One document matched: draft-ietf-rmt-sec-discussion-02.xml


<?xml version="1.0" encoding="US-ASCII"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<rfc category="info" docName="draft-ietf-rmt-sec-discussion-02" ipr="full3978">
  <!--
<?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?>
-->

  <?xml-stylesheet type='text/xsl'
href='http://xml.resource.org/authoring/rfc2629.xslt' ?>

  <?rfc toc="yes"?>

  <?rfc tocompact="yes"?>

  <?rfc tocdepth="3"?>

  <?rfc tocindent="yes"?>

  <?rfc symrefs="yes"?>

  <?rfc sortrefs="yes"?>

  <?rfc comments="yes"?>

  <?rfc inline="yes"?>

  <?rfc compact="yes"?>

  <?rfc subcompact="no"?>

  <?rfc iprnotified="no" ?>

  <?rfc strict="yes" ?>

  <front>
    <title abbrev="Security and RMT Protocols">Security and Reliable Multicast
    Transport Protocols: Discussions and Guidelines</title>

    <author fullname="Brian Adamson" initials="B." surname="Adamson">
      <organization>Naval Research Laboratory</organization>

      <address>
        <postal>
          <street></street>

          <city>Washington, DC</city>

          <code>20375</code>

          <country>USA</country>
        </postal>

        <email>adamson@itd.nrl.navy.mil</email>

        <uri>http://cs.itd.nrl.navy.mil</uri>
      </address>
    </author>

    <author fullname="Vincent Roca" initials="V." surname="Roca">
      <organization>INRIA</organization>

      <address>
        <postal>
          <street>655, av. de l'Europe</street>

          <street>Zirst; Montbonnot</street>

          <city>ST ISMIER cedex</city>

          <code>38334</code>

          <country>France</country>
        </postal>

        <email>vincent.roca@inria.fr</email>

        <uri>http://planete.inrialpes.fr/~roca/</uri>
      </address>
    </author>

    <date day="14" month="July" year="2008" />

    <area>Transport</area>

    <workgroup>RMT</workgroup>

    <keyword>security analysis</keyword>

    <abstract>
      <t>This document describes general security considerations for the
      Reliable Multicast Transport (RMT) Working Group set of building blocks
      and protocols. An emphasis is placed on risks that might be resolved in
      the scope of transport protocol design. However, relevant security
      issues related to IP Multicast control-plane and other concerns not
      strictly within the scope of reliable transport protocol design are also
      discussed. The document also begins an exploration of approaches that
      could be embraced to mitigate these risks. The purpose of this document
      is to provide a consolidated security discussion and provide a basis for
      further discussion and potential resolution of any significant security
      issues that may exist in the current set of RMT standards.</t>
    </abstract>
  </front>

  <middle>
    <section anchor="intro" title="Introduction">
      <t>The Reliable Multicast Transport (RMT) Working Group has produced a
      set of building block (BB) and protocol instantiation (PI)
      specifications for reliable multicast data transport. Some present PIs
      defined within the scope of RMT include <xref
      target="RFC3450">ALC</xref><xref
      target="draft-ietf-rmt-pi-alc-revised"></xref> , <xref
      target="RFC3940">NORM</xref>, and the <xref
      target="RFC3926">FLUTE</xref> application that is built on top of ALC.
      These can be considered "Content Delivery Protocols" (CDP) as described
      in<xref target="Neumann05"> </xref>. In this document, the term CDP will
      refer indifferently to either ALC or NORM, with their associated
      BBs.</t>

      <t>The use of these BBs and PIs raises some new security risks. For
      instance, these protocols share a novel set of Forward Error Correction
      (FEC) and congestion control building blocks that present some new
      capabilities for Internet transport, but may also pose some new security
      risks. Yet some security risks are not related to the particular BBs
      used by the PIs, but are more general. Reliable multicast transport
      sessions are expected to involve at least one sender and multiple
      receivers. Thus, the risk of and avenues to attack are implicitly
      greater than that of point-to-point (unicast) transport sessions. Also
      the nature of IP multicast can expose other coexistent network flows and
      services to risk if malicious users exploit it. The classic any-source
      multicast (ASM) model of multicast routing allows any host to join an IP
      multicast group and send traffic to that group. This poses many
      potential security challenges. And, while the emerging single-source
      multicast (SSM) model that allows only a single sender to send traffic
      to a group simplifies some challenges, there remain some specific
      issues. For instance, possible areas of attack include those against the
      control plane where malicious hosts join IP multicast groups to cause
      multicast traffic to be directed to parts of the network where it is not
      needed or desired. This can indirectly cause denial-of-service (DoS) to
      other network flows. Also, attackers may transmit erroneous or corrupt
      messages to the group or employ strategies such as replay attack within
      the "data plane" of protocol operation.</t>

      <t>The goals of this document are therefore to: <list style="numbers">
          <t>Define the possible general security goals; i.e., define what we
          want to protect, i.e. the network itself, and/or the protocol,
          and/or the content.</t>

          <t>List the possible elementary security services that will make it
          possible to fullfil the general security goals. Some of these
          services are generic (e.g. object and/or packet integrity), while
          others are specific to RMT protocols (e.g. congestion control
          specific security schemes).</t>

          <t>List some technological building blocks and solutions that can
          provide the desired security services.</t>

          <t>Highlight the CDP specificities that will impact security and
          define some use-cases. Indeed, the set of solutions proposed to
          fulfill the security goals will greatly be impacted by the target
          use case.</t>
        </list> In some cases, the existing RMT documents already discuss the
      risks and outline approaches to solve them, at least partially. The
      purpose of this document is to consolidate this content and provide a
      basis for further discussion and potential resolution of any significant
      security issues that may exist.</t>

      <section title="Conventions Used in this Document">
        <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"></xref>.</t>
      </section>
    </section>

    <section title="Quick Introduction to RMT Protocols and their Use">
      <section title="The Two Families of CDP">
        <t>The ALC/LCT and NORM classes of CDP are designed to reliably
        deliver nncontent to a group of multicast receivers, but each with a
        different set of features and limitations. The ALC/LCT class supports
        a unidirectional delivery model where there is no feedback from the
        receivers to senders, relying upon transmission redundant FEC coding
        capable of recovering missing packet content for reliability. With
        appropriate FEC encoding techniques, the transmission stream can
        deliver data at different rates to different receivers, thus offering
        the potential for multirate congestion control. This allows
        scalability for delivery of bulk content to potentially very large
        group sizes. While NORM supports the same use of FEC as ALC/LCT, it
        leverages Negative Acknowledgement techniques to control the senders'
        transmission of content. The advantage of NORM is that the sender need
        not transmit any more information than necessary to satisfy the
        receivers' need to achive reliable transfer. But, while NORM specifies
        feedback control techniques to allow it to scale to considerably large
        group sizes, it is not as massively scalable as the ALC/LCT approach
        when feedback is used. Additionally, the NORM feedback control
        mechanisms add some additional header content and protocol
        implementation complexity as compared ALC/LCT. The appropriate choice
        of CDP depends upon application needs, deployment constraints, and
        network connectivity considerations. And while there are many common
        security considerations for these two classes of CDP, there are also
        some unique considerations for each.</t>
      </section>

      <section anchor="protocolCharacteristics"
               title="RMT Protocol Characteristics">
        <t>This section focuses on the RMT protocol characteristics that will
        impact the choice of the technological building blocks, and the way
        they can be applied. Both ALC and NORM have been designed with
        receiver group size scalability. While ALC targets massively scalable
        sessions (e.g. with millions of receivers), NORM is less ambitious,
        essentially because of the use of feedback messages to the source.
        Ideally, the use of security mechanisms should not break these
        scalability features.</t>

        <t>The ALC and NORM protocols differ in the communication paths: <list
            style="symbols">
            <t>sender to receivers: ALC and NORM, for bulk data transfer and
            signaling messages;</t>

            <t>receivers to sender: NORM only, for feedback messages;</t>

            <t>receivers to receivers: NORM only for control messages;</t>
          </list> But the fact that ALC is capable of working on top of purely
        unidirectional networks does not mean that no back-channel will be
        available (see <xref target="useCaseCharacteristics"></xref>). The
        NORM and ALC protocols support a variety of content delivery models
        where transport may be carefully coordinated among the sender and
        receivers or with looser coordination and interaction. This leads to a
        number of different use cases for these protocols.</t>
      </section>

      <section anchor="useCaseCharacteristics"
               title="Target Use Case Characteristics">
        <t>This section focuses on the target use cases and their special
        characteristics. These details will impact both the choice of the
        technological building blocks and the way they can be applied. One can
        distinguish the following use case features: <list style="symbols">
            <t>Purely unidirectional transport versus symmetric bidirectional
            transport versus asymmetric bidirectional transport. Most of the
            time, the amount of traffic flowing to the source is limited, and
            one can overlook whether the transport channel is symmetric or
            not. The nature of the underlying transport channel is of
            paramount importance, since many security building blocks will
            require a bidirectional communication;</t>

            <t>Massively scalable versus moderately scalable session. Here we
            do not define precisely what the terms "massively scalable" and
            "moderately scalable" mean.</t>

            <t>Known set of receivers versus unknown set of receivers: I.e.,
            does the source know at any point of time the set of receivers or
            not? Of course, knowing the set of receivers is usually not
            compatible with massively scalable sessions;</t>

            <t>Dynamic set of receivers versus fixed set of receivers: I.e.,
            does the source know at some point of time the maximum set of
            receivers or will it evolve dynamically?</t>

            <t>High rate data flow versus small rate data flow: Some security
            building blocks are CPU-intensive and are therefore incompatible
            with high data rate sessions (e.g. solutions that digitally sign
            all packets sent).</t>

            <t>Protocol stack available at both ends: A solution that requires
            some unusual features within the protocol stack will not always be
            usable. Some target environments (e.g. embedded systems) provide a
            minimum set of features and extending them (e.g. to add IPsec) is
            not necessarily realistic;</t>

            <t>Multicast routing and other layer-3 protocols in use: E.g., SSM
            routing is often seen as one of the key service to improve the
            security within multicast sessions, and some security building
            blocks require specialized versions of layer-3 protocols (e.g.
            IGMP/MLD with security extensions). In some cases these
            assumptions might not be realistic.</t>
          </list>Depending on the target goal and the associated security
        building block used, other features might be of importance. For
        instance TESLA requires a loose time synchronization between the
        source and the receivers. Several possible techniques are available to
        provide this, but some of them may be feasible only if the target use
        case has the appropriate characteristics.</t>
      </section>
    </section>

    <section title="Known Security Threats">
      <t>The IP architecture provides common access to notional control and
      data planes to both end and intermediate systems. For the purposes of
      discussion here, the "control plane" mechanisms are considered those
      with message exchanges between end systems (typically computers) and
      intermediate systems (typically routers) (or among intermediate systems)
      while the "data plane" encompasses messages exchanged among end systems,
      usually pertaining to the transfer of application data. The security
      threats described here are introduced within the taxonomy of control
      plane and data plane IP mechanisms.</t>

      <section title="Control-Plane Attacks">
        <t>In this discussion, "control-plane" in the context of Internet
        Protocol systems refers to signaling among end systems and
        intermediate systems to facilitate routing and forwarding of packets.
        For IP multicast, this notably includes Internet Group Management
        Protocol (IGMP) and multicast routing protocol messaging. While
        control-plane attacks may be considered outside of the scope of the
        transport protocol specfications discussed here, it is important to
        understand the potential impact of such attacks with respect to the
        deployment and operation of these protocols. For example, awareness of
        possible IP Multicast control-plane manipulation that can lead to
        unauthorized (or unexpected) monitoring of data plane traffic by
        malicious users may lead a transport application or protocol
        implementation to support encryption to ensure data confidentiality
        and/or privacy. Also, these types of attack also have bearing on
        assessing the real risks of potentially more complex attacks against
        the transport mechanisms themselves. In some cases, the solutions to
        these control-plane risk areas may reduce the impact or possibility of
        some data-plane attacks that are discussed in this document.</t>

        <t>The presence of these types of attack may necessitate that
        policy-based controls be emplaced in routers to limit the distribution
        (including transmission and reception) of multicast traffic (on a
        group-wise and/or traffic volume basis) to different parts of the
        network. Such policy-based controls are beyond the scope of the RMT
        protocol specifications. However, such network protection mechanisms
        may reduce the opportunities for or effectiveness of of some of the
        data-plane attacks discussed later. For example, reverse-path checks
        can significantly limit opportunities for attackers to conduct replay
        attacks when hosts actually do use IPSec. Also, future IP Multicast
        control protocols may wish to consider providing security mechanism to
        prevent unauthorized monitoring or manipulation of messages related to
        group membership, routing, and activity. The sections below describe
        some variants of control-plane attacks.</t>

        <section title="Control Plane Monitoring">
          <t>While this may not be a direct attack on the transport system, it
          may be possible for an attacker to gain useful information in
          advancing attack goals by monitoring IP Multicast control plane
          traffic including group membership and multicast routing
          information. Indentification of hosts and/or routers participating
          in specific multicast groups may readily identify systems vulnerable
          to protocol-specific exploitation. And, with regards to user privacy
          concerns, such "side information" may be relevant to this emerging
          aspect of network security.</t>
        </section>

        <section title="Unauthorized (or Malicious) Group Membership">
          <t>One of the simplest attacks is that where a malicious host joins
          an IP multicast group so that potentially unwanted traffic is routed
          to the host's network interface. This type of attack can turn a
          legitimate source of IP traffic into a "attacker" without requiring
          any access privileges to the source host or routers involved. This
          type of attack can be used for denial-of-service purposes or for the
          real attacker (the malicious joiner) to gain access to the
          information content being sent. Similarly, some routing protocols
          may permit any sender (whether joined to the specific group or not)
          to transmit messages to a multicast group.</t>

          <t>It is possible that malicious hosts could also spoof IGMP
          messages, joining groups posing as legitimate hosts (or spoof source
          traffic from legitimate hosts). This may be done at intermediate
          locations in the network or by hosts co-resident with the authorized
          hosts on local area networks. Such spoofing could be done by raw
          packet generation or with replay of previously-recorded control
          messages. For the sake of completeness, it should be noted that
          multicast routing protocol control messaging may be subject to
          similar threats if insufficient protocol security mechanisms are
          enabled in the routing infrastructure.</t>
        </section>
      </section>

      <section title="Data-Plane Attacks">
        <t>This section discusses some types of active attacks that might be
        conducted "in-band" with respect to the reliable multicast transport
        protocol operating within the data plane of network data transfer.
        I.e., the "data-plane" here refers to IP packets containing end-to-end
        transport content to support the reliable multicast transfer. The
        passive attack of unauthorized data-plan monitoring is discussed above
        since such activity might be made possible by the vulnerabilities of
        the IP Multicast control plane. To cover the two classes of RMT
        protocols, the active data-plane attacks are categorized as 1) those
        where the attacker generates messages posing as a data sender, and 2)
        those where the attacker generates messages posing as a receiver
        providing feedback to the sender(s) or group. Additionally, a common
        threat to protocol operation is that of brute-force, rogue packet
        generation. This is discussed briefly below, but the more subtle
        attacks that might be conducted are given more attention as those fall
        within the scope of the RMT transport protocol design. Additionally,
        special consideration is given to that of the "replay attack" [see
        <xref target="ReplayAttacks"></xref>], as it can be applied across
        these different categories.</t>

        <section title="Rogue Traffic Generation">
          <t>If an attacker is able to successfully inject packets into the
          multicast distribution tree, one obvious denial-of-service attack is
          for the attacker to generate a large volume of apparently
          authenticate (and when authentication mechanisms are used, a
          "replay" attack strategy might be used) traffic. The impact of this
          type of attack can be significant since the potential for routers to
          relay the traffic to multiple portions of a networks (as compared to
          a single unicast routing path). However, other than the amplified
          negative impact to the network, this type of attack is no different
          than what is possible with rogue unicast packet generation and
          similar measures used to protect the network from such attacks could
          be used to contain this type of brute-force attack. Of course, the
          pragmatic question of whether current implementations of such
          protection mechanisms support IP Multicast SHOULD be considered.</t>
        </section>

        <section title="Sender Message Spoofing">
          <t>These types of attacks are applicable to both general types of
          RMT protocols: ALC (sender-only transmission) and NORM
          (sender-receiver exhanges). Without an authentication mechanism, an
          attacker can easily generate sender messages that could disrupt a
          reliable multicast transfer session. And with FEC-based transport
          mechanisms, a single packet with an apparently-correct FEC payload
          identifier <xref target="RFC3452"></xref> but a corrupted FEC
          payload could potentially render an entire block of transported data
          invalid. Thus, a modest injection rate of corrupt traffic could
          cause severe impairment of data transport. Additionallly, such
          invalid sender packets could convey out-of-bound indices (e.g. bad
          symbol or block identifiers) that can lead to buffer overflow
          exploits or similar issues in implementations that insufficient
          check for invalid data.</t>

          <t>An indirect use of sender message spoofing would be to generate
          messages that would cause receivers to take inappropriate
          congestion-control action. In the case of the layered congestion
          control mechanisms proposed for ALC use, this could lead to the
          receivers erroneusly leaving groups associated with higher bandwidth
          transport layers and suffering unnecessarily low transport rates.
          Similarly, receivers may be misled to join inappropriate groups
          directing unwanted traffic to their part of the network. Attacks
          with similar effect could be conducted against the TFMCC approach
          proposed for NORM operation with spoofing of sender messages
          carrying congestion control state to receivers.</t>
        </section>

        <section title="Receiver Message Spoofing">
          <t>These atacks are limited to RMT protocols that use feedback from
          receivers in the group to influence sender and other receiver
          operation. In the NORM protocol, this includes
          negative-acknowledgement (NACK) messages fed back to the sender to
          achieve reliable transfer, congestion control feedback content, and
          the optional positive acknowledgement features of the specification.
          It is also important to note that for ASM operation, NORM receivers
          pay attention to the messages of other receivers for the purpose of
          suppression to avoid feedback implosion as group size grows
          large.</t>

          <t>An attacker that can generate false feedback can manipulate the
          NORM sender to unnecessarily transmit repair information and reduce
          the goodput of the reliable transfer regardless of the sender's
          transmit rate. Contrived congestion control feedback could also
          cause the sender to transmit at an unfairly low rate.</t>

          <t>As mentioned, spoofed receiver messaging may not be directed only
          at senders, but also at receivers participating in the session. For
          example, an attacker may direct phony receiver feedback messages to
          selected receivers in the group causing those receivers to suppress
          feedback that might have otherwise been transmitted. This attack
          could compromise the ability of those receivers to achieve reliable
          transfer. Also, suppressed congestion control feedback could cause
          the sender to perhaps transmit at a rate unfair to those attacked
          receivers if their fair congestion control rate were lower than
          other receivers in the group.</t>
        </section>

        <section anchor="ReplayAttacks" title="Replay Attacks">
          <t>The infamous "replay attack" (injection of a previously
          transmitted packet (or at least its payload) into the reliable
          transport group or directly to one or more of its participants) is
          given special attention here because of the special consequences it
          can have on RMT protocol operation. Without specific protection
          mechanisms against replay (e.g. duplicate message detection), it is
          possible for these attacks to be successful even when security
          mechanisms such as packet authentication and/or encryption are
          employed.</t>

          <section title="Replay of Sender Messages">
            <t>Generally, replay of recent protocol messages from the sender
            will not harm transport, and could potentially assist it, unless
            it is of sufficient volume to result in the same type of impact as
            the "rogue traffic generation" described above. However, it is
            possible that replay of sufficiently old messages may cause
            receivers to think they are "out of sync" with the sender and
            reset state, compromising the transfer. Also, if sender transport
            data identifiers are reused (object identifiers, FEC payload
            identifiers, etc), it is possible that replay of old messages
            could corrupt data of a current transfer.</t>
          </section>

          <section title="Replay of Receiver Messages">
            <t>Replay of receiver messages are problematic for the NORM
            protocol, because replay of NACK messages could cause the sender
            unnecessarily transmit repair information for an FEC coding block.
            Similarly, the sender transmission rate might be manipulated by
            replay of congestion control feedback messages from receivers in
            the group. And the way that NORM senders estimate group round-trip
            timing (GRTT) could allow a replay attack to manipulate the
            senders' GRTT estimate to an unnecessarily large value, adding
            latency to the reliable transport process.</t>
          </section>
        </section>
      </section>
    </section>

    <section anchor="generalGoals" title="General Security Goals">
      <t>The term "security" is extremely vast and encompasses many different
      meanings. The goal of this section is to clarify what "security" means
      when considering the reliable multicast transport (RMT) protocols being
      defined in the IETF RMT working group. The scope can also encompass
      additional group communication applications, for instance streaming
      applications. This section only focuses on the desired general goals.
      The following sections will then discuss the possible elementary
      services that will be required to fulfill these general goals, as well
      as the underlying technological building blocks.</t>

      <t>The possible final goals include, in decreasing order of importance:
      <list style="symbols">
          <t>network protection: the goal is to protect the network from
          attacks, no matter whether these attacks are voluntary (i.e.
          launched by one or several attackers) or non voluntary (i.e. caused
          by a misbehaving system, where "system" can designate a building
          block, a protocol, an application, or a user);</t>

          <t>protocol protection: the goal is to protect the RMT protocol
          itself, e.g. to avoid that a misbehaving receiver prevents other
          receivers to get the content, no matter whether this is done
          intentionally or not;</t>

          <t>and content protection: to goal is to protect the content itself,
          for instance to guaranty the integrity of the content, or to make
          sure that only authorized clients can access the content.</t>
        </list></t>

      <section title="Network Protection">
        <t>Protecting the network is of course of primary importance. An
        attacker should not be able to damage the whole infrastructure by
        exploiting some features of the RMT protocol. Unfortunately, recent
        past has shown that the multicast routing infrastructure is relatively
        fragile, as well as the applications built on top of it. Since the RMT
        protocols may use congestion control mechanisms to regulate sender
        transmission rate, the protocol security features should ensure that
        the sender may not be manipulated to transmit at incorrect rates (most
        importantly not at an excessive rate) to any parts of the the receiver
        group. In the case of NORM, the security mechanisms should ensure that
        the feedback suppression mechanisms are protected to prevent
        badly-behaving network nodes from purposefully causing feedback
        implosion. In the case of ALC, where layered congestion control may be
        used via dynamic grou/layer membership, this extends to considerations
        of excessive manipulation of the multicast router control plane.</t>
      </section>

      <section title="Protocol Protection">
        <t>Protecting the protocols is also importance, since the higher the
        number of clients, the more serious the consequences of an attack.
        This is all the more true as scalability is often one of the desired
        goals of RMT protocols. Ideally, receivers should be sufficiently
        isolated from one another, so that a single misbehaving receiver does
        not affect others. Similarly, an external attacker should not be able
        to break the system, i.e. resulting in unreliable operation or
        delivery of incorrect content.</t>
      </section>

      <section title="Content Protection">
        <t>Finally, the content itself should be protected when meaningful.
        This level of security is often the concern of the content provider
        (and its responsibility). For instance, in case of confidential (or
        non-free) content, the typical solution consists in encrypting the
        content. It can be done within the upper application, i.e. above the
        RMT protocol, or within the transport system.</t>

        <t>But other requirements may exist, like verifying the integrity of a
        received object, or authenticating the sender of the received packets.
        To that goal, one can rely on the use of building blocks integrated
        within, or above, or beneath the RMT protocol.</t>

        <t>One may also consider that offering the packet sender
        authentication and content integrity services are basic requirements
        that should fulfill any RMT system that operates within an open
        network, where any attacker can easily inject spurious traffic in an
        ongoing NORM or ALC session. In that case this goal is not the
        responsibility of the content provider but the responsibility of the
        administrator who deploys the RMT system itself.</t>
      </section>
    </section>

    <section title="Elementary Security Techniques">
      <t>The goals defined in <xref target="generalGoals"></xref> will be
      fulfilled by means of underlying security techniques, provided by one or
      several technological building blocks. This section only focuses on
      these elementary security techniques. Some general techniques
      traditionally available are: </t>

      <texttable title="General Security Techniques">
        <ttcol width="20%">Technique</ttcol>

        <ttcol>Goal</ttcol>

        <c>packet integrity</c>

        <c>Enable session participants to verify that a packet has not been
        inappropriately modified in transit.</c>

        <c>packet source authentication</c>

        <c>Enable a receiver to verify the source of a packet.</c>

        <c>packet group authentication</c>

        <c>Enable a receiver to verify that a packet originated or was
        modified only within the group and has not been modified by nonmembers
        in transit; Additionally, if attribution of any modifications by the
        group is required, certain group authentication mechanisms may provide
        this capability.</c>

        <c>packet non-repudiation</c>

        <c>Enable any third party to verify the source of a packet such that
        the source cannot repudiate having sent the packet.</c>

        <c>packet anti-replay</c>

        <c>Enable a receiver to detect that a packet is the same as a
        previously-received packet</c>

        <c>object integrity</c>

        <c>Enable a receiver to verify the integrity of a whole object. Such
        bject integrity verification should be possible for any singular
        object or any composition of sub-objects which together constitute a
        larger object structure.</c>

        <c>object source authentication</c>

        <c>Enable a receiver to verify the source of an object.</c>

        <c>object confidentiality</c>

        <c>Enable a source to guarantee that only authorized receivers can
        access the object data.</c>
      </texttable>

      <t>Some additional techniques are specific to the RMT protocols:</t>

      <texttable title="RMT-Specific Security Techniques">
        <ttcol width="20%">Technique</ttcol>

        <ttcol>Goal</ttcol>

        <c>congestion control security</c>

        <c>Prevent an attacker from modifying the congestion control protocol
        normal behavior (e.g. by reducing the transmission (NORM) or reception
        (ALC) rate, or on the opposite increasing this rate up to a point
        where congestion occurs)</c>

        <c>group management</c>

        <c>Ensure that only authorized receivers (as defined by a certain
        group management policy) join the RMT session and possibly inform the
        source</c>

        <c>backward group secrecy</c>

        <c>Prevent a new group member to access the information in clear sent
        to the group before he joined the group</c>

        <c>forward group secrecy</c>

        <c>Prevent a former group member to access the information in clear
        sent to the group after he left the group</c>
      </texttable>

      <t>These technques are usually achieved by means of one or several
      technological building blocks. The target use case where the RMT system
      will be deployed will greatly impact the choice of the technological
      building block(s) used to provide these services, as explained in <xref
      target="useCaseCharacteristics"></xref>.</t>
    </section>

    <section title="Technological Building Blocks">
      <t>Here is a list of techniques and building blocks that are likely to
      fulfill one or several of the goals listed above: <list style="symbols">
          <t>IPsec;</t>

          <t>Use of TESLA within RMT;</t>

          <t>Use of Group MAC within RMT;</t>

          <t>Use of Digital signatures within RMT;</t>

          <t>use of SSM (Source Specific Multicast) multicast routing;</t>

          <t>Digital Signature;</t>

          <t>(TBD) add other BBs</t>
        </list> Each of them is now quickly discussed. In particular we
      identify what service it can offer, its limitations, and its field of
      application (adequacy W.R.T. the CDP and the target use case).</t>

      <section title="IPsec">
        <section title="Benefits">
          <t>One direct approach using existing standards is to apply IPSec
          <xref target="RFC2401"></xref> to achieve the following properties
          for message transmission: <list style="numbers">
              <t>Authentication (IPSec AH or ESP)</t>

              <t>Confidentiality (IPSec ESP)</t>
            </list></t>
        </section>

        <section title="Requirements">
          <t>It is expected that the approach to apply IPSec for reliable
          multicast transport sessions is similar to that described for OSPFv3
          security<xref target="RFC4552"></xref>. The following list proposes
          the IPSec capabilities needed to support a similar approach to RMT
          protocol security: <list style="numbers">
              <t>Mode - Transport mode IPSec security is required;</t>

              <t>Selectors - source and destination addresses and ports,
              protocol.</t>

              <t>For some uses, preplaced manual key support may be required
              to support application deployment and operation. For automated
              key management for group communication the Group Secure
              Association Key Management Protocol (GSAKMP) described in <xref
              target="RFC4535"></xref> may be used to emplace the keys for
              IPSec operation.</t>
            </list> Note that a periodic rekeying procedure similar to that
          described in RFC 4552 can also be applied with the additional
          benefit that the reliable transport aspects of the RMT protocols
          provide robustness to any message loss that might occur due to ANY
          timing discrepencies among the participants in the reliable
          multicast session.</t>
        </section>

        <section title="Limitations">
          <t>It should be noted that current IPSec implementations may not
          provide the capability for anti-replay protection for multicast
          operation. In the case of the NORM protocol, a sequence number is
          provided for packet loss measurement to support congestion control
          operation. This sequence number can also be used within a NORM
          implementation for detecting duplicate (replayed) messages from
          sources (senders or receivers) within the transport session group.
          In this way, protection against replay attack can be achieved in
          conjunction with the authentication and possibly confidentiality
          properties provided by an IPSec encapsulation of NORM messages. NORM
          receivers generate a very low volume of feedback traffic and it is
          expected that the 16-bit sequence space provided by NORM will be
          sufficient for replay attack protection. When a NORM session is
          long-lived, the limits of the sender repair window are expected to
          provide protection from replayed NACKs as they would typically be
          outside of the sender's current repair window. It is suggested that
          IPSec implementations that can provide anti-replay protection for IP
          Multicast traffic, even when there are multiple senders within a
          group, be adopted. The GSAKMP document has some discussion in this
          area.</t>
        </section>
      </section>

      <section title="Use of TESLA within RMT">
        <section title="Benefits">
          <t>The use of <xref target="TESLA_4_ALC_NORM">TESLA</xref> within
          the RMT protocols offers a loss tolerant, lightweight,
          authentication/integrity service for the packets generated by the
          session's sender. Depending on the time synchronization method and
          bootstrap method used, TESLA is compatible with massively scalable
          sessions. Because TESLA eavily relies on fast symmetric
          cryptographic building blocks, CPU processing remains limited both
          at the sender and receiver sides, which makes it suitable for high
          data rate transmissions, and/or lightweight terminals. Finally, the
          transmission overhead remains limited.</t>
        </section>

        <section title="Requirements">
          <t>The security offered by TESLA relies heavily on time. Therefore
          the session's sender and each receiver need to be loosely
          synchronized in a secure way. To that purpose, several methods
          exist, depending on the use case: direct time synchronization (which
          requires a bidirectional transport channel), using a secure NTP
          infrastructure (which also requires a bidirectional transport
          channel), or a GPS device, or a clock with a time-drift that is
          negligible in front of the TESLA time accuracy requirements.</t>

          <t>The various bootstrap parameters must also be communicated to the
          receivers, using either an in-band or out-of-band mechanism,
          sometimes requiring bidirectional communications.</t>

          <t>So, depending on the time synchronization scheme and the
          bootstrap mechanism method, TESLA can be used with either
          bidirectional or unidirectional transport channels.</t>
        </section>

        <section title="Limitations">
          <t>A first limitation is that TESLA does not protect the packets
          that are generated by receivers, for instance the feedback packets
          of NORM. These packets must be protected by other means.</t>

          <t>Another limitation is that TESLA requires some buffering
          capabilities at the receivers in order to enable the delayed
          authentication feature. This is not considered though as a major
          issue in the general case (e.g. FEC decoding of objects within an
          ALC session already requires some buffering capabilities, that often
          exceed that of TESLA), but it might be one in case of embedded
          environments.</t>
        </section>
      </section>

      <section title="Use of Group MAC within CDP">
        <section title="Benefits">
          <t>The use of Group MAC (Message Authentication Codes) within the
          CDP <xref target="SIMPLE_AUTH_4_ALC_NORM"></xref> is a simple
          solution to provide a loss tolerant group authentication/integrity
          service for all the packets exchanged within a session (i.e. the
          packets generated by the session's sender and the session's
          receivers). This scheme is easy to deploy since it only requires
          that all the group members share a common secret key. Because Group
          MAC heavily relies on fast symmetric cryptographic building blocks,
          CPU processing remains limited both at the sender and receiver
          sides, which makes it suitable for high data rate transmissions,
          and/or lightweight terminals. Finally, the transmission overhead
          remains limited.</t>
        </section>

        <section title="Requirements">
          <t>This scheme only requires that all the group members share a
          common secret key, possibly associated to a re-keying mechanism
          (e.g. each time the group membership changes, or on a periodic
          basis).</t>
        </section>

        <section title="Limitations">
          <t>This scheme cannot protect against attacks coming from inside the
          group, where a group member impersonates the sender and sends forged
          messages to other receivers. It only provides a group-level
          authentication/integrity service, unlike the TESLA and Digital
          Signature schemes.</t>

          <t>Note that the Group MAC and Digital Signature schemes can be
          advantageously used together, as explained in <xref
          target="SIMPLE_AUTH_4_ALC_NORM"></xref>.</t>
        </section>
      </section>

      <section title="Use of Digital Signatures within CDP">
        <section title="Benefits">
          <t>The use of Digital Signatures within the CDP <xref
          target="SIMPLE_AUTH_4_ALC_NORM"></xref> is a simple solution to
          provide a loss tolerant authentication/integrity service for all the
          packets exchanged within a session (i.e. the packets generated by
          the session's sender and the session's receivers). This scheme is
          easy to deploy since it only requires that the participants know the
          packet sender's public key, which can be achieved with either Public
          Key Infrastructre (PKI) or by pre-deploying these keys.</t>
        </section>

        <section title="Requirements">
          <t>This scheme is easy to deploy since it only requires that the
          participants know the packet sender's public key, which can be
          achieved either thanks to a PKI or by pre-deploying these keys.</t>
        </section>

        <section title="Limitations">
          <t>When RSA asymmetric cryptography is used, digital signatures has
          two major shortcommings: <list style="symbols">
              <t>it is limited by high computational costs, especially at the
              sender, and</t>

              <t>it is limited by high transmission overheads.</t>
            </list>This scheme is well suited to low data rate flows, when
          transmission overheads are not a major issue. For instance it can be
          used as a complement to TESLA for the feedback traffic coming from
          the session's receivers.</t>

          <t>The use of ECC ("Eliptic Curve Cryptography") significantly
          relaxes these constraints, especially when seeking for higher
          security levels. For instance, the following key size provide
          equivalent security:</t>

          <texttable>
            <ttcol align="center">Symmetric Key Size</ttcol>

            <ttcol align="center">RSA Key Size</ttcol>

            <ttcol align="center">ECC Key Size</ttcol>

            <c>80 bits</c>

            <c>1024 bits</c>

            <c>160 bits</c>

            <c>112 bits</c>

            <c>2048 bits</c>

            <c>224 bits</c>
          </texttable>

          <t>However in some cases, the Intellectual Property Rights (IPR)
          considerations for ECC may limit its use, so the other techniques
          are presented here as well.</t>

          <t>Note that the Group MAC and Digital Signature schemes can be
          advantageously used together, as explained in <xref
          target="SIMPLE_AUTH_4_ALC_NORM"></xref>.</t>
        </section>
      </section>

      <section title="SSM Multicast Routing">
        <t> <xref target="RFC3569">Source-specific Multicast (SSM)</xref>
        amends the classical Any-source Multicast (ASM) model creating logical
        IP multicast "channels" that are defined by the multicast destination
        address <spanx style="emph">and</spanx> the specific source address.
        Thus for a given "channel", only one specific source can inject
        packets that are distributed to receivers that have joined. This form
        of multicast has group management benefits since a source can
        independently control the "channels" it creates. Additionally, there
        are some security benefits of this multicast paradigm.</t>

        <section title="Benefits">
          <t>Since data-plane traffic for an SSM "channel" is limited to that
          of a single, specific source address, it is possible that network
          intermediate systems may impose mechanims that prevent injection of
          traffic to the group from inappropriate (perhaps malicious) nodes.
          This can reduce the risk for denial-of-service and some of the other
          attacks described in this document. While SSM alone is not a
          complete security solution, it can simplify secure RMT
          operation.</t>
        </section>

        <section title="Requirements">
          <t>Use of SSM requires that the network intermediate systems
          explicitly support it. Additionally, hosts are required to support
          the IGMPv3 extensions for SSM and applications and RMT
          implementations will need to support use of IGMPv3 including
          management of the <sourceAddr:dstMcastAdd> "channel"
          identifier.</t>
        </section>

        <section title="Limitations">
          <t>RMT protocols such as NORM that use signaling from receivers to
          multicast senders will need to use unicast addressing for feedback
          messages. In the case of NORM, its timer-based feedback suppression
          requires support of the sender NORM_CMD(REPAIR_ADV) message to
          control receiver feedback. In some topologies, use of unicast
          feedback may require some additional latency (increased backoff
          factor) for safe operation. The security of the unicast feedback
          from the receivers to sender will need to be addressed separately
          since the IP multicast model, including SSM, does not provide the
          sender knowledge of authorized group members.</t>
        </section>
      </section>

      <section title="Summary">
        <t>The following table summarizes the pros/cons of each
        authentication/integrity scheme used at application/transport
        level:</t>

        <texttable>
          <ttcol align="left"></ttcol>

          <ttcol align="center">RSA Digital Signature</ttcol>

          <ttcol align="center">ECC Digital Signature</ttcol>

          <ttcol align="center">Group MAC</ttcol>

          <ttcol align="center">TESLA</ttcol>

          <c>True auth and integrity</c>

          <c>Yes</c>

          <c>Yes</c>

          <c>No (group security)</c>

          <c>Yes</c>

          <c>Immediate auth</c>

          <c>Yes</c>

          <c>Yes</c>

          <c>Yes</c>

          <c>No</c>

          <c>Processing load</c>

          <c>--</c>

          <c>+</c>

          <c>++</c>

          <c>+</c>

          <c>Transmission overhead</c>

          <c>--</c>

          <c>+</c>

          <c>++</c>

          <c>+</c>

          <c>Complexity</c>

          <c>++</c>

          <c>++</c>

          <c>++</c>

          <c>--</c>
        </texttable>
      </section>
    </section>

    <section title="Security Infrastructure">
      <t>Deploying the elementary technological building blocks often requires
      that a security infrastructure exists. Such security infrastructure can
      provide: <list style="symbols">
          <t>Public Key Infrastructure (PKI) for trusted third party vetting
          of, and vouching for, user identities. PKI also allows the binding
          of public keys to users, usually by means of certificates.</t>

          <t>Group Key Management with rekeying schemes that are either
          periodic or triggered by some higher level event. It is required in
          particular when the group is dynamic and forward/backward secrecy
          are important. This is also required to improve the scalability of
          the CDP (since key management is done automatically, using a key
          server topology), or the security provided by the CDP (since the
          underlying cryptographic keys will be changed frequently)</t>
        </list></t>

      <t>It is expected that some RMT protocol deployments may use existing
      client-server security infrastructure models so that receivers may
      acquire any necessary session keys, etc and be authenticated or
      validated as needed for group participation. Then, the reliable delivery
      of session data content will be provided via the applicable RMT
      protocols. Note that in this case the security infrastructure itself may
      limit the scalability of the group size or other aspects of reliable
      multicast transfer. </t>

      <t>The IETF MSEC Working Group has developed some protocols that can be
      applied to achieve more scalable and effective group communication
      security infrastructure<xref target="RFC4046"></xref>. It is encouraged
      that these mechanisms be considered in the development of security for
      RMT protocols.</t>
    </section>

    <section title="New Threats Introduced by the Security Scheme Itself">
      <t>Introducing a security scheme, as a side effect, can sometimes
      introduce new security threats. For instance, signing all packets with
      asymmetric cryptographic schemes (to provide a source
      authentication/content integrity/anti-replay service) opens the door to
      DoS attacks. Indeed, verifying asymmetric-based cryptographic signatures
      is a CPU intensive task. Therefore an attacker can easily overload a
      receiver (or a sender in case of NORM) by injecting a significant number
      of faked packets.</t>
    </section>

    <section title="Consequences for the RMT and MSEC Working Group">
      <t>To meet the goals outlined in this document, it is expected that the
      RMT and Multicast Security (MSEC) WG may need to develop some supporting
      protocol security mechanisms.</t>

      <section title="RMT Transport Message Security Encapsulation Header">
        <t>An alternative approach to using IPSec to provide the necessary
        properties to protect RMT protocol operation from the application
        attacks described earlier, is to extend the RMT protocol message set
        to include a message encapsulation option. This encapsulation header
        could be used to provide authentication, confidientiality, and
        anti-replay protection as needed. Since this would be independent of
        the IP layer, the header might need to provide a source identifier to
        be used as a "selector" for recalling security state (including
        authentication certificate(s), sequence state, etc) for a given
        message. In the case of the NORM protocol, a NormNodeId field exists
        that could be used for this purpose. In the case of ALC, the security
        encapsulation mechanism would need to add this function. The security
        encapsulation mechanism, although resident "above" the IP layer, could
        use <xref target="RFC4535">GSAKMP</xref> or a similar approach for
        automated key managment.</t>
      </section>

      <!--
FROM THE JULY 2006 RMT MINUTES:

2- RMT BBs to Standard Track and Security

(point raised by Magnus Westerlund on the ML)

Magnus explains that having specifications (BBs, PIs) on the STD track
without any detailed guidelines on how to implement security services is
not reasonable. Hence this discussion...
Lorenzo answers there are many possible goals which makes this discussion
(and recommendations) difficult. The highest security goal is probably
to protect the network, a lower goal is to protect the protocol itself,
and an even lower goal is to protect the objects being transported.
Magnus agrees that there is a hierarchy of possible goals, but it does
not mean that we can ignore them altogether.
Lorenzo reminds that one initial RMT's goal was scalability, and therefore
it was decided that the source should not know the identity of the
receivers. As a consequence, it makes some security requirements more
complex to achieve.
George Gross: Scalability was one of the goals also for MSEC too
(e.g. for key management). Many BBs developed in MSEC will come for free.
Lorenzo: Do we need any modification to the protocols defined in MSEC
to use them in RMT?
George reminds the group that IPsec has been redesigned recently in order
to support multicast.
Lorenzo: We are not standardizing the application, but the transport.
As RMT is providing transport solutions, is there any need to bother
with application topics?
George: IPsec is one solution, but the question is whether we want to
use it (there are limitations) or want provide security in the
application layer?
Magnus: If we want to do that on a lower layer (i.e. with IPsec) we
need to  specify a profile for it here.

The discussion now focuses on the above three goals (content, network
and protocol protection):

1) Content protection:
Magnus: The provider of the content has its own opinion on how to protect
the content.
Lorenzo: It's not the problem of RMT to provide content protection.
We can say we need a WG in the application level to address this problem.
Magnus agrees with Lorenzo.
Mark: Yet but we must keep in mind that there's an urgent need to make
progress. Many people are using FLUTE/ALC now...
Magnus: Confidentiality issue is the last problem to solve. There are
more important problems to solve first.

2) Network protection:
Lorenzo: Network protection in our case means protecting the network
from congestion. What is the situation ?
Brian: There is a potential attack to TFMCC (used by NORM). The sender
can slow down because of a malicious receivers who says to be a bad
node.
George explains that against this kind of receiver feedback attack, the
only solution is to provide group key management and to use digital
signatures. This technique can enable the sender to decide whether
a receiver who sends feedback is an authorized member of the group or
not. But a requirement is to use a key management infrastructure...

3) Protocol protection:
One possible goal is to protect the transport protocol against nodes
that want to stop the protocol.
Lorenzo: Can we mandate the use of SSM to protect NORM?
Magnus answers that an attacker having control over the network can
easily spoof the source, so using SSM does not really solve the problem.
George: Using public keys, a certificate authority and digital signature
can help... The new version of IPsec, that features multicast extensions,
is another  way to achieve that goal.
Lorenzo suggests to ask the MSEC WG to have it a WG item. He also
suggests to add in the security requirement of NORM a note explaining
that this scheme can be used.

Gorry: Why do we go to SSM?
Lorenzo explains there's a potential solution for TFMCC if we mandate the
use of SSM. It seems a realistic assumption.
Brian explains that with SSM, a receiver can still slow down the sender
but cannot break the network. 
Lorenzo: A possible solution, for TFMCC, is to say that if the sender
receives contradicting feedback, then he should trust the worst one.

Gorry explains that in case of the ALC layer congestion control
scheme, there's an incentive to receive at a higher rate even if it
breaks the network (i.e. by not being fair with other TCP flows).
Lorenzo: Is there any work in unicast to address the problem?
What about a TCP receiver that acknowledges each packet even if
he didn't receive them? It can happen in NORM and ALC but also in TCP.
It seems to be a very tough problem since it needs to involve the routers.
Somebody explains that the problem is that we trust the host to do
the congestion control. In TCP, if a receiver misbehaves, he will
probably be the only one to be hurt.
Lorenzo: So we have a very big problem here... But we are only
opening issues, we are not solving them!

-->
    </section>

    <section title="Security Considerations">
      <t>This document is a general discussion of security for the RMT
      protocol family. But specific security considerations are not applicable
      as this document does not introduce any new techniques.</t>
    </section>

    <section title="Acknowledgments">
      <t>The authors would like acknowledge Magnus Westerlund for stimulating
      the working group activity in this area. Additionally George Gross and
      Ran Atkinson contributed many ideas to the discussion here.</t>
    </section>
  </middle>

  <back>
    <references title="Normative References">
      <?rfc include='reference.RFC.2119'?>

      <?rfc include='reference.RFC.2401'?>

      <?rfc include='reference.RFC.3450'?>

      <!--
	<?rfc include='reference.RFC.3451'?>
-->

      <?rfc include='reference.RFC.3452'?>

      <?rfc include='reference.RFC.3926'?>

      <?rfc include='reference.RFC.3940'?>

      <?rfc include='reference.RFC.4535'?>

      <?rfc include='reference.RFC.4552'?>

      <?rfc include='reference.RFC.4654'?>

      <?rfc include='reference.RFC.4046'?>

      <reference anchor="draft-ietf-rmt-pi-alc-revised">
        <front>
          <title>Asynchronous Layered Coding (ALC) Protocol
          Instantiation</title>

          <author initials="M." surname="Luby">
            <organization></organization>
          </author>

          <author initials="M." surname="Watson">
            <organization></organization>
          </author>

          <author initials="L." surname="Vicisano">
            <organization></organization>
          </author>

          <date month="February" year="2007" />
        </front>

        <seriesInfo name=""
                    value="draft-ietf-rmt-pi-alc-revised-04.txt (work in progress)" />
      </reference>

      <reference anchor="TESLA_4_ALC_NORM">
        <front>
          <title>The Use of TESLA in the ALC and NORM Protocols</title>

          <author fullname="Vincent Roca" initials="V." surname="Roca">
            <organization></organization>
          </author>

          <author fullname="Aurelien Francillon" initials="A."
                  surname="Francillon">
            <organization></organization>
          </author>

          <author fullname="Sebastien Faurite" initials="S." surname="Faurite">
            <organization></organization>
          </author>

          <date month="July" year="2007" />
        </front>

        <seriesInfo name="Internet-Draft"
                    value="draft-ietf-msec-tesla-for-alc-norm-02.txt" />

        <format target="http://www.ietf.org/internet-drafts/draft-ietf-msec-tesla-for-alc-norm-02.txt"
                type="TXT" />
      </reference>

      <reference anchor="SIMPLE_AUTH_4_ALC_NORM">
        <front>
          <title>Simple Authentication Schemes for the ALC and NORM
          Protocols</title>

          <author fullname="Vincent Roca" initials="V." surname="Roca">
            <organization></organization>
          </author>

          <date month="June" year="2007" />
        </front>

        <seriesInfo name="Internet-Draft"
                    value="draft-roca-rmt-simple_auth-for-alc-norm-00.txt" />

        <format target="http://www.ietf.org/internet-drafts/draft-roca-rmt-simple_auth-for-alc-norm-00.txt"
                type="TXT" />
      </reference>

      <reference anchor="RFC3569">
        <front>
          <title>An Overview of Source-Specific Multicast (SSM)</title>

          <author fullname="S. Bhattacharyya" initials="S."
                  surname="Bhattacharyya">
            <organization></organization>
          </author>

          <date month="July" year="2003" />
        </front>

        <seriesInfo name="IETF" value="RFC 3569" />
      </reference>
    </references>

    <references title="Informative References">
      <reference anchor="Neumann05">
        <front>
          <title>Large Scale Content Distribution Protocols</title>

          <author fullname="Christoph Neumann" initials="C." surname="Neumann">
            <organization>CN2005</organization>

            <address>
              <postal>
                <street></street>

                <city></city>

                <region></region>

                <code></code>

                <country></country>
              </postal>

              <phone></phone>

              <facsimile></facsimile>

              <email></email>

              <uri></uri>
            </address>
          </author>

          <author fullname="Vincent Roca" initials="V." surname="Roca">
            <organization>V</organization>
          </author>

          <author fullname="Rod Walsh" initials="R." surname="Walsh">
            <organization></organization>

            <address>
              <postal>
                <street></street>

                <city></city>

                <region></region>

                <code></code>

                <country></country>
              </postal>

              <phone></phone>

              <facsimile></facsimile>

              <email></email>

              <uri></uri>
            </address>
          </author>

          <date month="October" year="2005" />

          <abstract>
            <t></t>
          </abstract>
        </front>

        <seriesInfo name="ACM Computer Communications Review (CCR)"
                    value="Vol. 35 No. 5" />
      </reference>
    </references>
  </back>
</rfc>

PAFTECH AB 2003-20262026-04-23 10:57:55