.¶
ID-based architectures¶
A third class of approaches exists that redefines IP communication principles (i.e. network and transport layers) around location-independent network identifiers of node/traffic. The interest of such architectures is highlighted in [I-D.vonhugo-5gangip-ip-issues] (referred to as ID-oriented Networking) for it removes the limitations introduced by locators and simplifies the management of mobility. The draft however question the possibility to realize such an architecture where node status and mobility would not affect routing table stability.¶
The work done around Information-Centric Networking (ICN) falls into such class of approaches that we refer to as purely ID-based, also known as name-based [I-D.irtf-icnrg-terminology], although as we will see, mobility management often departs from this principle.¶
ICN is a new networking paradigm centering network communication around named data, rather that host location. Network operations are driven by location-independent data names, rather than location identifiers (IP addresses) to gracefully enable user-to-content communication.¶
Although there exist a few proposals, they share the same set of core principles, resulting in several advantages including a simplified mobility management [RFC7476]. For clarify, this section we focus on hICN [I-D.muscariello-intarea-hicn] an ICN implementation for IPv6.¶
Hybrid ICN (hICN) is an ICN architecture that defines integration of ICN semantics within IPv6. The goal of hICN is to ease ICN insertion in existing IP infrastructure by:¶
hICN architecture is described in [I-D.muscariello-intarea-hicn].¶
hICN, together with MAP-Me [I-D.irtf-icnrg-mapme], forms the basis for the mobility management architecture we describe in the rest of this document. Due to the pull based nature of hICN architecture, we distinguish consumer and producer nodes, for which mobility is handled differently¶
The consumer end-point is the logical communication termination that receives data. Due to the pull-based and connection-less properties of hICN communications, consumer mobility comes natively with ICN. It is indeed sufficient that the consumer reissues pending interests from the new point-of-attachment to continue the communication. Consumer mobility is anchorless by design, and managed without any impact on the transport session. It is however necessary to have an appropriate transport layer on top able to cope with eventual disruptions and path variations caused by the mobility event.¶
The producer end-point is the logical communication termination that sends data. Producer mobility is not natively supported by the architecture, rather handled in different ways according to the selected producer mobility management scheme, some of which diverge from the concept of pure ID-based architecture through their use of locators. Additional procedures have to be performed to maintain reachability as it moves in the network.¶
In fact, many schemes proposed for ICN are adaptations to the vast amount of work made in IP over the last two decades [RFC6301]. Surveys for the ICN family, resp. for CCN/NDN-specific solutions, are available in [SURVEYICN], respectively [SURVEY1] and [SURVEY2]. There has been however a recent trend towards anchorless mobility management, facilitated by ICN design principles, that has led to new proposals and an extension of previous classifications in [I-D.irtf-icnrg-mapme] and [MAPME] to the four following categories:¶
In an hICN network, regular routing protocols such as BGP, ISIS or OSPF can be used for propagating all prefix announcements and populate routers' FIBs. However, these protocols are not appropriate and should not be used to manage name prefix mobility, for scalability and consistency reasons.¶
The default mobility management for hICN is designed following the same principles and protocols as MAP-Me, an anchorless producer mobility management protocol initially proposed for ICN [I-D.irtf-icnrg-mapme] [MAPME]. It builds on an initial forwarding state bootstrapped by the routing protocol, and performs a lightweight path repair process as the producer moves. For simplicity, we refer to it as simply MAP-Me in the rest of the document.¶
In the rest of this section, we describe the specific realization of the protocol in an hICN context. Additional background and details are available in [I-D.irtf-icnrg-mapme] and [MAPME]. The solution is based on a path repair mechanism following mobility events, using dynamic FIB updates. Using data plane mechanisms for ensuring connectivity has been previously proposed in [DATAPLANE] to handle link failures, and has been proven more reactive than relying on typical control plane messaging.¶
Signalization messages follow hICN design principles and use data plane packets for signalization. Signalization messages and acknowledgement are respectively Interest and Data packet (requests and replies) according to hICN terminology [I-D.muscariello-intarea-hicn]. Upon processing of those advertisements, the network will send an acknowledgement back to the producer using the name prefix as the source and the locator of the producer as the destination, plus the sequence number allowing the producer to match which update has been acknowledged.¶
Pending signalization interests that are not acknowledged are retransmitted after a given timeout.¶
The producer is responsible for mobility updates and should be hICN-enabled. It stores a sequence number incremented at each mobility event.¶
Faces in the producer are assumed synchronized with layer 2 adjacencies, upon joining a new point-of-attachment, a new face should be created. Face creation will trigger the increase of the sequence number, and per-prefix advertisement packets to be sent to the joined network.¶
Those advertisements should contain:¶
Upon producer departures from a PoA, the corresponding face is destroyed. If this leads to the removal of the last next hop, then faces that were previously saved are restored in FIB to preserve the original forwarding tree and thus global connectivity.¶
Based on the information transmitted in the packet, and its local the network's local policy, the network might decide to update its forwarding state to reflect this change.¶
The update process consists in updating a few routers on the path between the new and a former point-of-attachment. More precisely, this is done in a purely anchorless fashion by sending a signalling message from the new location towards the name prefix itself. This packet will be forwarded based on the now-stale FIB entries, and will update forwarding entries to point to the ingress interfaces of each traversed router.¶
{fig-mapme-update}} illustrates the situation of a P moving between access routers AR1 and AR2 while serving user requests:¶
P (radio link) AR1 AR2 GW Internet
| | | | |
1 | | | |<~~~~~~~~~~~~~|
| |<~~~~~~~~~~~~~~|~~~~~~~~~~~~~~~| Interest |
|<~~~~~~~~~~~~~~| | | |
|-------------->| | | |
2 [] Data |---------------|-------------->| |
/ | | | |------------->|
| |(attach to AR2)| | | |
-X]...............|...............| | |
3 |===============|==============>o FIB update | |
| Update | |==============>o FIB update |
4 | o<==============|===============| |
| FIB update | | | |
5 | | | |<~~~~~~~~~~~~~|
| | |<~~~~~~~~~~~~~~| |
|<~~~~~~~~~~~~~~|~~~~~~~~~~~~~~~| | |
|---------------|-------------->| | |
| | |-------------->| |
| | | |------------->|
| | | | |
Through this process, MAP-Me trades-off path optimality for a fast, lightweight and loop-free forwarding update. It has been shown in [MAPME] that stretch is bounded and typically kept low in scenarios of interest. This results from the fact that MAP-Me operations preserve the initial structure of the forwarding tree/DAG (direct acyclic graph), by only flipping its edges toward the new producer location.¶
MAP-Me does not perform a routing update. The protocol is complementary to the routing protocol in that it can run in between routing updates to handle producer mobility, at a faster timescale, and with a lower overhead. Routing updates can be performed at a lower-pace, to re-optimize routes once a node has relocated for instance.¶
The update messages should contain :¶
At the reception of advertisement/update packets, each router performs a name-based Longest Prefix Match lookup in FIB to compare sequence number from the received packet and from FIB}. According to that comparison:¶
The update protocol is responsible for reestablishing global connectivity with minimal changes to the FIBs. In order to further improve the reactivity of the scheme and better support QoS constraints of latency-sensitive traffic, we propose an additional mechanism named Notifications. It assumes hICN-enabled routers at the edge, and the existence of links between access routers, which are typically used for handover, and proposes to exploit them during mobility events, or to delay updates when possible.¶
Upon receiving a valid advertisement, the point-of-attachment will remember the presence of the producer, update its corresponding FIB entry but send no update. Previous next hops should be saved and restored upon face deletion so as to preserve the forwarding tree/DAG structure.¶
The rationale is that during mobility events, the producer will move access connected and neighbouring base stations. It will be then sufficient to make a scoped discovery around the last position known to forwarding to find the producer within a few hops.¶
Figure 2 illustrate such as situation where an Interest is sent to the producer before it had the time to complete the update.¶
P (radio link) AR1 AR2 GW Internet
| | | | |
1 | | | |<~~~~~~~~~~~~~|
| |<~~~~~~~~~~~~~~|~~~~~~~~~~~~~~~| Interest |
|<~~~~~~~~~~~~~~| | | |
|-------------->| | | |
2 [].....Data.....-)---------------|-------------->| |
/ | | | |------------->|
| |(attach to AR2)| | | |
-Y]...............|...............+ | |
3 | | | |<~~~~~~~~~~~~~|
4 | X--|<~~~~~~~~~~~~~~|~~~~~~~~~~~~~~~| |
| | | | |
5 |===============|==============>o FIB update | |
| Notification | | | |
| |~~~~~~~~~~~~~~>| (scoped discovery) |
6 |<~~~~~~~~~~~~~~|~~~~~~~~~~~~~~~| | |
7 |---------------|-------------->| | |
| |<--------------| | |
| |---------------|-------------->| |
| | | |------------->|
| | | | |
We now review the potential benefits of the general architecture we have presented using features from the hICN data plane for supporting consumer and producer mobility.¶
Native mobility : This mobility management process follows and exploits hICN design principles. It makes producer mobility native in the architecture by preserving all benefits of hICN even when consumers and/or producers are moving.¶
Anchorless mobility : Such approach belongs to the category of pure ID-based mobility management schemes whose objective is (i) to overcome the limitations of traditional locator-based solutions like Mobile IP (conf)using locators as identifiers and requiring tunnels, and (ii) to remove the need for a global mapping system as the one required by locator-identifier separation solutions. The result is a fully anchorless solution both in the data plane and in the control plane.¶
Local and decentralized mobility management : Mobility updates are handled locally and the routers that are affected are those on path between successive positions of the producer. In particular, remote endpoints are not affected by the event. Mobility is managed in a fully distributed manner and no third party is required. This does not prevent any centralized control (as discussed in [I-D.auge-dmm-hicn-mobility-deployment-options]), but makes the network robust to disconnectivity events.¶
The next section will discuss more in depth the following advantages¶
As emerges from the points raised in the previous section, consumer mobility is transparently supported by an hICN network in virtue of the pull-based model and the way the forwarding path works. After moving, a consumer can just reissue pending interests once attached to the new access router at layer 2, without requiring any more information from L3 and above. This ensures a fast and simple handover, which can be further enhanced through additional mechanisms such as in-network caching. Consumer mobility is fully anchorless with hICN, and does not incur any signalization nor tunneling overhead.¶
This is particularly interesting considering that most mobile users are consumers only (e.g. linear video distribution, or large scale video conferencing where we typically have few presenters and most users are simply consumers).¶
Another aspect of using the unifying hICN architecture in replacement of the traditional tunnel-based mobile core is that it removes the need to maintain state for consumers and producers which are not mobile (eg. IoT sensors), or not currently moving.¶
While this is not strictly linked to mobility, we first illustrate the benefits of caching content close to the edge to reduce user latencies. This is particularly important for wireless networks such as WiFi or LTE which have non-negligible latencies especially during connection setup, or after an idle period. The characteristics of those radio networks are thus of interest for the performance of the mobility architecture as a whole.¶
The example in Figure 3 considers a mobile node that can move access two accesses linked to Access Routers (AR) AR1 and AR2, both connected to the Internet though a common gateway (GW). This same setup will be later used to illustrate the flow of packets during mobility events, eventually specializing AR into AP or eNB when it makes more sense.¶
+-----+ .--.
_,--+ AR1 +--, .-~ ~-( )_ _
+-----+ / +-----+ \ +----+ | ~-. +-----+
| C += =+ GW +---+ Internet \--//--+ P |
+-----+ \_ +-----+ / +----+ \ .' +-----+
'--+ AR2 +--' ~-.__________.-~
+-----+
We represent in Figure 4 a simple data flow between the different entities involved in the communication between the mobile node M, and a remote producer available over the Internet.¶
C AR1 AR2 GW Internet
| | | | |
1 |~~~~~~~~~~~~~~>| | | |
| Interest X |~~~~~~~~~~~~~~~|~~~~~~~~~~~~~~>| |
| | | |~~~~~~~~~~~~~>|
| | | | ...
| | | |<-------------|
| Data X |<--------------|---------------| |
|<--------------| | | |
| | | | |
2 |~~~~~~~~~~~~~~>| | | |
3 | Interest Y |~~~~~~~~~~~~~~~|~~~~~~~~~~~~~~>X Cache hit |
| |<--------------|---------------X |
|<--------------| | | |
| Data Y | | | |
LEGEND:
~~~~>: Interest <---- Data ====> Signalization
....-: L2 detach .....+ L2 attach X failure o event
Numbers on the left refer to the following comments relative to the data flow:¶
Mobile networks might consist in unreliable access technologies, such as WiFi, responsible for packet losses. Figure 5 considers a similar scenario with a lossy channel (eg. WiFi) between the mobile and the first access router.¶
C (radio link) AR1 AR2 GW Internet
| | | | |
1 |~~~~~~~X | | | |
2 |~~~~~~~~~~~~~~>| | | |
| Interest |~~~~~~~~~~~~~~~|~~~~~~~~~~~~~~>| |
| | | |~~~~~~~~~~~~~>|
| | | | ...
| | | |<-------------|
| Data |<--------------|---------------| |
3 | X-------| | | |
| | | | |
4 |~~~~~~~~~~~~~~>X Cache hit | | |
|<--------------X | | |
| | | | |
The same mechanism can be used to recover from mobility losses after the consumer reconnects to the new network as the content will already be available in the cache at the junction router between the two accesses. This is particularly interesting in case of micro-mobility between access routers that are topologically close in the network.¶
This is also the opportunity to manage those losses on behalf of the transport protocol, so that only losses due to congestion are exposed to it, and don't perturb the feedback loop by unnecessarily reducing the transfer rate.¶
C (radio link) AR1 AR2 GW Internet
| | | | |
1 |~~~~~~~~~~~~~~>| | | |
| Interest |~~~~~~~~~~~~~~~|~~~~~~~~~~~~~~>| |
| | | |~~~~~~~~~~~~~>|
|(detach fm AR1)| | | |
2 ..|...............- | |<-------------|
| |<--------------|---------------| |
3 | Data X-| | | |
| | | | |
|(attach to AR2 | | | |
4 ..|...............|...............+ | |
5 |~~~~~~~~~~~~~~~|~~~~~~~~~~~~~~>| | |
6 | | |~~~~~~~~~~~~~~>X Cache hit |
| | |<--------------X |
|---------------|---------------| | |
| | | | |
Because mobility is implemented at layer 3 and is thus agnostic to the physical layer, this allows a mobile consumer to seamlessly switch to a different access layer, eg. WiFi to LTE, following the unreachability of the preferred radio access.¶
Figure 7 illustrates a fallback to LTE which can be performed transparently for the application.¶
C WiFi GW LTE enB PGW Internet | | | | | | 1 |~~~~~~~~~~~>| | | | | | |~~~~~~~~~~~>| | | | | | |~~~~~~~~~~~~|~~~~~~~~~~~~|~~~~~~~~~~~>| | | | | | | | | |<-----------|------------|------------| | |<-----------| | | | |<-----------| | | | | 2 | X | | | | 3 |~~~~~~~~~~~~X~~~~~~~~~~~~|~~~~~~~~~~~>| | | | X | |~~~~~~~~~~~>| | | X | | |~~~~~~~~~~~>| | | | | | | | | | | |<-----------| | | | |<-----------| | |<-----------|------------|------------| | | 4 |~~~~~~~~~~~>|... | | | | | | | | | |
hICN handles consumer mobility from one access to the other (e.g. WiFi to LTE or vice-versa) without any a-priori knowledge of the multiple networks to use as it is the case of MPTCP or QUIC approaches. Moving rate and congestion control at the receiver end results to be a significant advantage w.r.t. all existing alternatives in controlling dynamically multiple and new discovered network accesses in presence of mobility.¶
This use case highlights the importance of having a compatible transport, as the WiFi and LTE paths will have much different characteristics in terms of delay, jitter and capacity.¶
Evaluations of the scheme have shown this scheme to preserve the performance of flows like mobile video distribution up to very high switching rates in the order of a second, not even considering optimization occurring from network support.¶
Mobility is handled transparently at the network layer with very fast handover times, due to the connection-less property of hICN. This makes it possible to offer reliable WiFi connectivity, besides its lossy nature as observed in previous use case and besides the frequency of mobility from one network access to the other.¶
Bandwidth aggregation can be realized dynamically through a congestion-aware load-balancing forwarding strategy at the client, with no a priori knowledge of paths. This is done similarly in the network by hICN forwarders, allowing a combination of multi-homing, multipath and multi-source data transfers. As all the paths are used at the same time, hICN offers the full network capacity to the users and tends to smooth fluctuations due to the radio channels.¶
Over an heterogeneous network access, hICN also offers a simple and cost-effective realization of heterogeneous channel bonding allowing an user to seamlessly roam across different radios or fixed lines (for increased reliability or reduced costs), or aggregate their bandwidth for high-throughput applications such as video streaming.¶
We illustrate a simplified data flows with one mobile consumer C alternating interests between a WiFi and LTE access points. The load balancing strategy would in that case optimize the split ratio between the two access to realize an optimal split. Note that in that case the relative distances on the vertical axis have not been respected here for readability.¶
C WiFi GW LTE enB PGW Internet | | | | | | |~~~~~~~~~~~~|~~~~~~~~~~~~|~~~~~~~~~~~>| | | |~~~~~~~~~~~>| | |~~~~~~~~~~~>|~~~~~~~~~~~>| | |~~~~~~~~~~~>| | | | | | |~~~~~~~~~~~~|~~~~~~~~~~~~|~~~~~~~~~~~>| | | | | | | | | | | |<-----------| | | | |<-----------| | |<-----------|------------|------------| | | | | |<-----------|------------|------------| | |<-----------| | | | |<-----------| | | | | | | | | | | | | | | | |¶
Fine grained control from the application allows fully exploiting available bandwidth, resulting in an aggregated throughput equal to the sum of access throughputs, which is hard to achieve with existing solutions.¶
As a natural consequence of its anchorless behavior, an hICN network can continue to operate in disconnected mode, for local mobility, even though no upstream entity is available.¶
In order to illustrate this, we slightly extend the previous topology in Figure 8 with an additional access network. To show that local mobility induces no traffic on upstream links, we further assume a failure on the link between the gateway (GW) and the Internet.¶
+-----+
_,--+ AR1 +--,
+-----+ / +-----+ \ .-~~~-.
| P += \ .- ~ ~-( )_ _
+-----+ \_ +-----+ +----+ | ~ -.
'--+ AR2 +-----+ GW +----X----+ Internet \
+-----+ +----+ FAIL \ .'
/ ~- . _____________ . -~
+-----+ +-----+ /
| C +------+ AR3 +-'
+-----+ +-----+
The data flow represented in Figure 9 illustrates the communication between a consumer connected to AR3, and a mobile producer moving from AR1 to AR2.¶
C P AR1 AR2 AR3 GW X Internet | | | | | | X | |~~~~~~~~~|~~~~~~~~~~|~~~~~~~~~~|~~~~~~~~~>| | X | | | | | |~~~~~~~~~>| X | | | |<~~~~~~~~~|~~~~~~~~~~|~~~~~~~~~~| X | | |<~~~~~~~~~| | | | X | | |--------->| | | | X | | | |----------|----------|--------->| X | | | | | |<---------| X | |<--------|----------|----------|----------| | X | | | | | | | X | | |..........- | | | X | | |..........|..........+ | | X | | |==========|=========>o | | X NO | | | Update | |==========|=========>o X TRAFFIC | | | o<=========|==========|==========| X | | | | | | | X | |~~~~~~~~~|~~~~~~~~~~|~~~~~~~~~~|~~~~~~~~~>| | X | | | | | |~~~~~~~~~>| X | | | | |<~~~~~~~~~|~~~~~~~~~~| X | | |<~~~~~~~~~|~~~~~~~~~~| | | X | | |----------|--------->| | | X | | | | |----------|--------->| X | | | | | |<---------| X | |<--------|----------|----------|----------| | X | | | | | | | X |
We see that both the data and the signalization remain local to the zone where the mobility occurs, and that communications during mobility are not affected by the failure of the link upstream. This is a sign that the mobile core is not loaded with unnecessary traffic, and that communications remain local, thus improving user flow latencies. The offload of both data and signalization allows reducing the cost of the infrastructure by increasing the diversity of resources used at edge, mutualizing their capacity, and lower requiring network and compute capacity in the mobile core. A direct consequence is also a more robust and reliable network.¶
The realization of the architecture in a partial hICN deployment where some routers are not extended to support hICN mechanisms requires either to introduce additional functionalities or protocol support, or to reuse existing protocols achieving similar objectives (following hICN design).¶
One such example is the combination of ICMP redirect messages and Neighbour Discovery Proxies (NDProxy), that partially realizes the objectives of the update process:¶
Identified concerns might be about the unexpected use of such protocols, the lack of available implementation for NDProxy, and security aspect related to redirect messages. The latter shares the fate of source routing, which has long been advocated against, and has recently gained popularity within the SPRING context. A proper security scheme is certainly the right way to address this problem, and we believe the set of benefits that we have listed are worth reconsidering such aspects.¶
As indicated in previous sections, signalization messages transmitted across trust boundaries must be secured. The choice of the solution will intimately depend on the selected protocols.¶
The use of ICMP packet might allow reusing existing security schemes such as AH headers [RFC4302], or SEND [RFC3971] (and its proxy extensions [RFC6496], [RFC5909]).¶
Alternatively, [SEC] has reviewed standard approaches from the literature and proposes a fast, lightweight and distributed approach that can be applied to MAP-Me and fits its design principles.¶
Both consumer and producer mobility support multiple paths, however the support of mobility for a multihomed producer, is left for future updates of the present document.¶
Similarly, the proposed producer mobility solution is appropriate for the management of micro-mobility; its extension to multiple domains is out of scope.¶
This memo includes no request to IANA.¶