One document matched: draft-tsirtsis-netext-controversy-00.xml


<?xml version="1.0" encoding="US-ASCII"?>
    <?xml-stylesheet type='text/xsl' href='../xml2rfc-1.33/rfc2629.xslt' ?>
    <?rfc toc="yes" ?>
    <?rfc symrefs="yes" ?>
    <?rfc sortrefs="yes"?>
    <?rfc strict="no" ?>
    <?rfc rfcedstyle="yes"?>
    <?rfc subcompact="no"?>
<!DOCTYPE rfc SYSTEM "../xml2rfc-1.33/rfc2629.dtd" [
    <!ENTITY rfc0792 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.0792.xml'>
    <!ENTITY rfc0894 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.0894.xml'>
    <!ENTITY rfc2004 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.2004.xml'>
    <!ENTITY rfc2119 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml'>
    <!ENTITY rfc2225 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.2225.xml'>    
    <!ENTITY rfc5226 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.5226.xml'>
    <!ENTITY rfc3344 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.3344.xml'>  
    <!ENTITY rfc3775 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.3775.xml'>  
    <!ENTITY rfc4830 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.4830.xml'>
    <!ENTITY rfc4861 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.4861.xml'>
    <!ENTITY rfc4960 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.4960.xml'>
    <!ENTITY rfc5213 PUBLIC '' 
      'http://xml.resource.org/public/rfc/bibxml/reference.RFC.5213.xml'>

]>
<rfc category="info" docName="draft-tsirtsis-netext-controversy-00.txt"
    ipr="full3978">
    <front>
        <title abbrev="NETEXT">Discussion of Controversial PMIP Extensions</title>
        <author initials="G." surname="Tsirtsis" fullname="George Tsirtsis">
            <organization>Qualcomm</organization>
            <address>
                <email>tsirtsis@gmail.com</email>
            </address>
        </author>
        <date month="April" year="2009"/>
        <abstract>
            <t>This document discusses the recent controversy regarding PMIP
                extensions for inter-technology handoffs and multihoming. Many
                of the arguments presented below have been discussed in NETEXT
                BOF and subsequent discussions on the mailing list. They are
                written here in an attempt to explain why some of the proposed
                PMIP extensions are so controversial.</t>
        </abstract>
    </front>
    <middle>
        <section title="Requirements Notation">
            <t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
                NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and
                "OPTIONAL" in this document are to be interpreted as described
                in <xref target="RFC2119"/>.</t>
        </section>
        <section title="Introduction">
            <t>This document discusses the recent controversy regarding PMIP
                extensions for inter-technology handoffs and multihoming. Many
                of the arguments presented below have been discussed in NETEXT
                BOF and subsequent discussions on the mailing list. They are
                written here in an attempt to explain why some of the proposed
                PMIP extensions are so controversial. </t>
            <t>Following this introduction, the draft reminds readers of how
                interfaces and hosts are normally viewed by network layer
                protocols, in <xref target="IFs"/>, something that is important
                to keep in mind while reading the rest of the document.</t>
            <t>The draft then establishes what is currently wrong with <xref
                    target="RFC5213"/>, in <xref target="wrong"/>. The draft
                specifically argues that:<list>
                    <t>-PMIPv6 is at best incomplete (and at worst fundamentally
                        broken) because it relies on parameters not available to
                        it by itself or by any other IETF defined protocols.
                        This affects its fundamental operation, in that: <list>
                            <t>a) for single interface mobile nodes it is only
                                applicable for some link layers.</t>
                            <t>b) inter-technology handoffs are broken since the
                                protocol does not define how a MAG/LMA knows
                                when two different link layer technologies are
                                somehow coupled at a given node, nor does it
                                define how the MAG knows if a MN is reachable
                                over a particular link layer. </t>
                            <t>c) The narrow support for multihoming is broken
                                since the protocol does not define how a MAG/LMA
                                knows when handoff vs new mobility session is to
                                be provided to the multiple interfaces.</t>
                        </list>
                    </t>
                </list>
            </t>
            <t> The draft continues with outlining the controversial nature of
                some of the proposed PMIP extensions, in <xref target="sig"/>.
                The arguments presented can be summarized as follows: <list>
                    <t>a) To fix Inter-technology handoffs and to support
                        multihoming properly, MN involvement is required.</t>
                    <t>b) The MN involvement must be at the network layer (or
                        above), and must be defined by the IETF.</t>
                    <t>c) The need for (b) breaks the original reason for
                        defining PMIPv6 in the IETF i.e., a mobility management
                        protocol that does not require MN involvement.</t>
                    <t>d) If (c) is ignored, and PMIPv6 extensions are
                        considered then the following issues arise:<list>
                            <t>- The definition of an MN-MAG protocol,
                                essentially turns MAGs into FA-like entities;
                                but FAs were designed out of MIPv6 for many
                                reasons. Why do we need FA-like functionality
                                back and in what form?</t>
                            <t>- Such a protocol, would bring PMIPv6 in direct
                                competition with MIPv6, contributing to
                                undesirable proliferation of redundant tools in
                                the Internet, which requires justification that
                                is not currently available.</t>
                        </list>
                    </t>
                </list>
            </t>
            <t>The fundamental question then becomes, if MN involvement is
                unavoidable, and if such involvement has to be at network layer,
                what is the justification for extending PMIPv6, when the whole
                premise of PMIPv6 is "no MN involvement", and when MIPv6 fully
                defines a protocol implementing these functions *with* MN
                involvement?</t>
            <t>The draft finally concludes, in <xref target="conc"/>, by first
                arguing that the justification for the proposed work on PMIPv6
                inter-technology and multihoming support, is nonexistent and
                lists specific questions that need to be answered by the various
                factions of the NETEXT community supporting this work (<xref
                    target="just"/>), and by then proposes a way forward for
                PMIPv6 work (<xref target="future"/>).</t>
        </section>
        <section title="The Internet, the Interface, and the Host" anchor="IFs">
            <t>This section provides some background on how "hosts" and
                "interfaces" are viewed in network layer specifications of the
                IETF. This context is important since a lot of the debate around
                PMIPv6 centers around the nature, availability and awareness of
                these entities at various places in the network e.g., MAGs
                according to <xref target="RFC5213"/> have to regularly ask the
                question: "Is interface X part of host Y, and associated with
                Interface Z (i.e., are interfaces X and Y under the same virtual
                interface?"</t>
            <t> The Internet at the network layer has no notion of a host. The
                end points of the Internet are IP Interfaces, identified by
                lower layers as a link layer address (or other low layer
                identifier) and by the higher layers by their IP addresses. The
                link layer end point associated with a given interface can be
                presumed to be on the same or different link with other such end
                point interfaces, but it is mostly impossible to tell which of
                these link layer end points are grouped under the same interface
                and even harder to know which interfaces are grouped inside the
                same computer unit (end node or host). Several protocols do that
                but always at a higher layer (network or above) and always by
                means of explicit signaling; see more on this in <xref
                    target="compl"/>. </t>
            <!--   <t>The Internet is using the network layer as the Inter-networking
                layer, which primary job is to offer a uniform base over which a
                diverse range of transport protocols and applications have been
                developed. The reason this Internetworking layer was successful
                is its minimal set of requirements it sets to the lower layers
                (e.g., <xref target="RFC0894"/>) combined with the uniform
                simplicity it provides to all upper layers.</t> -->
            <t> Each of the link layers on top of which IP can operate, in the
                end represents an interface connected to a link over which the
                Internet Protocol runs.</t>
            <t>The basic IETF protocols defined for end node to access router
                communications (e.g., <xref target="RFC0792"/> and <xref
                    target="RFC4861"/>), are link-specific and treat each link
                layer end point on a given link, as an independent interface
                with no concern of whether two interfaces are part of the same
                node or not. For example <xref target="RFC4861"/> says: <list>
                    <t> " Routers and multihomed hosts have multiple interfaces.
                        The remainder of this document assumes that all sent and
                        received Neighbor Discovery messages refer to the
                        interface of appropriate context.... "</t>
                </list>
            </t>
            <t>Indeed <xref target="RFC4861"/>, in Appendix A, also indicates
                that NDP operation for multihomed nodes (on the same link) is
                "not straightforward". The RFC discusses various implications of
                same-link multihoming with respect to redirect and
                load-balancing functions, and places the responsibility for
                making this work on the end node which is the only entity that
                really knows what interface configuration it has. The RFC of
                course has nothing to say about multihomed nodes on different
                links, since these are in fact invisible to the network layer.</t>
            <t> It is therefore expected that the default behavior of any access
                router is to treat each interface attaching to it as a distinct
                entity, independent from all other interfaces. This is then, one
                of the low level building blocks of the Internet i.e.,
                individual, independent from each other, end-Interfaces. </t>
            <t>On top of this low level infrastructure of interconnected end
                points, however, it is still possible to create more complex
                behaviors that are, importantly, explicitly signaled and thus,
                do not interfere with the fundamental operation of the Internet
                nor do they hinder the operation of simple nodes not equipped to
                handle such additional complexity.<list>
                    <t>And this is where some of the controversy around PMIP
                        starts. </t>
                </list>
            </t>
        </section>
        <section title="What is wrong with PMIP so far" anchor="wrong">
            <t>This section ignores any philosophical issue, of which the author
                has many, against PMIP, and focuses on specific technical issues
                that have to do with <xref target="RFC5213"/>.</t>
            <t>Several technical issues are identified in following subsection,
                after the following observations: <list>
                    <t>1) PMIP is at best incomplete (at worst fundamentally
                        broken), because it relies on information not available
                        by PMIP itself or any other network layer protocols
                        defined by the IETF. This is direct result of the 'no MN
                        involvement" assumption based on which the protocol was
                        built resulting in not defining an MN-MAG protocol at
                        the network layer.</t>
                    <t>2) PMIP for single interface mobiles can work for
                        specific link layers only i.e., when the link layer used
                        presents an identifier for the interface e.g., a MAC
                        address, and when these links allow the MAG to
                        explicitly identify an interface when it attaches to
                        them. </t>
                    <t>3) PMIP's Inter-technology handoff support suffers from
                        (1), but is also incomplete in the sense that <xref
                            target="RFC5213"/> does not define when two
                        different link layer technologies are somehow coupled at
                        a given node (e.g., under a given virtual interface).
                        Inter-technology handoff support breaks the
                        'no-MN-changes' clause of <xref target="RFC4830"/> since
                        it requires some form of virtual interface support in
                        the node, even if there is no MN-MAG protocol to be
                        implemented.</t>
                    <t>4) PMIP as defined in <xref target="RFC5213"/> supports
                        multihoming in a very narrow manner as defined in
                        section 5.4 of the RFC. It requires that, if two
                        interfaces of the same node attach to the same PMIP
                        domain there are two options; a) the two interfaces are
                        under the same mobility session in which case only one
                        can be used to forward traffic to at any one time, b)
                        the two interfaces belong to different mobility
                        sessions, in which case they are treated as interfaces
                        of different MNs (i.e., each has its own HoA and
                        associated binding state in the LMA). This is good, but
                        the PMIP protocol is again incomplete since it does not
                        define when (a) vs (b) is supposed to happen.</t>
                </list>
            </t>
            <section title="Signaling for Complexity" anchor="compl">
                <t>To maintain the Internet's integrity, while allowing for
                    arbitrarily complex behaviors, the way complexity is added
                    on top of the rather simple IP layer is by explicit
                    signaling of additional, more complex protocols. Here are
                    some relevant examples: <list>
                        <t>SCTP (<xref target="RFC4960"/>: binds multiple IP
                            addresses to the same transport layer pipe between
                            two end nodes</t>
                        <t>Mobile IP <xref target="RFC3344"/>, <xref
                                target="RFC3775"/>: binds a stable home address
                            to one (or more) temporary care-of addresses</t>
                    </list>
                </t>
                <t>What is important about these protocols is that they
                    explicitly signal their intentions.</t>
                <t> SCTP is signaled end to end, the routers in the path are
                    oblivious to it, and the peer end node will accept an SCTP
                    pipe only if it also supports SCTP. The complexity of
                    handling multiple IP addresses as part of the same transport
                    pipe has, therefore, no impact on any node in the Internet
                    that does not support SCTP.</t>
                <t>Even more relevant is the example of Mobile IP. End nodes
                    using Mobile IP enabled nodes (Mobile Nodes, Home Agents,
                    CNs, and Foreign Agents in MIPv4), all signal their
                    capabilities. MNs indeed explicitly signal their desire to
                    use Mobile IP which can be ignored or rejected automatically
                    reverting the Mobile Node to operations (with no Mobile IP
                    support). Thus, again, the Mobile IP family of protocols has
                    no impact on any node that does not also support Mobile IP.</t>
                <t>In contrast, PMIPv6 <xref target="RFC5213"/> operates
                    differently. Similarly to Mobile IP it binds a stable home
                    address to one or more temporary MAG addresses. It does
                    that, however, without the explicit IETF-standardized
                    involvement of the MN. The next section discusses some
                    implications of this approach</t>
            </section>
            <section title="PMIP without Host Signaling">
                <t>PMIP was created under the premise of no MN involvement. The
                    NETLMM charter
                    (http://www.ietf.org/html.charters/netlmm-charter.html)
                    clearly indicates: <list>
                        <t>"This working group is tasked with defining a
                            network-based local mobility management protocol,
                            where local IP mobility is handled without
                            involvement from the mobile node."</t>
                    </list>
                </t>
                <t>The "no MN involvement" clause was a fundamental part of
                    justifying the need for a new WG, since the IETF has already
                    defined a whole family of mobility management protocol "with
                    MN involvement", namely <xref target="RFC3344"/> and <xref
                        target="RFC3775"/> with all their extensions. </t>
                <t>Given this "no MN involvement" clause, the solution protocol,
                    (PMIPv6) has to work within these confines resulting in the
                    following characteristics:</t>
                <t>PMIP and Single-Interface end nodes <list>
                        <t>A single Interface node can operate in a PMIP domain
                            without significant problems. This is because when a
                            single interface is attached to a PMIP domain, the
                            end node is made to think it is essentially directly
                            connected to the LMA. If this interface is
                            disconnected from one MAG and connected to another
                            MAG, again the same illusion of direct connectivity
                            to the LMA is preserved.</t>
                        <t>Still, this only works for link layers that present
                            an appropriate interface identifier, e.g., a MAC
                            address. This allows MAGs (or actually LMAs) to
                            recognise a new link with a given MN as a link of a
                            given node handing-off from another MAG.</t>
                    </list>
                </t>
                <t>PMIP and a Multi-Interface end nodes <list>
                        <t>According to <xref target="RFC5213"/> a PMIP MAG
                            somehow has knowledge of whether an interface
                            connected to it falls under one of the following
                            categories, expressed in the form of a Handover Hint
                            defined in section 8.4 of the RFC. <list>
                                <t>1: Attachment over a new interface</t>
                                <t> 2: Handoff between two different interfaces
                                    of the mobile node</t>
                                <t> 3: Handoff between mobile access gateways
                                    for the same interface</t>
                                <t> 4: Handoff state unknown</t>
                                <t> 5: Handoff state not changed
                                    (Re-registration) </t>
                            </list>
                        </t>
                        <!--           <t>The effects of getting this wrong are well known
                            e.g., it could result in allocating the same address
                            two different interfaces, which would break a number
                            of host implementations.</t> -->
                    </list>
                </t>
                <t> Unfortunately, <xref target="RFC5213"/> does not say how
                    this information is obtained. Lack of standardized signaling
                    for the determination of this parameter is causing
                    significant concerns. The concerns stem from the fact that
                    this determination will most likely take place according to
                    some proprietary solution that is not under the control of
                    the IETF and thus, it's accuracy across multiple link layers
                    is at best unpredictable. </t>
                <t> PMIPv6 <xref target="RFC5213"/> attempts to instruct
                    implementers to correctly set the Handoff Indication
                    parameter but the protocol has no internal knowledge of how
                    to set this value. For example, 3GPP has defined layer 2
                    procedures (and assume other link layers support the same
                    layer 2 procedures) for the MN to indicate to the MAG the
                    Handover Hint. This is perfectly reasonable for a body that,
                    unlike the IETF, designs full systems and can place
                    requirement at any layer and entity of that system.</t>
                <t>The IETF, however, has no way of ensuring that a MAG
                    implementation will behave appropriately, independently from
                    the Link Layer used between MN and MAG. This could easily
                    result in incompatibilities since each SDO specific system
                    tends to make assumptions that are not necessarily true, in
                    the general Internet. </t>
            </section>
            <section title="PMIP and Virtual Interfaces">
                <t>An IP Interface is a software construct and as such can take
                    many forms. Many IP interfaces have a one to one
                    relationship with a physical interface e.g., an Ethernet
                    adaptor.</t>
                <t>Almost as often, however, IP Interfaces are virtualized in
                    some way or another. Examples of such Virtual Interfaces are
                    IP in IP Tunnels, VPN Tunnels, Mobile IP Interfaces, and
                    PMIP interfaces.</t>
                <t>All of these virtual interfaces, which are so common in IP
                    networks are either manually configured at the relevant ends
                    (e.g., IP in IP tunnels) or explicitly signaled by a
                    protocols e.g., a VPN client may use IKEv2 to established an
                    IPSEC tunnel, presenting, Mobile IP uses signaling to
                    establish a home address to care-of address binding etc.</t>
                <t>What differentiates PMIP virtual interfaces, is that the
                    formation of a PMIP Interface is not explicitly communicated
                    to anyone at the network layer. It is assumed that MAGs
                    "somehow know" whether a link layer connection from a given
                    end node is under the same PMIP virtual interface with some
                    other such link layer connection. This assumption, that a
                    MAG can somehow know of the existence of a virtual interface
                    or not, has always been a stretch and a point of contention
                    in NETLMM WG.</t>
                <t>Again, not only the need for virtual interfaces brakes the
                    'no MN changes' clause of <xref target="RFC4830"/>, but
                    because of the 'no-MN-invovement' basis for PMIP,the MN can
                    not indicate at the network layer (or above) exactly which
                    links/interfaces relate to each other and in what way,
                    resulting in an incomplete, non-general solution.</t>
            </section>
            <section title="PMIP and Link Layer Signaling">
                <t>It is true that most functions defined at the network layer
                    can also be defined in some form or another at a lower
                    layer. So, there is nothing technically strange or difficult
                    about creating a link layer that supports any number of link
                    layer protocols, which can be used to allow PMIPv6 to
                    perform not only inter-technology handoffs, but also
                    multihoming, flow bindings and most other functions one can
                    build on top of a mobility protocol.</t>
                <t>So, arguments against the idea of relying on link layers for
                    such triggers are not about the feasibility of such an
                    approach for a given link layer. Instead they are about how
                    this works between all the different link layers and node
                    configurations. For example think of an MN with several
                    interfaces. <list>
                        <t>- IF1 and IF2 under PMIP VI1</t>
                        <t>- IF3 separate</t>
                        <t>- IF4 and IF4 under PMIP VI2</t>
                    </list> All the interfaces happen to be connected on the
                    same PMIP domain. Say IF1 (of VI1) and IF4 (of VI2) are
                    already connected and now one more IF is getting connected.
                    When this new IF connects to a MAG, what information does
                    the MAG have to know which one it is and how it relates to
                    other IFs?<list>
                        <t> - A link layer based handover hint by itself is not
                            enough because it does not tell the MAG whether the
                            new IF is IF2 (associated with VI1) or IF5
                            (associated with VI2)</t>
                        <t> - The MNs authorization Identifier (NAI, IMSI or
                            whatever) is clearly not enough because again it
                            does not say whether the new IF is one of the other
                            VI IFs (IF3, IF5) or the independent IF3. </t>
                        <t> - The L2 MAC address (if it exist) again does not
                            provide enough information, since at best it
                            identifies one interface and not the ones related to
                            it.</t>
                    </list>
                </t>
                <t>It should be clear then that link layer signaling is not
                    appropriate for such function since it can never provide
                    information on other links.</t>
                <!--            <t>Note that
                    http://tools.ietf.org/html/draft-devarapalli-netext-multi-interface-support
                    also identifies these limitations and concludes that MN
                    involvement is necessary.</t> -->
                <!--         <t> We can then summarize with the following:</t>
                
                <t>3) Competition with Network Layer:<list>
                        <t>In the end, even if such triggers are implemented,
                            the functionality comes in direct competition to
                            functionality provided at the network layer by
                            Mobile IP. The PMIPv6 vs MIPv6 or, proxy vs client
                            mobility debate has never been about comparing
                            PMIPv6 with MIPv6. It has always been about
                            comparing an IETF standardised network layer
                            protocol MIPv6 with a set of proprietary Link Layer
                            triggers.</t>
                    </list>
                </t>-->
            </section>
        </section>
        <section title="Why extending PMIP is controversial?" anchor="contr">
            <section title="Intertech handoff, Multihoming, and Host Signaling"
                anchor="sig">
                <t> Part of the work proposed for NETEXT WG, is about a) fixing
                    inter-technology handoff support <xref target="RFC5213"/>,
                    which as discussed earlier is broken (or at best incomplete)
                    and b) extending PMIP to support multihoming. Multihoming in
                    this context refers to a node with multiple interfaces (of
                    same or different technology) connected to the Internet.
                    These interfaces can be connected to the same or different
                    links, access routers, or even domains. The extensions
                    proposed would also allow for flow bindings to be used to
                    direct different flows to different links of the same end
                    node. </t>
                <t>This document concludes that to really support
                    Inter-technology handoffs and Multihoming, network layer
                    signaling from the end node is absolutely required. The
                    following summarizes the reasons for this conclusion.</t>
                <t>1) Applicability for ALL link layers: <list>
                        <t>The IETF is concerned with the Internet as a whole,
                            which operates over an ever expanding variety of
                            link layers. Indeed mobility in the Internet means
                            that any node can move from any link to any other
                            link. While seamlessness of such movement is not
                            guaranteed, the network must operate correctly and
                            provide deterministic behaviors to the end nodes.
                            This is why common functions, needed over different
                            link layers, are always defined at the network
                            layer.</t>
                    </list>
                </t>
                <t>2) The impossibility of global knowledge:<list>
                        <t> a) Inter-technology Handoffs: <list>
                                <t>It is not possible for a given link layer,
                                    under a given Interface, to know, and to be
                                    able to signal correctly, which other link
                                    layers and interfaces are associated with it
                                    under a PMIPv6 Virtual Interface. This is
                                    because by definition, a link layer has
                                    access only to information of its own layer.
                                    It is also not possible to have
                                    preconfigured knowledge of such
                                    relationships in the network (a.g., AAA)
                                    since the configuration of an end node can
                                    change at any time, without the AAA being
                                    notified (e.g., an device is changed or
                                    upgraded).</t>
                                <t/>
                            </list>
                        </t>
                        <t> b) Multihoming and Flow Bindings: <list>
                                <t>The same arguments but even stronger hold in
                                    this case. If neither the link layers nor
                                    the network can not be expected to handle
                                    (a) then they can definitely not handle
                                    multihoming and flow bindings which require
                                    a lot more information regarding application
                                    mix running in the end node and
                                    instantaneous condition of different link
                                    layers available.</t>
                                <t/>
                            </list>
                        </t>
                    </list>
                </t>
                <t>It should now be clear that to fix inter-technology handoff
                    support in <xref target="RFC5213"/> and to extend it further
                    to support multihoming and flow bindings, a network layer
                    MN-MAG protocol is required.</t>
            </section>
            <section title="PMIP with Host Signaling">
                <t>If one concludes that host signaling is required for these
                    advanced features, it then should be reasonable to ask:<list>
                        <t> "why host signaling is such a controversial issue
                            for PMIP?" </t>
                    </list>
                </t>
                <t>Some proposals, like [draft-krishnan-netlmm-pmip-sel-00],
                    [draft-larsson-netext-pmipv6-sma-flow-mobility-00], and even
                    [draft-devarapalli-netext-multi-interface-support], openly
                    talk about the need for such IETF defined signaling. At
                    least some, however, in the NETEXT mailing list discussions
                    present significant resistance in admitting this is
                    necessary and the NETEXT BOF presentations where at vague
                    and evasive on the subject.</t>
                <t>There are several reasons for this, discussed in the
                    following subsections.</t>
                <section title="Historic Reasons">
                    <t>The formation of NETLMM WG was strongly resisted by part
                        of the community (see for example
                        [draft-soliman-netlmm-harmful]). So much so that the
                        WG's formation was even the subject of satire and
                        sarcasm; remember NetLemmings?
                        [draft-bombadil-netlemmings].</t>
                    <t> The group was finally formed under the assumption that
                        it would not introduce any changes to the end nodes.
                        Indeed the original NETLMM charter
                        (http://www.ietf.org/html.charters/HISTORY/netlmm-charter.2006-07-24.15.html)
                        said: <list>
                            <t>"...no specific mobile node to network protocol
                                will be required for localized mobility
                                management itself. " </t>
                        </list>
                    </t>
                    <t>The charter continued saying: <list>
                            <t>"...The (PMIPv6) protocol itself will be agnostic
                                with respect to the last hop link layer protocol
                                between the mobile node and the access
                            router."</t>
                        </list>
                    </t>
                </section>
                <section title="MAG ... the new FA?">
                    <t>It is not often expressed explicitly but a PMIP MAG is
                        very similar to a MIPv4 Foreign Agent (<xref
                            target="RFC3344"/>, the main functional difference
                        being that an <xref target="RFC5213"/> compliant MAG
                        does not operate a PMIP-related protocol with the end
                        nodes. Instead it relies on "lower layer" triggers for
                        its operation, and suffers for that, for example by
                        having to deal with rather complex ordering of PBUs
                            <xref target="RFC5213"/>, Section 5.5. </t>
                    <t>Introduction of MN to MAG signaling at the network layer,
                        would indeed make MAGs even more similar to MIPv4 <xref
                            target="RFC3344"/> Foreign Agents (FA). This which
                        would raise two important issues: <list>
                            <t>During the design of MIPv6, foreign agents were
                                considered undesirable. Based on this, FAs were
                                designed out of the MIPv6 protocol. The PMIPv6
                                community has yet to make a case on why we need
                                to re-create them now. Some of the reasons FAs
                                are considered undesirable are: <list>
                                    <t>a)FAs required support from the Access
                                        Networks thus impeding the MN's ability
                                        to move freely without be concerned of
                                        whether or exactly what each access
                                        system supports. </t>
                                    <t>b) FAs are unnecessary in IPv6 since
                                        there is no shortage of address space
                                        (i.e., no need for shared care-of
                                        addresses in MIPv4-FA mode. </t>
                                    <t>c) FAs require a chain of trust between
                                        MNs, FAs, and HAs is increased
                                        complexity compared to the one hop
                                        association of MN-HA in MIPv6.</t>
                                </list>
                            </t>
                            <t>One could, however, make an argument for why
                                removing FAs from this Mobile IP architecture
                                was a mistake and we need them back in. Indeed
                                the introduction of MAGs may point to that
                                conclusion but this debate has never taken place
                                in the IETF in these terms.</t>
                            <t>If the resurrection of FAs in their MAG form can
                                be argued, however, one should also ask the
                                question: "If we really need FA-like
                                functionality in IPv6 mobility management, why
                                is it not defined as part of the mainstream
                                MIPv6 solution itself?" <list>
                                    <t>a) On one hand it is rather obvious for
                                        example, that if we incorporate MAGs
                                        into the Mobile IPv6 protocol structure,
                                        we are much more likely to have better
                                        interoperability and handoff between
                                        domains using MAGs and others that do
                                        not.</t>
                                    <t>b) On the other hand, it is not clear
                                        that any of the reasons FAs were
                                        designed out of MIPv6 are no longer
                                        valid and thus the whole endeavor is
                                        questionable.</t>
                                </list>
                            </t>
                        </list>
                    </t>
                </section>
                <section title="Proliferation of Redundant Tools">
                    <t>Once the IETF developed a tool to handle a specific
                        function e.g., Mobile IP for Mobility, the barrier for
                        additional tools tackling the same problem space is, and
                        should be high.</t>
                    <t>This is both reasonable and necessary to ensure wide
                        adoption of these tools and it is the reason why it is
                        not so easy to define a new transport protocol to
                        replace TCP, or a new Web protocol to replace HTTP. This
                        does not mean that a new tool is impossible, it is just
                        a matter of having a high bar for the adoption of such
                        redundant tools.</t>
                    <t>During the formation of the NETLMM WG a case was made for
                        basic PMIPv6, based on the idea that for a single
                        interface mobility (i.e. intra-technology handovers), it
                        should be possible to define a mobility management
                        protocol that, unlike Mobile IP, does not rely on end
                        node signaling and provides mobility transparently to
                        the end node IP stack, without any host changes (<xref
                            target="RFC4830"/>.</t>
                    <t>As explained in detail in <xref target="sig"/>, these
                        initial assumptions are not possible for
                        inter-technology handoffs and multihoming. </t>
                    <t>It should now be clear that introduction of host
                        signaling in the PMIP protocol defeats the purpose of
                        NETLMM/PMIP's existence since that was to provide
                        mobility transparently to the IP stack of end nodes,
                        unlike what MIPv6 <xref target="RFC3775"/> provides
                        based on host signaling. </t>
                </section>
            </section>
        </section>
        <section title="Conclusions" anchor="conc">
            <section title="Lack of Justification for PMIPv6 Extensions"
                anchor="just">
                <t>During the NETEXT BOF and subsequent discussions on the
                    mailing list a lot of time has been expending around the
                    justification arguments for enhancing PMIPv6 further with
                    inter-technology and multihoming features.
                    [draft-jeyatharan-netext-multihoming-ps-01] attempts to
                    capture such justification but it falls short in the
                    following respects.</t>
                <t>As discussed earlier, the NETLMM WG was established and
                    PMIPv6 protocol was defined based on the "no MN involvement"
                    assumption. The "no MN involvement" assumption restricts the
                    operation of this protocol, and makes advanced features like
                    multihoming and flow movement seem unreasonable in the
                    context of such restrictions.</t>
                <t>Some in the NETEXT community argue that any required MN
                    involvement can be done at lower layers. This part of the
                    community has to address the following issues: <list>
                        <t>- It is a well understood fact that this is not
                            possible for all link layers. It is actually not
                            clear at all which link layers have such capability
                            and whether the triggers are compatible or
                            equivalent between different technologies. </t>
                        <t>- It is important for the integrity of the Internet,
                            that the IETF defines standardized mechanisms
                            providing all the necessary parameters for its own
                            protocols to operate. The author is not aware of any
                            IETF protocol that this is not currently true. This
                            does not seem possible when a protocol depends on
                            external triggers not controlled by any other IETF
                            protocol.</t>
                    </list>
                </t>
                <t>There are at least some in the NETEXT community, however, who
                    recognize that MN involvement at the network layer is
                    necessary to make these advanced features work with PMIPv6.
                    This part of the community has to address a different set of
                    issues. <list>
                        <t>- The definition of an MN-MAG network layer protocol,
                            invalidates the main reason why PMIPv6 was created
                            in the first place (i.e., mobility management with
                            no MN involvement). It was always assumed that MIPv6
                            would then handle any advance functions that require
                            MN involvement. What is the justification for
                            changing this assumption? </t>
                    </list>
                </t>
                <t>As a conclusion the discussion so far comes down to one
                    point. What is the justification for extending PMIPv6's
                    inter-technology and multihoming capabilities? What is
                    missing from the IETF arsenal of tools to handle such
                    features? Why are existing tools not sufficient? These are
                    fundamental questions that MUST be answered before any such
                    work can be taken on.</t>
            </section>
            <section title="What should be done with PMIPv6" anchor="future">
                <t>A number of extensions proposed in the context for NETEXT WG
                    are natural to this work and at the time of writing were
                    approved as part of NETEXT WG Charter, e.g., Bulk
                    Registrations, LMA Redirection etc. These tasks should
                    indeed go ahead.</t>
                <t>It is the opinion of this author, however, that PMIPv6 and
                    all its extensions should be limited to single interface
                    operations, without any inter-technology handoff and without
                    multihoming support beyond what is already defined in <xref
                        target="RFC5213"/>.</t>
                <t> This document actually explains in earlier sections and in
                    detail that <xref target="RFC5213"/> includes a number of
                    features that are incomplete or even broken due to lack of
                    MN-MAG protocol. <xref target="RFC5213"/> should therefore
                    be revised, not to add to these features, but to either
                    change them, so no MN involvement is required, or to remove
                    them from the RFC altogether. </t>
            </section>
        </section>
        <section title="Security Considerations">
            <t>This document does not introduce any Security Considerations </t>
        </section>
        <section title="IANA Considerations">
            <t>This document does not introduce any IANA Considerations</t>
        </section>
        <section title="Acknowledgements">
            <t>Thanks to Marcelo Bagnulo who poked a number of holes on my
                original arguments causing me significant pain :-), but in the
                end helping me refine and clarify them significantly. Also
                thanks to Gerardo Giaretta and Hesham Soliman for useful
                comments and suggestions.</t>
            <t>Many of the arguments presented below are not solely mine but
                have been discussed in the past in NETLMM WG mailing list, and
                more recently in NETEXT BOF and subsequent discussions on the
                mailing list.</t>
        </section>
    </middle>
    <back>
        <references title="Informative References"> &rfc0792; &rfc0894;
            &rfc2004; &rfc2119; &rfc2225; &rfc5226;
            &rfc3344; &rfc3775; &rfc4830; &rfc4861;
            &rfc4960; &rfc5213; </references>
    </back>
</rfc>

PAFTECH AB 2003-20262026-04-23 20:44:26