[Docs] [txt|pdf] [Tracker] [Email] [Diff1] [Diff2] [Nits]

Versions: 00 01 02 03 04 05 06 07

Internet Engineering Task Force                                R. Winter
Internet-Draft                                   NEC Laboratories Europe
Intended status: Informational                                  M. Faath
Expires: September 22, 2016      University of Applied Sciences Augsburg
                                                                A. Ripke
                                                 NEC Laboratories Europe
                                                          March 21, 2016

           Multipath TCP Support for Single-homed End-systems


   Multipath TCP relies on the existence of multiple paths between end-
   systems.  These are typically provided by using different IP
   addresses obtained by different ISPs at the end-systems.  While this
   scenario is certainly becoming increasingly a reality (e.g. mobile
   devices), currently most end-systems are single-homed (e.g. desktop
   PCs in an enterprise).  It seems also likely that a lot of network
   sites will insist on having all traffic pass a single network element
   (e.g. for security reasons) before traffic is split across multiple
   paths.  This memo therefore describes mechanisms to make multiple
   paths available to multipath TCP-capable end-systems that are not
   available directly at the end-systems but somewhere within the

Status of This Memo

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

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF).  Note that other groups may also distribute
   working documents as Internet-Drafts.  The list of current Internet-
   Drafts is at http://datatracker.ietf.org/drafts/current/.

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

   This Internet-Draft will expire on September 22, 2016.

Winter, et al.         Expires September 22, 2016               [Page 1]

Internet-Draft             single-homed MPTCP                 March 2016

Copyright Notice

   Copyright (c) 2016 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
   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.  Code Components extracted from this document must
   include Simplified BSD License text as described in Section 4.e of
   the Trust Legal Provisions and are provided without warranty as
   described in the Simplified BSD License.

Table of Contents

   1.  Introduction  . . . . . . . . . . . . . . . . . . . . . . . .   2
   2.  Approaches to Use Multiple Paths in the Network . . . . . . .   3
     2.1.  Exposing Multiple Paths Through End-host Auto-
           configuration . . . . . . . . . . . . . . . . . . . . . .   3
     2.2.  Heuristic Use of Multiple Paths . . . . . . . . . . . . .   5
   3.  Other scenarios and extensions  . . . . . . . . . . . . . . .   6
   4.  Alternative approaches  . . . . . . . . . . . . . . . . . . .   6
   5.  Acknowledgements  . . . . . . . . . . . . . . . . . . . . . .   6
   6.  IANA Considerations . . . . . . . . . . . . . . . . . . . . .   6
   7.  Security Considerations . . . . . . . . . . . . . . . . . . .   7
   8.  References  . . . . . . . . . . . . . . . . . . . . . . . . .   7
     8.1.  Normative References  . . . . . . . . . . . . . . . . . .   7
     8.2.  Informative References  . . . . . . . . . . . . . . . . .   7
   Authors' Addresses  . . . . . . . . . . . . . . . . . . . . . . .   8

1.  Introduction

   The IETF has specified a multipath TCP (MPTCP) architecture and
   protocol where end-systems operate a modified standard TCP stack
   which allows packets of the same TCP connection to be sent via
   different paths to an MPTCP-capable destination ([RFC6824],
   [RFC6182]).  Paths are defined by sets of source and destination IP
   addresses.  Using multiple paths has a number of benefits such as an
   increased reliability of the transport connection and an effect known
   as resource pooling [resource_pooling].  Most end-systems today do
   not have multiple paths/interfaces available in order to make use of
   multipath TCP, however further within the network multiple paths are
   the norm rather than the exception.  This memo therefore describes
   ways how these multiple paths in the network could potentially be
   made available to multipath TCP-capable hosts that are single-homed.

Winter, et al.         Expires September 22, 2016               [Page 2]

Internet-Draft             single-homed MPTCP                 March 2016

   In order to illustrate the general mechanism we make use of a simple
   reference scenario shown in Figure 1.

                               | DHCP  |
     +-------+      +----------+ Server|
     |       |      |          |       |
     | Host  +------+          +-------+
     |       |      |      +-------+        ISP 1
     +-------+      +------+       |----------
                           | Gatew.|
                           |       |----------
                           +-------+        ISP 2

                       Figure 1: Reference Scenario

   The scenario in Figure 1 depicts e.g. a possible SOHO or enterprise
   setup where a gateway/router is connected to two ISPs and a DHCP
   server gives out leases to hosts connected to the local network.
   Note that both, the gateway and the DHCP server could be on the same
   device (similar to current home gateway implementations).  Also, the
   two ISPs could really be two different access technologies (e.g.  LTE
   and DSL) provided by a single ISP.

   The host is running a multipath-capable IP stack, however it only has
   a single interface.  The methods described in the following sections
   will let the host make use of the gateway's two interfaces without
   requiring modifications to the MPTCP implementation.

2.  Approaches to Use Multiple Paths in the Network

   All approaches in this document do not require changes to the wire
   format of MPTCP and both communicating hosts need to be MPTCP-
   capable.  The benefit this approach has is that a) it has no
   implications on MPTCP standards, b) it will hopefully encourage the
   deployment of MPTCP as the number of scenarios where MPTCP brings
   benefits vastly increases and c) these approaches do not require
   complex middle-boxes to implement MPTCP-like functionality in the
   network as other approaches have suggested before.

2.1.  Exposing Multiple Paths Through End-host Auto-configuration

   Multipath TCP distinguishes paths by their source and destination IP
   addresses.  Assuming a certain level of path diversity in the
   Internet, using different source and destination IP addresses for a
   given subflow of a multipath TCP connection will, with a certain
   probability, result in different paths taken by packets of different
   subflows.  Even in case subflows share a common bottleneck, the

Winter, et al.         Expires September 22, 2016               [Page 3]

Internet-Draft             single-homed MPTCP                 March 2016

   proposed multipath congestion control algorithm [RFC6356] will make
   sure that multipath TCP will play nicely with regular TCP flows.

   In order to not require changes to the TCP implementation, we keep
   the above assumptions multipath TCP makes, i.e. working with
   different IP addresses to use different paths.  Since the end-system
   is single-homed, all IP addresses are bound to the same physical
   interface.  In our reference scenario in Figure 1, the host would
   e.g. receive more than one RFC1918 [RFC1918] private IP address from
   the DHCP server as depicted in Figure 2.

       Host                Gateway

                         +-----------------+     ISP1
     +--------+          | src.            |
     |  virt. | |  __.+----------
     |        +---+      |        __.--'   |
     |  phys. |   |      |  __.--'      N  |
     |        +----------+.:_           A  |
     |        | |   `-.._      T  |
     +--------+          | src.   `-.._    |     ISP2
                         | `-..+----------
                         |                 |

                        Figure 2: Gateway internals

   The gateway that is shown in Figure 2 has received two IP addresses,
   one from each ISP that it is connected to (ISP1 and ISP2).  The NAT
   that the gateway is implementing needs to "map" each private IP
   address of the host consistently to a one of the addresses received
   by the ISPs, i.e. each private IP to a different public IP.  Packets
   sent by the host to the gateway are then routed based on the source
   address found in the packets as illustrated in the figure.  In other
   words, depending on the source address of the host, the packets will
   either go through ISP 1 or ISP 2 and TCP will balance the traffic
   across those two links using its built-in congestion control

   The way the gateway has received its public IP addresses is not
   relevant.  It could be via DHCP, IPCP or static configuration.  In
   order to configure the hosts behind the gateway, we propose to make
   use of provisioning domains [RFC7556], more specifically one
   provisioning domain per external gateway interface (the two
   interfaces to ISP1 and ISP2 in Figure 2).  The DHCPv6 specification
   for encoding provisioning domains can be found in

Winter, et al.         Expires September 22, 2016               [Page 4]

Internet-Draft             single-homed MPTCP                 March 2016

   In order to signal to the host, that each provisioning domain will
   result in a different path towards the Internet, this memo introduces
   a new DHCP option called EXT_ROUTE, which will be included in each
   provisioning domain sent by the server.  The option value will
   determine which external interface is used to sent the traffic when
   using the configuration information present in the respective
   provisioning domain.

   Upon receipt of a DHCP offer including multiple provisioning domains,
   or multiple offers each including one or more provisioning domains,
   the client SHOULD create up to n virtual interfaces, where n is one
   less than the number of different EXT_ROUTE option values found in
   all received provisioning domains.  Each virtual interface will
   contact the DHCP server and will request configuration information
   for the respective provisioning domains, excluding the configuration
   of the physical interface.

2.2.  Heuristic Use of Multiple Paths

   The auto-configuration mechanism above has the advantage that
   available paths and information on how to use them are directly sent
   to the end-host.  In other words, there is an explicit signalling of
   the availability of multiple paths to the end-host.  This has the
   advantage that the host can efficiently use these paths.

   This method works well when multiple paths are available close to the
   end-host and means for auto-configuration are available.  But that is
   not always the case.  Another method to use different paths in the
   network without prior knowledge of their existence is to apply
   heuristics in order to exploit setups where Equal Cost Multi-path
   [RFC2991], a widely deployed technology [ECMP_DEPLOYMENT], or similar
   per-flow load-balancing algorithms are employed.

   The ADD_ADDR option defined in [RFC6824] can be used to advertise the
   same address but a different port to open another subflow.
   Additionally, the MP_JOIN option can also be used to open another
   subflow with the same IP address and e.g. a different source port
   given that a different address ID is used.  This means there are
   multiple scenarios possible (e.g. either sender-initated or receiver-
   initiated) where single-homed end-hosts can influence the 5-tuple
   (source and destination IP addresses and port numbers plus protocol
   number) which is often used as the basis for per-flow load balancing.
   Changing the 5-tuple will only with a certain probability result in
   using a different path unless the load-balancing algorithm that is
   used is known to the MPTCP implementation (an assumption we cannot
   generally make).  This means that a number of subflows might end up
   on the same path.  Fortunately, the MPTCP congestion control

Winter, et al.         Expires September 22, 2016               [Page 5]

Internet-Draft             single-homed MPTCP                 March 2016

   algorithm will make sure that the collection of subflows on that path
   will not be more agressive than a single TPC flow.

3.  Other scenarios and extensions

   The reference scenario is only one conceivable setting.  Other
   scenarios such as DSL broadband customers or mobile phones are
   conceivable as well.  As an example, take the DSL scenario.  The home
   gateway could be provided with multiple IP addresses using extensions
   to IPCP.  The home gateway in turn can then implement the DHCP server
   and gateway functionality as described before.  More scenarios will
   be described in future versions of this document.

4.  Alternative approaches

   One alternative is that a DHCP server always sends n offers, where n
   is the number of interfaces at the gateway to different ISPs.  The
   client could then accept all or a subset of these offers.  This
   approach seems interesting in environments where there are multiple
   DHCP servers, one for each ISP connection (think multiple home
   gateways).  However, accepting multiple offers based on a single DHCP
   request is not standard's compliant behavior (at least for the DHCPv4
   case).  Also, to cater for a scenario that only contains a single
   DHCP server, server changes are needed in any case.  Finally, correct
   routing is not always guaranteed in these scenarios.

   An interesting alternative is the use of ECMP at the gateway for load
   distribution and let MPTCP use different port numbers for subflows.
   Assuming that ECMP is available at the gateway, this approach would
   work fine today.  The only drawback of the approach is that it
   involves a little trial and error to find port numbers that actually
   hash to different paths used by ECMP [RFC2991].

5.  Acknowledgements

   Part of this work was supported by Trilogy (http://www.trilogy-
   project.org), 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.

6.  IANA Considerations

   One new DHCP options is required by this version of this document.

Winter, et al.         Expires September 22, 2016               [Page 6]

Internet-Draft             single-homed MPTCP                 March 2016

7.  Security Considerations


8.  References

8.1.  Normative References

   [RFC1918]  Rekhter, Y., Moskowitz, B., Karrenberg, D., de Groot, G.,
              and E. Lear, "Address Allocation for Private Internets",
              BCP 5, RFC 1918, DOI 10.17487/RFC1918, February 1996,

   [RFC2991]  Thaler, D. and C. Hopps, "Multipath Issues in Unicast and
              Multicast Next-Hop Selection", RFC 2991, DOI 10.17487/
              RFC2991, November 2000,

   [RFC7556]  Anipko, D., Ed., "Multiple Provisioning Domain
              Architecture", RFC 7556, DOI 10.17487/RFC7556, June 2015,

8.2.  Informative References

              Augustin, B., Friedman, T., and R. Teixeira, "Measuring
              Multipath Routing in the Internet", October 2011,

              Krishnan, S., Korhonen, J., and S. Bhandari, "Support for
              multiple provisioning domains in DHCPv6", draft-ietf-mif-
              mpvd-dhcp-support-02 (work in progress), October 2015.

   [RFC6182]  Ford, A., Raiciu, C., Handley, M., Barre, S., and J.
              Iyengar, "Architectural Guidelines for Multipath TCP
              Development", RFC 6182, DOI 10.17487/RFC6182, March 2011,

   [RFC6356]  Raiciu, C., Handley, M., and D. Wischik, "Coupled
              Congestion Control for Multipath Transport Protocols", RFC
              6356, DOI 10.17487/RFC6356, October 2011,

   [RFC6824]  Ford, A., Raiciu, C., Handley, M., and O. Bonaventure,
              "TCP Extensions for Multipath Operation with Multiple
              Addresses", RFC 6824, DOI 10.17487/RFC6824, January 2013,

Winter, et al.         Expires September 22, 2016               [Page 7]

Internet-Draft             single-homed MPTCP                 March 2016

              Wischik, D., Handley, M., and M. Bagnulo Braun, "The
              Resource Pooling Principle", October 2008,

Authors' Addresses

   Rolf Winter
   NEC Laboratories Europe
   Kurfuersten-Anlage 36
   Heidelberg  69115

   Email: rolf.winter@neclab.eu

   Michael Faath
   University of Applied Sciences Augsburg
   An der Hochschule 1
   Augsburg  86161

   Email: michael.faath@hs-augsburg.de

   Andreas Ripke
   NEC Laboratories Europe
   Kurfuersten-Anlage 36
   Heidelberg  69115

   Email: andreas.ripke@neclab.eu

Winter, et al.         Expires September 22, 2016               [Page 8]

Html markup produced by rfcmarkup 1.128b, available from https://tools.ietf.org/tools/rfcmarkup/