Internet Engineering Task Force                             A. Ford, Ed.
Internet-Draft                                       Roke Manor Research
Intended status: Experimental                                  C. Raiciu
Expires: November 8, 2009                                     M. Handley
                                               University College London
                                                                S. Barre
                                                Universite catholique de
                                                             May 7, 2009

     TCP Extensions for Multipath Operation with Multiple Addresses

Status of this Memo

   This Internet-Draft is submitted to IETF in full conformance with the
   provisions of BCP 78 and BCP 79.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups.  Note that
   other groups may also distribute working documents as Internet-

   Internet-Drafts are draft documents valid for a maximum of six months
   and may be updated, replaced, or obsoleted by other documents at any
   time.  It is inappropriate to use Internet-Drafts as reference
   material or to cite them other than as "work in progress."

   The list of current Internet-Drafts can be accessed at

   The list of Internet-Draft Shadow Directories can be accessed at

   This Internet-Draft will expire on November 8, 2009.

Copyright Notice

   Copyright (c) 2009 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents in effect on the date of
   publication of this document (
   Please review these documents carefully, as they describe your rights
   and restrictions with respect to this document.

Ford, et al.            Expires November 8, 2009                [Page 1]

Internet-Draft                Multipath TCP                     May 2009


   Often endpoints are connected by multiple paths, but the nature of
   TCP/IP restricts communications to a single path per socket.
   Resource usage within the network would be more efficient were these
   multiple paths able to be used concurrently.  This should enhance
   user experience through higher throughput and improved resilience to
   network failure.  This document presents extensions to TCP in order
   to transparently provide this multi-path functionality at the
   transport layer, if at least one endpoint is multi-addressed.

Ford, et al.            Expires November 8, 2009                [Page 2]

Internet-Draft                Multipath TCP                     May 2009

Table of Contents

   1.  Introduction . . . . . . . . . . . . . . . . . . . . . . . . .  4
     1.1.  Motivations  . . . . . . . . . . . . . . . . . . . . . . .  4
     1.2.  Design Assumptions . . . . . . . . . . . . . . . . . . . .  4
     1.3.  Layered Representation . . . . . . . . . . . . . . . . . .  5
     1.4.  Operation Summary  . . . . . . . . . . . . . . . . . . . .  6
     1.5.  Open Issues  . . . . . . . . . . . . . . . . . . . . . . .  7
     1.6.  Requirements Language  . . . . . . . . . . . . . . . . . .  8
   2.  Terminology  . . . . . . . . . . . . . . . . . . . . . . . . .  8
   3.  Semantic Issues  . . . . . . . . . . . . . . . . . . . . . . .  9
   4.  MPTCP Protocol . . . . . . . . . . . . . . . . . . . . . . . .  9
     4.1.  Session Initiation . . . . . . . . . . . . . . . . . . . . 10
     4.2.  Address Knowledge Exchange (Path Management) . . . . . . . 11
       4.2.1.  Explicit Path Management . . . . . . . . . . . . . . . 11  Adding Addresses . . . . . . . . . . . . . . . . . 11  Remove Address . . . . . . . . . . . . . . . . . . 12
       4.2.2.  Implicit Path Management . . . . . . . . . . . . . . . 13  Request-SYN  . . . . . . . . . . . . . . . . . . . 14  Request-FIN (Remove Address) . . . . . . . . . . . 15
     4.3.  Starting a New Subflow . . . . . . . . . . . . . . . . . . 15
     4.4.  General MPTCP Operation  . . . . . . . . . . . . . . . . . 16
       4.4.1.  Subflow Policy . . . . . . . . . . . . . . . . . . . . 17
       4.4.2.  Retransmissions  . . . . . . . . . . . . . . . . . . . 19
       4.4.3.  Resync Packet  . . . . . . . . . . . . . . . . . . . . 19
     4.5.  Closing a Connection . . . . . . . . . . . . . . . . . . . 20
     4.6.  Error Handling . . . . . . . . . . . . . . . . . . . . . . 22
   5.  Security Considerations  . . . . . . . . . . . . . . . . . . . 22
   6.  Interactions with Middleboxes  . . . . . . . . . . . . . . . . 23
   7.  Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 23
   8.  Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 23
   9.  IANA Considerations  . . . . . . . . . . . . . . . . . . . . . 23
   10. References . . . . . . . . . . . . . . . . . . . . . . . . . . 24
     10.1. Normative References . . . . . . . . . . . . . . . . . . . 24
     10.2. Informative References . . . . . . . . . . . . . . . . . . 24
   Appendix A.  Functional Separation . . . . . . . . . . . . . . . . 25
     A.1.  Motivations  . . . . . . . . . . . . . . . . . . . . . . . 25
     A.2.  TCP Performance  . . . . . . . . . . . . . . . . . . . . . 26
     A.3.  Architecture overview  . . . . . . . . . . . . . . . . . . 26
     A.4.  PM/MPS interface . . . . . . . . . . . . . . . . . . . . . 28
   Appendix B.  Notes on use of TCP Options . . . . . . . . . . . . . 29
   Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 29

Ford, et al.            Expires November 8, 2009                [Page 3]

Internet-Draft                Multipath TCP                     May 2009

1.  Introduction

   This section describes the motivations behind the design of Multipath
   TCP (henceforth referred to as MPTCP), a set of extensions for
   regular TCP [RFC0793] to allow one TCP connection to be spread across
   multiple paths.  The following sections go on to describe the
   extensions themselves, and its operation.

1.1.  Motivations

   As the Internet evolves, demands on Internet resources are ever-
   increasing, but often these resources (in particular, bandwidth)
   cannot be fully utilised due to protocol constrains on both the end-
   systems and within the network.  By the application of resource
   pooling [WISCHIK], these resources can be 'pooled' such that they
   appear as a single logical resource to the user.  Multipath TCP
   achieves resource pooling by combining multiple TCP sessions running
   over multiple paths, and presenting them as a single TCP connection
   to the application.

   This form of resource pooling bring two key benefits:

   o  To increase the efficiency of the resource usage, and thus
      increase the network capacity available to end hosts.

   o  To increase the resilience of the connectivity by providing
      multiple paths, protecting end hosts from the failure of one.

   The protocol presented in this document still follows the same
   service model as TCP [RFC0793]: byte oriented, in order reliable
   delivery.  This leads to a high level goal of the resulting protocol,
   where 'subflows' on different paths will function independently of
   one another, i.e. failure of one path should not result in reduced
   throughput on the other paths.

1.2.  Design Assumptions

   In order to limit the potentially huge design space, the authors
   imposed two key constraints on the multipath TCP design presented in
   this document:

   o  It must be backwards-compatible with current, regular TCP, to
      increase its chances of deployment

   o  It can be assumed that one or both endpoints are multihomed and

   To simplify the design we assume that the presence of multiple

Ford, et al.            Expires November 8, 2009                [Page 4]

Internet-Draft                Multipath TCP                     May 2009

   addresses at an endpoint is sufficient to indicate the existence of
   multiple paths.  These paths need not be entirely disjoint: they may
   share one or many routers between them.  Even in such a situation
   making use of multiple paths will improve resource utilisation.

   There are three aspects to the backwards-compatibility listed above:

   External Constraints:  The protocol must function through the vast
      majority of existing middleboxes such as NATs, firewalls and
      proxies, and as such must resemble existing TCP as far as possible
      on the wire.  In addition, therefore, we cannot rely on the TCP
      packets (both headers and payloads) remaining unchanged end-to-

   Application Constraints:  The protocol must be usable with no change
      to existing applications that use the standard TCP API (although
      it is reasonable that not all features would be available to such
      legacy applications).

   Fall-back:  The protocol should be able to fall back to standard TCP
      with no interference from the user, to be able to communicate with
      legacy hosts.

   Areas for further study:

   o  In theory, since this is purely a TCP extension, it should be
      possible to use MPTCP with both IPv4 and IPv6 on dual-stack hosts,
      thus having the additional possible benefit of aiding transition.

   o  Some features of the design presented here could be extended to
      work with non-multi-addressed hosts by using packet marking or
      partial multipath.

   o  Some features of the design presented here could be combined with
      mechanisms such as shim6 [I-D.ietf-shim6-proto].

   It is important to note that this document deliberately avoids any
   discussion of algorithms for coupling congestion windows in order to
   achieve optimum performance.  Work in this area is ongoing and will
   be presented in separate documents; considerable discussion can be
   found in [I-D.van-beijnum-1e-mp-tcp-00]

1.3.  Layered Representation

   MPTCP operates at the transport layer, and its existence aims to be
   transparent to both higher and lower layers.  It is a set of
   additional features on top of standard TCP, and as such MPTCP is
   designed to be usable by legacy applications with no changes.  A

Ford, et al.            Expires November 8, 2009                [Page 5]

Internet-Draft                Multipath TCP                     May 2009

   possible implementation would be for such a feature to be a system-
   wide setting: "Use multipath TCP by default?  Y/N".  Multipath-aware
   applications would be able to use an extended sockets API to have
   further influence on the behaviour of MPTCP.  Figure 1 illustrates
   this architecture.

                                      |           Application         |
      +---------------+               +-------------------------------+
      |  Application  |               |             MPTCP             |
      +---------------+               + - - - - - - - + - - - - - - - +
      |      TCP      |               |      TCP      |      TCP      |
      +---------------+               +-------------------------------+
      |      IP       |               |       IP      |      IP       |
      +---------------+               +-------------------------------+

      Figure 1: Comparison of Standard TCP and MPTCP Protocol Stacks

   Detailed discussion of an architecture for developing a multipath TCP
   implementation, especially regarding the functional separation by
   which different components should be developed, is given in
   Appendix A.

1.4.  Operation Summary

   This section will, very briefly, provide a high-level summary of the
   normal operation case of MPTCP, and is illustrated by the scenario
   shown in Figure 2.  A detailed description of operation is given in
   Section 4.

   o  To a non-MPTCP-aware application, MPTCP will be indistinguishable
      from normal TCP.  All MPTCP operation is handled by the MPTCP
      implementation, although extended APIs could provide additional
      control.  An application begins by opening a TCP socket in the
      normal way.

   o  An MPTCP connection begins as a single TCP session.  This
      illustrated in Figure 2 as being between Addresses A1 and B1 on
      Hosts A and B respectively.

   o  If extra paths are available, additional TCP sessions are created
      on these paths, and are combined with the existing session, which
      continues to appear as a single connection to the applications at
      both ends.  The creation of the additional TCP session is
      illustrated between Address A2 on Host A and Address B1 on Host B.

   o  MPTCP identifies multiple paths by the presence of multiple
      addresses at endpoints.  Combinations of these multiple addresses

Ford, et al.            Expires November 8, 2009                [Page 6]

Internet-Draft                Multipath TCP                     May 2009

      equate to the additional paths.  In the example, other potential
      paths that could be set up are A1<->B2 and A2<->B2.  Although this
      additional session is shown as being initiated from A2, it could
      equally have been initiated from B1.

   o  The discovery and setup of additional TCP sessions (termed
      'subflows') can be achieved through alternative mechanisms, two of
      which are described in this document for comment.

   o  The exact properties of these TCP sessions that are logically
      bonded are dependent upon the congestion and flow control
      characteristics of the endpoints' MPTCP implementation.

   o  MPTCP adds connection-level sequence numbers in order to
      reassemble the data stream in-order from multiple subflows.
      Connections are terminated by connection-level FIN packets as well
      as those relating to the individual subflows.

               Host A                               Host B
      ------------------------             ------------------------
      Address A1    Address A2             Address B1    Address B2
      ----------    ----------             ----------    ----------
          |             |                      |             |
          |     (initial connection setup)     |             |
          |----------------------------------->|             |
          |<-----------------------------------|             |
          |             |                      |             |
          |            (additional subflow setup)            |
          |             |--------------------->|             |
          |             |<---------------------|             |
          |             |                      |             |
          |             |                      |             |

                  Figure 2: Example MPTCP Usage Scenario

1.5.  Open Issues

   This specification is a work-in-progress, and as such there are many
   issues that are still to be resolved.  This section lists many of the
   key open issues within this specification; these are discussed in
   more detail in the appropriate sections throughout this document.

   o  Congestion control, and especially mechanisms by which congestion
      windows should be coupled to best respond to congestion on a path.
      This is also related to retransmission algorithms, in particular
      how to decide when to retransmit packets on the same or different

Ford, et al.            Expires November 8, 2009                [Page 7]

Internet-Draft                Multipath TCP                     May 2009

   o  Correct path/address management scheme.  There are two schemes
      (implicit and explicit) presented in this document.  The authors
      generally tend towards the implicit scheme for simplicity, however
      both are presented to solicit feedback.  Other alternatives are
      also welcome!

   o  Best handshake mechanisms.  This document contains a proposed
      scheme by which connections and subflows can be set up.  It is
      felt that, although this is "no worse than regular TCP", there
      could be opportunities for significant improvements in security
      that could be included (potentially optionally) within this

   o  Issues around simulataneous opens, where both ends attempt to
      create a new subflow simultaneously, need to be investigated and
      behaviour specified.

   o  Appropriate mechanisms for controlling policy of subflow usage.
      The ECN signal is currently proposed but other alternatives,
      including path property options, could be employed instead.

1.6.  Requirements Language

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   document are to be interpreted as described in RFC 2119 [RFC2119].

2.  Terminology

   Path:  A sequence of links between a sender and a receiver, defined
      in this context by a source and destination address pair.

   Subflow:  A stream of TCP packets sent over a path.  A subflow is a
      component part of a connection between two endpoints.

   Connection:  A collection of one or more subflows, over which an
      application can communicate between two endpoints.  There is a
      one-to-one mapping between a connection and a socket.

   Token:  A unique identifier given to a multipath connection by an
      endpoint.  May also be referred to as a "Connection ID".

   Endpoint:  A host operating an MPTCP implementation, and either
      initiating or terminating a MPTCP connection.

Ford, et al.            Expires November 8, 2009                [Page 8]

Internet-Draft                Multipath TCP                     May 2009

3.  Semantic Issues

   In order to support multipath operation, the semantics of some TCP
   components have changed.  To aid clarity, this section collects these
   semantic changes as a reference.

   Sequence Number:  The TCP sequence number is subflow-specific, with a
      data sequence number used for reassembly for higher layers.

   FIN:  The FIN only applies to a subflow, not to a connection.  For a
      connection-level FIN, use the DATA FIN option.

   ACK:  The ACK acknowledges the subflow sequence number only, and the
      mapping to the data sequence number is handled out-of-band.

   RST:  The RST only applies to a subflow.  There is no connection-
      level RST, since it would be impossible to distinguish the two, as
      the link between a subflow and a connection is established at the
      SYN handshake.  A connection is considered reset if every subflow
      sends a RST in response.

   Address List:  The address management is handled per-connection to
      permit the application of per-connection local policy.

   IP Address:  The IP address presented to the application layer in a
      non-multipath-aware application is that of the first address
      connected to, even if that address has since been removed from the

4.  MPTCP Protocol

   This section describes the operation of the MPTCP protocol, and is
   subdivided into sections for each key part of the protocol operation.

   All MPTCP operations are signalled using optional TCP header fields.
   These TCP Options will have option numbers allocated by IANA, as
   discussed in Section 9, and are defined throughout the following

   This document currently presents two alternatives for management of
   addresses to set up additional subflows:

   Explicit Path Management:  Each endpoint shares a list of addresses
      on which it can be reached.  Either endpoint can then initiate new
      subflows between any pair of these addresses.

Ford, et al.            Expires November 8, 2009                [Page 9]

Internet-Draft                Multipath TCP                     May 2009

   Implicit Path Management:  A multihomed endpoint starts additional
      subflows, by connecting from an address not currently in use in
      the connection to a destination that is in use in an existing

   We present these alternatives in order to solicit feedback on the
   most appropriate mechanism to use for maximum compatibility and thus
   liklihood of take-up.

   Briefly, the key differences are that explicit path management
   provides additional flexibility in the ability of endpoints to use
   any combination of addresses (not just those already active), whereas
   implicit path management is relatively simpler (requiring fewer TCP
   options), and also has functionality to work around NATs.

4.1.  Session Initiation

   Session Initiation begins with a SYN, SYN/ACK exchange on a single
   path.  Each of these packets will additionally feature the Multipath
   Capable TCP option (Figure 3, which declares the sender's locally
   unique 32-bit token for this connection, and a version field.

   The "Multipath Capable" option declares an endpoint to be capable of
   operating Multipath TCP (or rather, more accurately, a desire to
   operate Multipath TCP on this particular connection).  As well as
   this declaration, this field presents a token, which is used when
   adding additional subflows to this connection.  This token is
   generated by the sender and has local meaning only, but it must be
   unique for the sender.  The token should be difficult for an attacker
   to guess, and thus it is recommended to be generated randomly.
   (However, see further discussions about security in Section 5.)

   This option is only present in packets with the SYN flag set.  It is
   only used in the first TCP session of a connection, in order to
   identify the connection; all following connections will use path
   management techniques to join the existing connection.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      | Kind=OPT_MPC  |  Length = 7   |(resvd)|Version|  Sender Token :
      : Sender Token (continued - 4 octets total)     |

                    Figure 3: Multipath Capable option

   The version field represents the version of MPTCP in use.  The

Ford, et al.            Expires November 8, 2009               [Page 10]

Internet-Draft                Multipath TCP                     May 2009

   version provided in this specification is 0.  The reserved bits may
   be used for connection-specific flags in later versions.

   If a SYN contains a "multipath capable" option but the SYN/ACK does
   not, it is assumed that the recipient is not multipath capable and
   thus the MPTCP session will operate as regular, single-path TCP.  If
   a SYN does not contain a "multipath capable" option, the SYN/ACK MUST
   NOT contain one in response.

   If these packets are unacknowledged, it is up to local policy to
   decide how to respond.  It is expected that a sender will eventually
   fall back to single-path TCP (i.e. without the Multipath Capable
   Option), in order to work around middleboxes that may drop packets
   with unknown options, however the number of multipath-capable
   attempts that are made first will be up to local policy.  In the case
   of out-of-order packets, i.e. if a multipath-capable SYN/ACK is
   received in response to a multipath-capable SYN, after a standard SYN
   has been sent, then once again it is up to the sender to choose how
   to behave.  For example, the sender could respond to new connections
   using the previously declared token, or it could simply drop any new
   multipath options within the flow.

   If an endpoint is known to be multiaddressed (e.g. through multiple
   addresses returned in a DNS lookup), alternative destination
   addresses should be tried first, before falling back to regular TCP.

4.2.  Address Knowledge Exchange (Path Management)

   This section presents two alternative path management techniques, as
   introduced at the start of Section 4.

4.2.1.  Explicit Path Management

   With explicit path management, the addresses over which a host is
   accessible are announced to the other party through in-band
   signalling, and then hosts can set up new TCP subflows on any subset
   of combinations of (source, destination) address pairs.  Either
   endpoint can initiate the creation of a new subflow.  Adding Addresses

   Announcing additional addresses that an endpoint can be reached on
   will be undertaken by the Add Address TCP Option (Figure 4), where an
   (index, address) pair can be announced to the other endpoint.
   Several addresses can be added if there is sufficient TCP option
   space, otherwise multiple TCP messages containing this option must be
   sent.  This option can be used at any time during a connection; not
   just at the initial SYN/ACK exchange.

Ford, et al.            Expires November 8, 2009               [Page 11]

Internet-Draft                Multipath TCP                     May 2009

   The Add Address option announces a list of alternative IP addresses,
   beyond the current one in use, that the sender can be contacted on.
   This option can be used multiple times until all available addresses
   have been announced, in order to get around TCP option space limits.
   It should be noted that every address has an index which can be used
   for address removal, and therefore endpoints must cache the mapping
   between index and address.  The index must be unique to the sender,
   and although it is expected to be sequential this is not mandated.

   This option is shown for IPv4.  For IPv6, the IPVer field will read
   6, and the length of the address will be 16 octets not 4, and thus
   the length of the option will be 2 + (18 * number_of_entries).
   Multiple addresses can be included, with an index following on
   immediately from the previous address, and their existance can be
   inferred through the option length and version fields.

   NB: by having a IPVer field, we get four free reserved bits.  These
   could be used in later versions of this protocol, e.g. one bit for
   "use now" or similar, to differentiate between subflows for backup
   purposes and those for throughput.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      | Kind=OPT_ADDR |     Length    |     Index     | IPVer |(resvd)|
      |                   Address (IPv4 - 4 octets)                   |
        ( ... further Index/Version/Address fields as required ... )

                  Figure 4: Add Address option (for IPv4)

   If an index is already in use, it should be treated as a request to
   remove the existing address (see Section followed by a new
   addition at that new index.  Remove Address

   If, during the lifetime of a MPTCP connection, a previously-announced
   address becomes invalid (e.g. if the interface disappears), the
   affected endpoint should announce this so that the other endpoint can
   remove subflows related to this address.

   This is achieved through the Remove Address option (Figure 5), which
   will remove a previously-added address (or list of addresses) from a
   connection and terminate any subflows currently using it.

   The sending and receipt of this message should trigger the sending of

Ford, et al.            Expires November 8, 2009               [Page 12]

Internet-Draft                Multipath TCP                     May 2009

   FINs by both endpoints on the affected subflow(s) (if possible), as a
   courtesy to cleaning up middlebox state, but endpoints may clean up
   their internal state without a long timeout.

   If there is no address at the requested indices, the receiver will
   silently ignore the request.

   Address removal is undertaken by index, so as to permit the use of
   (MPTCP-aware) NATs and other middleboxes, in the cases where new
   connections have been initiated but now want to be removed.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      |Kind=OPT_REMADR|  Length = 2+n |     Index     | ...

                      Figure 5: Remove Address option

4.2.2.  Implicit Path Management

   As opposed to the explicit path management presented above, this
   method does not exchange a list of addresses which are then
   (independently) used to set up subflows.  Instead, knowledge of an
   endpoint's alternative addresses is obtained only when an additional
   subflow is being set up (see Section 4.3).  Subflows are started,
   joining a pre-existing connection, with no pre-negotiation.  A
   "Request-SYN" option can also used to request a SYN in the reverse
   direction, in order to get around middleboxes, notably NATs.

   The implicit mechanism makes use of SYNs and connection identifiers
   in order to add new subflows to an existing connection.  The
   following is an example of how this should work:

   o  An endpoint that is multihomed starts an additional TCP session to
      an address/port pair that is already in use on the other endpoint,
      using a token to identify the flow (Section 4.3).  (A multihomed
      destination may open a new subflow from its new address to the
      source address and port, or a multihomed source may open a new
      subflow from its new address another connection to the existing
      destination and port).

   o  To expand upon this, say a connection is intiated from host "A" on
      (address, port) combination A1 to desintation (address, port) B1
      on host "B".  If host A is multihomed, it starts an additional
      connection from new (address, port) A2 to B1, using B's previously
      declared token.  Alternatively, if B is multhomed, it will try to
      set up a new TCP connection from B2 to A1, using A's previously

Ford, et al.            Expires November 8, 2009               [Page 13]

Internet-Draft                Multipath TCP                     May 2009

      declared token.

   o  Simultaneously, a "Request-SYN" option is sent on an existing TCP
      connection, asking the recipient to try to open a connection to
      the sender's additional address.  This is intended to permit new
      sessions to be opened if one endpoint is behind a NAT.

   o  Using the previous notation, this would be a Request-SYN packet
      sent from A1 to B1 requesting a SYN to be sent from B1 to A2.

   As can be seen, the implicit path management is designed for ease of
   deployment and operation through middleboxes such as NATs.  The main
   drawback is that new subflows can only be started with one of the two
   addresses being part of an existing subflow, since there is no
   separate exchange of addresses.  This improves security and
   simplicity but limits the flexibility and speed of being able to set
   up entirely disjoint subflows immediately on an address list
   exchange.  However, once multiple addresses exist at one endpoint,
   the other endpoint can target new connections at any or all of these.  Request-SYN

   This packet requests the recipient to send a SYN (with a join option,
   discussed in Section 4.3) to the presented IP address to initiate a
   new subflow.  The motivation for this is to get around NATs and
   firewalls that may block SYN packets in the forward direction.  This
   packet could be seen as fulfilling the same function as "Add Address"
   for explicit path management.

   This option is shown for IPv4.  For IPv6, the IPVer field will read
   6, and the length of the address will be 16 octets not 4, and thus
   the length of the option will be 19.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      |Kind=OPT_REQSYN|    Length     | IPVer |(resvd)|    Address... :
      : ... Address (4 octets - IPv4 version only)    |

                Figure 6: Request-SYN option (IPv4 version)

   OPEN ISSUES: Must the recipient reply from the same address?  Can a
   nonce be used for security here by echoing it in the "join" option in
   the SYN?  Do we need anything to prevent DoS here?  We also will need
   to define the logic of responding to this versus having already sent
   a SYN (related to the simultaneous open issue).

Ford, et al.            Expires November 8, 2009               [Page 14]

Internet-Draft                Multipath TCP                     May 2009  Request-FIN (Remove Address)

   OPEN ISSUE: Do we want to be able to do Request-FIN?  It would be
   used to do cleanups of other subflows, e.g. when an interface becomes
   unavailable (i.e. like "Remove Address" for explicit path
   management).  Assuming we need this option, somehow we need to be
   able to identify the existing subflows.  This is particularly
   difficult when there is no subflow identifier.

   The primary reason for this message is to allow a sender to tell its
   receiver that a particular inferface has been unexpectedly lost, and
   thus it should close any connections associated with it.  Although
   this is purely an efficiency and not essential to the operation of
   the protocol, it would nevertheless be useful to deploy such a
   mechanism.  As currently proposed, this option will not work through
   non-MPTCP-aware NATs, and so it should not be expected to be

   This option works by a sender identifying the source address that is
   no longer valid.  A Request-FIN requests the recipient to send a FIN
   on the affected subflow(s), and then it can close the subflows with a
   short timeout.  The sender should also send FINs, however the
   Request-FIN is used to help clean up state on middleboxes on subflows
   that have unexpectedly broken.

   This option is shown for IPv4.  For IPv6, the IPVer field will read
   6, and the length of the address will be 16 octets not 4, and thus
   the length of the option will be 19.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      |Kind=OPT_REQFIN|    Length     | IPVer |(resvd)|    Address... :
      : ... Address (4 octets - IPv4 version only)    |

                Figure 7: Request-FIN option (IPv4 version)

4.3.  Starting a New Subflow

   Endpoints have knowledge of their own multiple addresses, and can
   become aware of the other endpoint's addresses through a path
   management technique as described in Section 4.2.  Once this
   knowledge has been gathered, an endpoint will want to initiate a new
   subflow over a currently unused pair of addresses.

   A new subflow is started as a normal TCP SYN/ACK exchange, to (or

Ford, et al.            Expires November 8, 2009               [Page 15]

Internet-Draft                Multipath TCP                     May 2009

   from) a different address to one already in use.  The following TCP
   option is used to identify which connection the new subflow should
   become part of.  The token used is the locally unique token of the
   destination for the connection, as defined by the Multipath Capable
   option received in the first SYN/ACK exchange.

   It should be noted that, in theory, additional subflows can exist
   between any pair of ports, and as such it is this token that is used
   for demuxing at the receiver.  A receiver must store some mapping
   state, of (source_addr, dest_addr, source_port, dest_port) to its
   token, using information from the initial SYN exchange, in order to
   enable this.  In practice, however, it is envisaged that most new
   subflows will connect to a port that is already in use as the source
   or destination port of an existing subflow, in order to have a
   greater chance of getting through firewalls and other middleboxes,
   and to support traffic engineering of the flows.

   This option can only be present when the SYN flag is set.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      | Kind=OPT_JOIN |  Length = 6   |Receiver Token (4 octets total):
      :  Receiver Token (continued)   |

                     Figure 8: Join Connection option

4.4.  General MPTCP Operation

   This section discusses operation of MPTCP for data transfer,
   independent of the path management mechanism used.

   At a high level, the an MPTCP implementation will take one input data
   stream from an application, and split it into one or more subflows.
   The data stream as a whole can be reassembled through the use of the
   Data Sequence Number (Figure 9) option, which defines the sequence in
   the data stream of the first octet of the packet's payload, and this
   is used by the receiver to ensure in-order delivery to th
   applicationlayers.  Meanwhile, the subflow-level sequence numbers
   (i.e. the regular TCP header sequence numbers) have subflow-only
   relevance.  The only acknowledgements are those at the subflow-level,
   so the sender must be able to map these acknowledgements to the data
   sequence numbers that were contained in the relevant packets.  The
   sender thus knows, if subflow data goes unackowledged, which part of
   the original data stream this equates to, and thus what data must be
   retransmitted.  It is expected (but not mandated) that SACK [RFC2018]

Ford, et al.            Expires November 8, 2009               [Page 16]

Internet-Draft                Multipath TCP                     May 2009

   is used as an efficiency at the subflow level.  Each subflow will
   maintain its own congestion widow.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      | Kind=OPT_DSN  |    Length     |      Data Sequence Number... :
      : ... ( (length-2) octets )     |

                   Figure 9: Data Sequence Number option

   As a TCP option contains a length field, the length of the Data
   Sequence Number can be declared implicitly.  Although it is expected
   that initial implementations will use 32-bit sequence numbers (i.e. 4
   octets, so a length field of 6), setting the length field to 10 and
   including a 64-bit sequence number (of four octets) MUST be
   considered valid and processed appropriately.  This may have also
   have useful security implications, discussed in Section 5.

   As wth the standard TCP sequence number, the data sequence number
   should not start at zero, but at a random value to make session hi-
   jacking harder.

   The Data Sequence Number is included in every MPTCP packet that
   contains data (or a DATA FIN, see Section 4.5), even if only one path
   is in use, so long as the MPTCP handshake has been completed and the
   endpoints have therefore agreed to use MPTCP.

   The MPTCP data and subflow level sequence numbering could be said to
   be analogous to that used in SACK, however there are subtle
   differences.  The key similarity is that it is possible to have
   temporary "holes" in the received data sequence space - later data
   may have arrived earlier (most likely on a different subflow), but
   does not need to be retransmitted.  The "holes" are later filled in.
   The key difference, however, is that while SACK can rely on the
   regular TCP cumulative acknowledgements to indicate how much data has
   been successfully received (with no holes), there is no similar
   method in MPTCP.  Instead, the sender must keep track of the
   acknowledgements to derive what data has been successfully received.
   This leads to some oddities especially with session termination (see
   Section 4.5).

4.4.1.  Subflow Policy

   Within a local MPTCP implementation, a host may use any policy it
   wishes to decide how to share the traffic to be sent over the

Ford, et al.            Expires November 8, 2009               [Page 17]

Internet-Draft                Multipath TCP                     May 2009

   available paths.

   In the typical use case, where the goal is to maximise throughput, it
   is necessary to couple the congestion windows in use on each subflow,
   in order to react in the most appropriate way to congestion on
   subflows.  This is the subject of significant theoretical and
   practical research outside the scope of this document.

   In other use cases, a user may split traffic across available
   subflows according to local policy.  Typically such cases would be an
   'all-or-nothing' approach, i.e. have a second path ready for use in
   the event of failure of the first path, but alternatives could
   include entirely saturating one path before using an additional path
   (the 'overflow' case).  Such choices would be most likely based on
   the monetary cost of links, but may also be based on properties such
   as delay or bandwidth, in cases where the additional paths are
   significantly worse and not worth including in the base operation.
   Other metrics such as this could be wrapped into an overall "cost"
   metric for a link.

   The ability to make effective choices at the sender requires full
   knowledge of the path characteristics, which is unlikely to be the
   case.  There is no mechanism in MPTCP for a receiver to signal their
   own particular preferences for paths, but this is a necessary feature
   since receivers will often be the multihomed party, such as in the
   case of laptop computers with wired and wireless connectivity.
   Instead of incorporating complex signalling, it is proposed to use
   existing TCP features to signal priority implicitly.  If a receiver
   wishes to keep a path active as a backup but wishes to prevent data
   being sent on that path, this could be achieved by the receiver not
   sending ACKs for any data it receives on that path.  The sender would
   interpret this as severe congestion or a broken path and stop using
   it.  We do not advocate this method, however, since this is brutal,
   naive, and will result in unnecessary retransmissions.

   Therefore, it is proposed to use ECN [RFC3168] to to provide fake
   congestion signals on paths that a receiver wishes to stop being used
   for data.  This has the benefit of causing the sender to back off
   without the need to retransmit data unnecessarily, as in the case of
   a lost ACK.  This should be sufficient to allow a receiver to express
   their policy, although does not permit a rapid increase in throughput
   when switching to such a path.  A potential solution to this would be
   that, if there is significant congestion, or the set of available
   paths has changed, MPTCP should wipe all subflow state and restart
   the multiplicative increase on all paths that appear uncongested.
   ECN will stop any paths that are still not required immediately,
   while the receiver's desired backup path will be in use and
   throughput will increase quickly.  This proposal should be no worse

Ford, et al.            Expires November 8, 2009               [Page 18]

Internet-Draft                Multipath TCP                     May 2009

   than current TCP.

4.4.2.  Retransmissions

   This protocol specification does not mandate any mechanisms for
   handling retransmissions in the event of path failures, and much will
   be dependent upon local policy (as discussed in Section 4.4.1).  The
   data sequence number, as given in a TCP option, is used to reassemble
   the incoming streams before presentation to the application layers,
   so a sender is free to re-send data with the same data sequence
   number on a different subflow.

   When doing this, it may be necessary to use the re-sync packet
   (Section 4.4.3) in order to skip over the subflow sequence numbers
   that were not retransmitted on the original subflow.  Of course, such
   a retransmission will only occur if this is what local policy
   suggests.  Indeed, it may be equally valid to retransmit on the same
   subflow if alternative paths have considerably worse quality of
   service, or are only kept for backup purposes.  Similarly, local
   implementation/policy will also determine how to modify the treatment
   of paths after packet loss - for example, how long to wait until
   returning to treating it as the preferred path.  Additionally, it may
   be possible for some implementations to signal from lower layers if
   there are problems with the paths, and so more appropriate responses
   can occur.

4.4.3.  Resync Packet

   The resync packet is used in certain circumstances when a sender
   needs to instruct the receiver to skip over certain subflow sequence
   numbers (i.e. to treat the specified sequence space as having been
   received and acknowledged).

   The typical use of this option will be when packets are retransmitted
   on different subflows, after failing to be acknowledged on the
   original subflow.  In such a case, it becomes necessary to move
   forward the original subflow's sequence numbering so as not to later
   transmit different data with a previously used sequence number (i.e.
   when more data comes to be transmitted on the original subflow, it
   would be different data, and so must not be sent with previously-used
   (but unacknowledged) sequence numbering).

   The rationale for needing to do this is two-fold: firstly, when ACKs
   are received they are for the subflow only, and the sender infers
   from this the data that was sent - if the same sequence space could
   be occupied by different data, the sender won't know whether the
   intended data was received.  Secondly, certain classes of middleboxes
   may cache data and not send the new data on a previously-seen

Ford, et al.            Expires November 8, 2009               [Page 19]

Internet-Draft                Multipath TCP                     May 2009

   sequence number.

   Therefore, it is necessary to 're-sync' the expected sequence
   numbering at the receiving end of a subflow, using the following TCP
   option.  This packet declares a sequence number space (inclusive)
   which the receiving node should skip over, i.e. if the receiver's
   next expected sequence number was previously within the range
   start_seq_num to end_seq_num, move it forward to end_seq_num + 1.

   This option will be used on the first new packet on the subflow that
   needs its sequence numbering re-synchronised.  It will be continue to
   be included on every packet sent on this subflow until a packet
   containing this option has been acknowledged (i.e. if subflow
   acknowledgements exist for packets beyond the end sequence number).
   If the end sequence number is earlier than the current expected
   sequence number (i.e. if a resync packet has already been received),
   this option should be ignored.

                           1                   2                   3
       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
      |Kind=OPT_RESYNC|  Length = 10  |     Start Sequence Number    :
      :          (4 octets)           |      End Sequence Number     :
      :          (4 octets)           |

                         Figure 10: Resync option

4.5.  Closing a Connection

   Under single path TCP, a FIN signifies that the sender has no more
   data to send.  In order to allow subflows to operate independently,
   however, and with as little change from regular TCP as possible, a
   FIN in MPTCP will only affect the subflow on which it is sent.  This
   allows nodes to exercise considerable freedom over which paths are in
   use at any one time.  The semantics of a FIN remain as for regular
   TCP, i.e. it is not until both sides have ACKed each other's FINs
   that the subflow is fully closed.

   When an application calls close() on a socket, this indicates that it
   has no more data to send, and for regular TCP this would result in a
   FIN on the connection.  For MPTCP, an equivalent mechanism is needed,
   and this is the DATA FIN.  This option, shown in Figure 11, is
   attached to a regular FIN option on a subflow.

   A DATA FIN is an indication that the sender has no more data to send,

Ford, et al.            Expires November 8, 2009               [Page 20]

Internet-Draft                Multipath TCP                     May 2009

   and as such can be used as a rapid indication of the end of data from
   a sender.  Therefore, it is an optimisation to clean up state
   associated with a MPTCP connection, especially when some subflows may
   have failed.  Specifically, when a DATA FIN has been received, IF all
   data has been successfully received, timeouts on all subflows MAY be
   reduced.  Similarly, when sending a DATA FIN, once all data
   (including the DATA FIN has been acknowledged, FINs must be sent on
   every subflow.  This applies to both endpoints, and is required in
   order to clean up state in middleboxes.

   There are complex interactions, however, between a DATA FIN and
   subflow properties:

   o  A DATA FIN can only be sent on a packet which also has the FIN
      flag set.

   o  A DATA FIN occupies one octet (the final octet) of Data Sequence
      Number space.  Therefore, even if there is no user data, a Data
      Sequence Number option must be added to a packet containing the
      DATA FIN option.  This allows the receiver to easily determine the
      last data sequence number that should have been received.

   o  There is a one-to-one mapping between the DATA FIN and the
      subflow's FIN flag (and its associated sequence space and thus its
      acknowlegement).  In other words, when a subflow's FIN flag has
      been acknowledged, the associated DATA FIN is also acknowledged.

   o  As such, the acknowledgement of a FIN and DATA FIN DOES NOT
      indicate that all data has been successfully received; this must
      wait for all subflows to acknowledge.

   It should be noted that an endpoint may also send a FIN on an
   individual subflow to shut it down, but this impact is limited to the
   subflow in question.  If all subflows have been closed with a FIN,
   that is equivalent to having closed the connection with a DATA FIN.

       0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
      | Kind=OPT_DFIN |   Length = 2  |

                        Figure 11: DATA FIN option

Ford, et al.            Expires November 8, 2009               [Page 21]

Internet-Draft                Multipath TCP                     May 2009

4.6.  Error Handling


   Unknown token in MPTCP SYN should equate to an unknown port, e.g. a
   TCP reset?  We should make this as silent and tolerant as possible.
   Where possible, we should keep this close to the semantics of TCP.
   The amount of error handling required may also have an impact on the
   choice of path management schemes.  Issues may include odd cases
   where a data sequence number is missing from a subflow.  Will
   definitely need errors in those cases.

5.  Security Considerations


   (Token generation, handshake mechanisms, new subflow authentication,

   The development of a TCP extension such as this will bring with it
   many additional security concerns.  We have set out here to produce a
   solution that is "no worse" than current TCP, with the possibility
   that more secure extensions could be proposed later.

   The primary area of concern will be around the handshake to start new
   subflows which join existing connections.  The proposal set out in
   Section 4.1 and Section 4.3 is for the initiator of the new subflow
   to include the token of the other endpoint in the handshake.  The
   purpose of this is to indicate that the sender of this token was the
   same entity that received this token at the initial handshake.

   One area of concern is that the token could be simply brute-forced.
   The token must behard to guess, and as such could be randomly
   generated.  This may still not be strong enough, however, and so the
   use of 64 bits for the token would alleviate this somewhat.

   Use of these tokens only provide an indication that the token is the
   same as at the initial handshake, and does not say anything about the
   current sender of the token.  Therefore, another approach would be to
   bring a new measure of freshness in to the handshake, so instead of
   using the initial token a sender could request a new token from the
   receiver to use in the next handshake.

   Yet another alternative would be for the SYN packet to include a data
   sequence number.  This could either be used as a passive identifier
   to indicate an awareness of the current data sequence number
   (although a reasonable window would have to be allowed for delays).

Ford, et al.            Expires November 8, 2009               [Page 22]

Internet-Draft                Multipath TCP                     May 2009

   Or, the SYN could form part of the data sequence space - but this
   would cause issues in the event of lost SYNs (if a new subflow is
   never established), thus causing unnecessary delays for

   The "Request-FIN" option (if included) is possibly vulnerable to TCP-
   Reset style attacks, however the presense of the subflow and data-
   level sequence numbers should provide some level of freshness

6.  Interactions with Middleboxes


   How we get around NATs, firewalls.  Problems with TCP proxies.  How
   to make an MPTCP-aware middlebox, ...

7.  Interfaces


   Interface with applications, interface with TCP, interface with lower

8.  Acknowledgements

   The authors are supported by Trilogy
   (, a research project (ICT-216372)
   partially funded by the European Community under its Seventh
   Framework Program.  The views expressed here are those of the
   author(s) only.  The European Commission is not liable for any use
   that may be made of the information in this document.

   The authors gratefully acknowledge significant input into this
   document from many members of the Trilogy project, notably Iljitsch
   van Beijnum, Lars Eggert, Marcelo Bagnulo Braun, Robert Hancock, Pasi
   Sarolahti, Olivier Bonaventure, Toby Moncaster, Philip Eardley and
   Andrew McDonald.

9.  IANA Considerations

   This document will make a request to IANA to allocate new values for
   TCP Option identifiers, as follows:

Ford, et al.            Expires November 8, 2009               [Page 23]

Internet-Draft                Multipath TCP                     May 2009

   |   Symbol   |       Name      |    PM    |       Ref       | Value |
   |   OPT_MPC  |    Multipath    |     -    |   Section 4.1   | (tbc) |
   |            |     Capable     |          |                 |       |
   |  OPT_ADDR  |   Add Address   | Explicit | Section | (tbc) |
   | OPT_REMADR |  Remove Address | Explicit | Section | (tbc) |
   | OPT_REQSYN |   Request-SYN   | Implicit | Section | (tbc) |
   | OPT_REQFIN |   Request-FIN   | Implicit | Section | (tbc) |
   |  OPT_JOIN  | Join Connection |     -    |   Section 4.3   | (tbc) |
   |   OPT_DSN  |  Data Sequence  |     -    |   Section 4.4   | (tbc) |
   |            |      Number     |          |                 |       |
   | OPT_RESYNC |     Re-sync     |     -    |  Section 4.4.3  | (tbc) |
   |  OPT_DFIN  |     DATA FIN    |     -    |   Section 4.5   | (tbc) |

                      Table 1: TCP Options for MPTCP

10.  References

10.1.  Normative References

   [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
              Requirement Levels", BCP 14, RFC 2119, March 1997.

10.2.  Informative References

              Eddy, W. and A. Langley, "Extending the Space Available
              for TCP Options", draft-eddy-tcp-loo-04 (work in
              progress), July 2008.

              Nordmark, E. and M. Bagnulo, "Shim6: Level 3 Multihoming
              Shim Protocol for IPv6", draft-ietf-shim6-proto-12 (work
              in progress), February 2009.

              van Beijnum, I., "One-ended Multipath TCP",
              draft-van-beijnum-1e-mp-tcp-00 (work in progress),
              May 2009.

   [RFC0793]  Postel, J., "Transmission Control Protocol", STD 7,
              RFC 793, September 1981.

   [RFC2018]  Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP
              Selective Acknowledgment Options", RFC 2018, October 1996.

Ford, et al.            Expires November 8, 2009               [Page 24]

Internet-Draft                Multipath TCP                     May 2009

   [RFC3168]  Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
              of Explicit Congestion Notification (ECN) to IP",
              RFC 3168, September 2001.

   [RFC4960]  Stewart, R., "Stream Control Transmission Protocol",
              RFC 4960, September 2007.

   [RFC5061]  Stewart, R., Xie, Q., Tuexen, M., Maruyama, S., and M.
              Kozuka, "Stream Control Transmission Protocol (SCTP)
              Dynamic Address Reconfiguration", RFC 5061,
              September 2007.

   [RFC5062]  Stewart, R., Tuexen, M., and G. Camarillo, "Security
              Attacks Found Against the Stream Control Transmission
              Protocol (SCTP) and Current Countermeasures", RFC 5062,
              September 2007.

   [WISCHIK]  Wischik, D., Handley, M., and M. Bagnulo Braun, "The
              Resource Pooling Principle", ACM SIGCOMM CCR vol. 38 num.
              5, pp. 47-52, October 2008,

Appendix A.  Functional Separation

   [Potential to move to separate architectural document]

   This section describes the functional separation that drives the
   design of the MPTCP protocol.  Its main goal is to separate MPTCP in
   two parts that communicate through a well defined interface.  We
   first provide the motivations for this functional separation, then we
   describe in more details the two main components of the MPTCP

A.1.  Motivations

   The major goal behind MPTCP is to send data over different paths in
   the same time.  This assumes that an MPTCP implementation must be
   able to discover and use the multiple paths that connect two given
   hosts, when they exist.  However, different mechanisms can be
   envisioned for multipath discovery and use.  Examples are as follows:

   Use multiple addresses:  This is the method currently proposed in
      this document - if hosts are multi-addressed, different address
      pairs may take different routes.

Ford, et al.            Expires November 8, 2009               [Page 25]

Internet-Draft                Multipath TCP                     May 2009

   Use a path selector value:  An end-host might be able to tag packets
      with a path selector value, or "colour".  If some network nodes
      are able to read the colour and use it as a path selector, the
      host can influence the outgoing path of the packet.

   Next-hop selection:  In a network configuration where multiple next-
      hops can offer to forward packets, a host may decide to send some
      of its packets through one next-hop, and some through another.

   The above list is not exhaustive, and could grow as new network
   technologies are deployed.

A.2.  TCP Performance

   In addition to purely sending data over multiple paths, MTCP must do
   it in a way that will not affect TCP performance.  This raises the
   need for an efficient multipath congestion control algorithm.  While
   this specification does not mandate the use of any particular
   algorithm for congestion control, it ensures that the protocol is
   designed in such a way that any CC algorithm can be designed,
   independently of the particular path management mechanism available
   to the host.  Consequently our architecture for MTCP decouples the
   policy from the mechanism.  The policy is the decision of what path
   to use for each packet to send.  It is mainly driven by the
   implementation-dependent congestion control algorithm.  The mechanism
   is the technology used to ensure that a packet will be sent on the
   desired path.  This separation is intended to be relatively future-
   proof by allowing these components to evolve at different speeds.

A.3.  Architecture overview

Ford, et al.            Expires November 8, 2009               [Page 26]

Internet-Draft                Multipath TCP                     May 2009

            Control plane    <--     |     -->    Data plane
   |                     Multipath Scheduler (MPS)                 |
                ^                    |          |
                |                    |          |
                |Announcing new      |   +-------------+
                |paths. (referred    |   | Data packet |<--Path idx:3
                |to as path indices) |   +-------------+   attached
                |                    |          |          by MPS
                |                    |          V
   |                         Path Manager (PM)   \__________zzzzz  |
      /                   \          |                       \
     /---------------------\         |   /"\       /"\       /"\
     | Path key    Action  |         |   | |       | |       | |
     |     1        xxxxx  |         |   | |       | |       | |
     |     2        yyyyy  |         |   \./       \./       \./
     |     3        zzzzz  |         |  path1     path2     path3

                 Figure 12: Overview of MTCP architecture

   A general overview of the architecture is provided in Figure 12.  The
   Multipath Scheduler (MPS) learns about the number of available paths
   through notifications received from the Path Manager (PM).  From the
   point of view of the Multipath Scheduler, a path is just a number,
   called a Path Index.  Notifications from the PM to the MPS MAY
   contain supporting information about the paths, if relevant, so that
   the MPS can make more intelligent decisions about where to route
   traffic.  When the Multipath Scheduler initiates a communication to a
   new host, it can only send the packets to the default path.  But
   since the Path manager is layered below the MPS, it can detect that a
   new communication is happening, and tell the MPS about the other
   paths it knows about.

   From then on, it is possible for the MPS to attach a Path Index to
   the control structure of its packets (internal to the MTCP
   implementation), so that the Path Manager can map this Path Index to
   the corresponding action. (see table in the lower left part of
   Figure 12).  The particular action depends on the network mechanism
   used to select a path.  Examples are address rewriting, tunnelling or
   setting a path selector valude inside the packet.

   The applicability of the architecture is not limited to the MTCP
   protocol.  While we define in this document an MTCP MPS (MTCP
   Multipath Scheduler), other Multipath Schedulers can be defined.  For

Ford, et al.            Expires November 8, 2009               [Page 27]

Internet-Draft                Multipath TCP                     May 2009

   example, if an appropriate socket interface is designed, applications
   could behave as a Multipath Scheduler and decide where to send any
   particular data.  In this document we concentrate on the MTCP case,

   In this specification, we define the core protocol for Multipath TCP.
   The core protocol is not dependent on the Path Management technique
   that is chosen, and MUST be implemented in any MTCP MPS.  We also
   provide a default Path Manager that is based on declaring IP
   addresses, and carries control information in TCP options.  An
   implementation of Multipath TCP can use any Path Manager, but it MUST
   be able to fallback to the default PM in case the other end does not
   support the custom PM.  Alternative Path Managers may be specified in
   separate documents in the future.

A.4.  PM/MPS interface

   The minimal set of requirement for a Path Manager is as follows:

   o  Outgoing untagged packets: Any outgoing packet flowing through the
      Path Manager is either tagged or untagged (by the MPS) with a path
      index.  If it is untagged, the packet is sent normally to the
      Internet, as if no multi-path support were present.  Untagged
      packets can be used to trigger a path discovery procedure, that
      is, a Path Manager can listen to untagged packets and decide at
      some time to find if any other path than the default one is
      useable for the corresponding host pair.  Note that any other
      criteria could be used to decide when to start discovering
      available paths.  Note also that MPS scheduling will not be
      possible until the Path Manager has notified the available paths.
      The PM is thus the first entity coming into action.

   o  Outgoing tagged packets: The Path Manager maintains a table
      mapping path indices to actions.  The action is the operation that
      allows using a particular path.  Examples of possible actions are
      route selection, interface selection or packet transformation.
      When the PM sees a packet tagged with a path index, it looks up
      its table to find the appropriate action for that packet.  The tag
      is purely local.  It is removed before the packet is transmitted.

   o  Incoming packets: A Path Manager MUST ensure that incoming path is
      mapped unambiguously to exactly one outgoing path.  Note that this
      requirement implies that the same number of incoming/outgoing
      paths must be established.  Moreover, a PM MUST tag any incoming
      path with the same Path Index as the one used for the
      corresponding outgoing path.  This is necessary for MTCP to know
      what outgoing path in acknowledged by an incoming packet.

Ford, et al.            Expires November 8, 2009               [Page 28]

Internet-Draft                Multipath TCP                     May 2009

   o  Module interface: A PM MUST be able to notify the MPS about the
      number of available paths.  Such notifications MUST contain the
      path indices that are legal for use by the MPS.  In case the PM
      decides to stop providing service for one path, it MUST notify the
      MPS about path deletion.  Additionnaly, a PM MAY provide
      complementary path information when available, such as link
      quality or preference level.

Appendix B.  Notes on use of TCP Options

   The TCP option space is limited due to the length of the Data Offset
   field in the TCP header (4 bits), which defines the TCP header length
   in 32-bit words.  With the standard TCP header being 20 bytes, this
   leaves a maximum of 40 bytes for options, and many of these may
   already be used by options such as timestamp and SACK.

   As such, when doing address list manipulation, not all data may fit.
   This can be mitigated in one of two ways:

   o  Using an option to extend the option space, such as that proposed
      in [I-D.eddy-tcp-loo], which proposes an option providing a 16-bit
      header length field.  Such an option could only be used between
      nodes that support it, however, and so long options could not be
      used until a handshake is complete.

   o  Alternatively, since at least one IP address option field should
      be able to fit per packet, address list manipulation can be
      undertaken with one address per packet.  One method could be to
      wait for data to send, and then append one new address per packet.
      This would seem reasonable if the TCP session begins rapidly, but
      if it is required that the multipath session is ready before the
      first data is to be sent, address list manipulation would be
      required on empty data (signalling only) packets.  Issues may
      arise regarding acknowledged delivery of signalling versus data -
      this is discussed in Section 3 below.

Ford, et al.            Expires November 8, 2009               [Page 29]

Internet-Draft                Multipath TCP                     May 2009

Authors' Addresses

   Alan Ford (editor)
   Roke Manor Research
   Old Salisbury Lane
   Romsey, Hampshire  SO51 0ZN

   Phone: +44 1794 833 465

   Costin Raiciu
   University College London


   Mark Handley
   University College London

   Sebastien Barre
   Universite catholique de Louvain
   Pl. Ste Barbe, 2
   Louvain-la-Neuve  1348

   Phone: +32 10 47 91 03

Ford, et al.            Expires November 8, 2009               [Page 30]