RFC 8039

Internet Engineering Task Force (IETF)                        A. Shpiner
Request for Comments: 8039                                      Mellanox
Category: Experimental                                            R. Tse
ISSN: 2070-1721                                                Microsemi
                                                               C. Schelp
                                                              T. Mizrahi
                                                           December 2016

                     Multipath Time Synchronization


   Clock synchronization protocols are very widely used in IP-based
   networks.  The Network Time Protocol (NTP) has been commonly deployed
   for many years, and the last few years have seen an increasingly
   rapid deployment of the Precision Time Protocol (PTP).  As time-
   sensitive applications evolve, clock accuracy requirements are
   becoming increasingly stringent, requiring the time synchronization
   protocols to provide high accuracy.  This memo describes a multipath
   approach to PTP and NTP over IP networks, allowing the protocols to
   run concurrently over multiple communication paths between the master
   and slave clocks, without modifying these protocols.  The multipath
   approach can significantly contribute to clock accuracy, security,
   and fault tolerance.  The multipath approach that is presented in
   this document enables backward compatibility with nodes that do not
   support the multipath functionality.

Status of This Memo

   This document is not an Internet Standards Track specification; it is
   published for examination, experimental implementation, and

   This document defines an Experimental Protocol for the Internet
   community.  This document is a product of the Internet Engineering
   Task Force (IETF).  It represents the consensus of the IETF
   community.  It has received public review and has been approved for
   publication by the Internet Engineering Steering Group (IESG).  Not
   all documents approved by the IESG are a candidate for any level of
   Internet Standard; see Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at

Shpiner, et al.               Experimental                      [Page 1]

RFC 8039             Multipath Time Synchronization        December 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 ....................................................3
   2. Conventions Used in This Document ...............................4
      2.1. Abbreviations ..............................................4
      2.2. Terminology ................................................4
   3. Multiple Paths in IP Networks ...................................5
      3.1. Load Balancing .............................................5
      3.2. Using Multiple Paths Concurrently ..........................5
      3.3. Two-Way Paths ..............................................5
   4. Solution Overview ...............................................6
      4.1. Path Configuration and Identification ......................6
      4.2. Combining ..................................................6
   5. Multipath Time Synchronization over IP Networks .................7
      5.1. Overview ...................................................7
      5.2. Single-Ended Multipath Synchronization .....................8
           5.2.1. Single-Ended MPPTP Synchronization Message
                  Exchange ............................................8
           5.2.2. Single-Ended MPNTP Synchronization Message
                  Exchange ............................................9
      5.3. Dual-Ended Multipath Synchronization ......................10
           5.3.1. Dual-Ended MPPTP Synchronization Message Exchange ..10
           5.3.2. Dual-Ended MPNTP Synchronization Message Exchange ..11
      5.4. Using Traceroute for Path Discovery .......................12
      5.5. Using Unicast Discovery for MPPTP .........................13
   6. Combining Algorithm ............................................13
   7. Security Considerations ........................................14
   8. Scope of the Experiment ........................................14
   9. References .....................................................15
      9.1. Normative References ......................................15
      9.2. Informative References ....................................15
   Acknowledgments ...................................................17
   Authors' Addresses ................................................17

Shpiner, et al.               Experimental                      [Page 2]

RFC 8039             Multipath Time Synchronization        December 2016

1.  Introduction

   The two most common time synchronization protocols in IP networks are
   (1) the Network Time Protocol [NTP] and (2) the Precision Time
   Protocol (PTP) as defined in the IEEE 1588 standard [IEEE1588].

   The accuracy of the time synchronization protocols directly depends
   on the stability and the symmetry of propagation delays in both
   directions between the master and slave clocks.  Depending on the
   nature of the underlying network, time synchronization protocol
   packets can be subject to variable network latency or path asymmetry
   (e.g., [ASYMMETRY] [ASYMMETRY2]).  As time-sensitive applications
   evolve, accuracy requirements are becoming increasingly stringent.
   Using a single network path in a clock synchronization protocol
   closely ties the slave clock accuracy to the behavior of the specific
   path, which may suffer from temporal congestion, faults, or malicious
   attacks.  Relying on multiple clock servers, as in NTP, solves these
   problems but requires active maintenance of multiple accurate sources
   in the network, which is not always possible.  The usage of
   Transparent Clocks (TCs) in PTP solves the congestion problem by
   eliminating the queuing time from the delay calculations but does not
   address security or fault-tolerance aspects.

                            ______/    \_____
                        ___/                 \____
                   ____/                          \
       ____       /           path 1              /           ___
      /    \     /    ________________________    \          /   \
     /Master\____\___/                        \____\________/Slave\
     \Clock /    /   \________ _______________/     \       \Clock/
      \____/     \                                  /        \__ /
                  \____       path 2             __/
                       \_______           ______/

                      Figure 1: Multipath Connection

   Since master and slave clocks are often connected through more than
   one path in the network, as shown in Figure 1, [SLAVEDIV] suggested
   that a time synchronization protocol can be run over multiple paths,
   providing several advantages.  First, it can significantly increase
   the clock accuracy as shown in [SLAVEDIV].  Second, this approach
   provides additional security, allowing the mitigation of
   man-in-the-middle attacks against the time synchronization protocol
   [DELAY-ATT].  Third, using multiple paths concurrently provides an
   inherent failure protection mechanism.

Shpiner, et al.               Experimental                      [Page 3]

RFC 8039             Multipath Time Synchronization        December 2016

   This document introduces Multipath PTP (MPPTP) and Multipath NTP
   (MPNTP).  The functionality of the multipath approach is defined at
   the network layer and does not require any changes in PTP or NTP.

   MPPTP and MPNTP are defined over IP networks.  As IP networks
   typically combine ECMP routing, this property is leveraged for the
   multiple paths used in MPPTP and MPNTP.  The key property of the
   multipath approach is that clocks in the network can use more than
   one IP address.  Each {master IP, slave IP} address pair defines a
   path.  Depending on the network topology and configuration, the IP
   combination pairs can form multiple diverse paths used by the
   multipath synchronization protocols.  It has been shown [MULTI] that
   using multiple IP addresses over the wide Internet indeed allows two
   endpoints to attain multiple diverse paths.

   This document introduces two variants of the multipath approach:
   (1) a variant that requires both master and slave nodes to support
   the multipath functionality, referred to as the dual-ended variant,
   and (2) a backward-compatible variant that allows a multipath clock
   to connect to a conventional single-path clock, referred to as the
   single-ended variant.

2.  Conventions Used in This Document

2.1.  Abbreviations

   BMC      Best Master Clock [IEEE1588]

   ECMP     Equal-Cost Multipath

   LAN      Local Area Network

   MPNTP    Multipath Network Time Protocol

   MPPTP    Multipath Precision Time Protocol

   NTP      Network Time Protocol [NTP]

   PTP      Precision Time Protocol [IEEE1588]

2.2.  Terminology

   In the NTP terminology, a time synchronization protocol is run
   between a client and a server, while PTP uses the terms 'master' and
   'slave'.  Throughout this document, the sections that refer to both
   PTP and NTP generically use the terms 'master' and 'slave'.

Shpiner, et al.               Experimental                      [Page 4]

RFC 8039             Multipath Time Synchronization        December 2016

3.  Multiple Paths in IP Networks

3.1.  Load Balancing

   Traffic sent across IP networks is often load-balanced across
   multiple paths.  The load-balancing decisions are typically based on
   packet header fields: source and destination addresses, Layer 4
   ports, the Flow Label field in IPv6, etc.

   Three common load-balancing criteria are per-destination, per-flow,
   and per-packet.  The per-destination load balancers take a
   load-balancing decision based on the destination IP address.
   Per-flow load balancers use various fields in the packet header,
   e.g., IP addresses and Layer 4 ports, for the load-balancing
   decision.  Per-packet load balancers use flow-blind techniques such
   as round-robin without basing the choice on the packet content.

3.2.  Using Multiple Paths Concurrently

   To utilize the diverse paths that traverse per-destination
   load balancers or per-flow load balancers, the packet transmitter can
   vary the IP addresses in the packet header.  The analysis in [PARIS2]
   shows that a significant majority of the flows on the Internet
   traverse per-destination or per-flow load balancing.  It presents
   statistics that 72% of the flows traverse per-destination
   load balancing and 39% of the flows traverse per-flow load balancing,
   while only a negligible part of the flows traverse per-packet
   load balancing.  These statistics show that the vast majority of the
   traffic on the Internet is load-balanced based on packet header

   The approaches in this document are based on varying the source and
   destination IP addresses in the packet header.  Possible extensions
   have been considered that also vary the UDP ports.  However, some of
   the existing implementations of PTP and NTP use fixed UDP port values
   in both the source and destination UDP port fields and thus do not
   allow this approach.

3.3.  Two-Way Paths

   A key property of IP networks is that packets forwarded from A to B
   do not necessarily traverse the same path as packets from B to A.
   Thus, we define a two-way path for a master-slave connection as a
   pair of one-way paths: the first from master to slave and the second
   from slave to master.

Shpiner, et al.               Experimental                      [Page 5]

RFC 8039             Multipath Time Synchronization        December 2016

   If possible, a traffic engineering approach can be used to verify
   that time synchronization traffic is always forwarded through
   bidirectional two-way paths, i.e., that each two-way path uses the
   same route in the forward and reverse directions, thus allowing
   propagation time symmetry.  However, in the general case, two-way
   paths do not necessarily use the same path for the forward and
   reverse directions.

4.  Solution Overview

   The multipath time synchronization protocols we present here are
   comprised of two building blocks: one is the path configuration and
   identification, and the other is the algorithm used by the slave to
   combine the information received from the various paths.

4.1.  Path Configuration and Identification

   The master and slave clocks must be able to determine the path of
   transmitted protocol packets and to identify the path of incoming
   protocol packets.  A path is determined by a {master IP, slave IP}
   address pair.  The synchronization protocol message exchange is run
   independently through each path.

   Each IP address pair defines a two-way path and thus allows the
   clocks to bind a transmitted packet to a specific path or to identify
   the path of an incoming packet.

   If possible, the routing tables across the network should be
   configured with multiple traffic-engineered paths between the pair of
   clocks.  By carefully configuring the routers in such networks, it is
   possible to create diverse paths for each of the IP address pairs
   between two clocks in the network.  However, in public and provider
   networks, the load-balancing behavior is hidden from the end users.
   In this case, the actual number of paths may be less than the number
   of IP address pairs, since some of the address pairs may share common

4.2.  Combining

   Various methods can be used for combining the time information
   received from the different paths.  The output of the combining
   algorithm is the accurate time offset.  Combining methods are further
   discussed in Section 6.

Shpiner, et al.               Experimental                      [Page 6]

RFC 8039             Multipath Time Synchronization        December 2016

5.  Multipath Time Synchronization over IP Networks

5.1.  Overview

   This section presents two variants of MPPTP and MPNTP: single-ended
   multipath time synchronization and dual-ended multipath time
   synchronization.  In the first variant, the multipath approach is
   only implemented by the slave, and the master is not aware of its
   usage.  In the second variant, all clocks use multiple paths.

   The dual-ended variant provides higher path diversity by using
   multiple IP addresses at both ends, the master and slave, while the
   single-ended variant only uses multiple addresses at the slave.
   Consequently, the single-ended approach can interoperate with
   existing implementations that do not use multiple paths.  The
   dual-ended and single-ended approaches can coexist in the same
   network; each slave selects the connection(s) it wants to make with
   the available masters.  A dual-ended slave could switch to
   single-ended mode if it does not see any dual-ended masters
   available.  A single-ended slave could connect to a single IP address
   of a dual-ended master.

   Multipath time synchronization, in both variants, requires clocks to
   use multiple IP addresses.  Using multiple IP addresses introduces a
   trade-off.  A large number of IP addresses allows a large number of
   diverse paths, providing the advantages of slave diversity discussed
   in Section 1.  On the other hand, a large number of IP addresses is
   more costly, requires the network topology to be more redundant, and
   exacts extra management overhead.

   If possible, the set of IP addresses for each clock should be chosen
   in a way that enables the establishment of paths that are the most
   different.  If the load-balancing rules in the network are known, it
   is possible to choose the IP addresses in a way that enforces path
   diversity.  However, even if the load-balancing scheme is not known,
   a careful choice of the IP addresses can increase the probability of
   path diversity.  For example, choosing multiple addresses with
   different prefixes is likely to produce higher path diversity, as BGP
   routers are more likely to route these different prefixes through
   different routes.

   The use of Network Address Translation (NAT) may significantly reduce
   the effectiveness of multipath synchronization in some cases.  For
   example, if a master uses multiple IP addresses that are translated
   to a single IP address, the path diversity can be dramatically
   reduced compared to a network that does not use NAT.  Thus, path

Shpiner, et al.               Experimental                      [Page 7]

RFC 8039             Multipath Time Synchronization        December 2016

   discovery should be used to identify the possible paths between the
   master and slave.  Path discovery is further discussed in
   Section 5.4.

   The concept of using multiple IP addresses or multiple interfaces is
   well established and is being used today by various applications and
   protocols, e.g., [MPTCP].  Using multiple interfaces introduces some
   challenges and issues, which were presented and discussed in [MIF].

   The descriptions in this section refer to the end-to-end scheme of
   PTP but are similarly applicable to the peer-to-peer scheme.  MPNTP,
   as described in this document, refers to the NTP client-server mode,
   although the concepts described here can be extended to include the
   symmetric variant as well.

   Multipath synchronization by nature requires protocol messages to be
   sent as unicast.  Specifically in PTP, the following messages must be
   sent as unicast in MPPTP: Sync, Delay_Req, Delay_Resp, PDelay_Req,
   PDelay_Resp, Follow_Up, and PDelay_Resp_Follow_Up.  Note that
   [IEEE1588] allows these messages to be sent either as multicast or as

5.2.  Single-Ended Multipath Synchronization

   In the single-ended approach, only the slave is aware of the fact
   that multiple paths are used, while the master is agnostic to the
   usage of multiple paths.  This approach allows a hybrid network,
   where some of the clocks are multipath clocks and others are
   conventional one-path clocks.  A single-ended multipath clock
   presents itself to the network as N independent clocks, using N IP
   addresses, as well as N clockIdentity [IEEE1588] values (in PTP).
   Thus, the usage of multiple slave identities by a slave clock is
   transparent from the master's point of view, such that it treats each
   of the identities as a separate slave clock.

5.2.1.  Single-Ended MPPTP Synchronization Message Exchange

   The single-ended MPPTP message exchange procedure is as follows.

   o  Each single-ended MPPTP clock has a fixed set of N IP addresses
      and N corresponding clockIdentities.  Each clock arbitrarily
      defines one of its IP addresses and clockIdentity values as the
      clock primary identity.

   o  A single-ended MPPTP port sends Announce messages only from its
      primary identity, according to the BMC algorithm.

Shpiner, et al.               Experimental                      [Page 8]

RFC 8039             Multipath Time Synchronization        December 2016

   o  The BMC algorithm at each clock determines the master, based on
      the received Announce messages.

   o  A single-ended MPPTP port that is in the 'slave' state uses
      unicast negotiation to request the master to transmit unicast
      messages to each of the N slave clockIdentity values.  The slave
      port periodically sends N Signaling messages to the master, using
      each of its N identities.  The Signaling message includes the

   o  The master periodically sends unicast Sync messages from its
      primary identity, identified by the sourcePortIdentity [IEEE1588]
      and IP address, to each of the slave identities.

   o  The slave, upon receiving a Sync message, identifies its path
      according to the destination IP address.  The slave sends a
      Delay_Req unicast message to the primary identity of the master.
      The Delay_Req is sent using the slave identity corresponding to
      the path through which the Sync was received.  Note that the rate
      of Delay_Req messages may be lower than the Sync message rate, and
      thus a Sync message is not necessarily followed by a Delay_Req.

   o  The master, in response to a Delay_Req message from the slave,
      responds with a Delay_Resp message using the IP address and
      sourcePortIdentity from the Delay_Req message.

   o  Upon receiving the Delay_Resp message, the slave identifies the
      path using the destination IP address and the
      requestingPortIdentity [IEEE1588].  The slave can then compute the
      corresponding path delay and the offset from the master.

   o  The slave combines the information from all negotiated paths.

5.2.2.  Single-Ended MPNTP Synchronization Message Exchange

   The single-ended MPNTP message exchange procedure is as follows.

   o  A single-ended MPNTP client has N separate identities, i.e., N IP
      addresses.  The assumption is that the server information,
      including its IP address, is known to the NTP clients.  This is a
      fair assumption, as typically the address(es) of the NTP server(s)
      is provided to the NTP client by configuration.

   o  A single-ended MPNTP client initiates NTP with an NTP server
      N times, using each of its N identities.

   o  NTP is maintained between the server and each of the N client

Shpiner, et al.               Experimental                      [Page 9]

RFC 8039             Multipath Time Synchronization        December 2016

   o  The client sends NTP messages to the master using each of its
      N identities.

   o  The server responds to the client's NTP messages using the IP
      address from the received NTP packet.

   o  The client, upon receiving an NTP packet, uses the IP destination
      address to identify the path through which it came, and it uses
      the time information accordingly.

   o  The client combines the information from all paths.

5.3.  Dual-Ended Multipath Synchronization

   In dual-ended multipath synchronization, each clock has N IP
   addresses.  Time synchronization messages are exchanged between some
   of the combinations of {master IP, slave IP} addresses, allowing
   multiple paths between the master and slave.  Note that the actual
   number of paths between the master and slave may be less than the
   number of chosen {master IP, slave IP} address pairs.

   Once the multiple two-way connections are established, a separate
   synchronization protocol exchange instance is run through each
   of them.

5.3.1.  Dual-Ended MPPTP Synchronization Message Exchange

   The dual-ended MPPTP message exchange procedure is as follows.

   o  Every clock has N IP addresses but uses a single clockIdentity.

   o  The BMC algorithm at each clock determines the master.  The master
      is identified by its clockIdentity, allowing other clocks to know
      the multiple IP addresses it uses.

   o  When a clock sends an Announce message, it sends it from each of
      its IP addresses with its clockIdentity.

   o  A dual-ended MPPTP port that is in the 'slave' state uses unicast
      negotiation to request the master to transmit unicast messages to
      some or all of its N_s IP addresses.  This negotiation is done
      individually between a slave IP address and the corresponding
      master IP address with which the slave desires a connection.  The
      slave port periodically sends Signaling messages to the master,
      using some or all of its N_s IP addresses as the source, to the
      corresponding master's N_m IP addresses.  The Signaling message

Shpiner, et al.               Experimental                     [Page 10]

RFC 8039             Multipath Time Synchronization        December 2016

      ('N_s' and 'N_m' indicate the number of IP addresses of the slave
      and master, respectively.)

   o  The master periodically sends unicast Sync messages from each of
      its IP addresses to the corresponding slave IP addresses for which
      a unicast connection was negotiated.

   o  The slave, upon receiving a Sync message, identifies its path
      according to the {source IP, destination IP} addresses.  The slave
      sends a Delay_Req unicast message, swapping the source and
      destination IP addresses from the Sync message.  Note that the
      rate of Delay_Req messages may be lower than the Sync message
      rate, and thus a Sync message is not necessarily followed by a

   o  The master, in response to a Delay_Req message from the slave,
      responds with a Delay_Resp message using the sourcePortIdentity
      from the Delay_Req message and swapping the IP addresses from the

   o  Upon receiving the Delay_Resp message, the slave identifies the
      path using the {source IP, destination IP} address pair.  The
      slave can then compute the corresponding path delay and the offset
      from the master.

   o  The slave combines the information from all negotiated paths.

5.3.2.  Dual-Ended MPNTP Synchronization Message Exchange

   The MPNTP message exchange procedure is as follows.

   o  Each NTP clock has a set of N IP addresses.  The assumption is
      that the server information, including its multiple IP addresses,
      is known to the NTP clients.

   o  The MPNTP client chooses N_svr server IP addresses and N_c client
      IP addresses and initiates the N_svr*N_c instances of the
      protocol, one for each {server IP, client IP} address pair,
      allowing the client to combine the information from the N_s*N_c

      ('N_svr' and 'N_c' indicate the number of IP addresses of the
      server and client, respectively, with which a client chooses to

   o  The client sends NTP messages to the master using each of the
      source-destination address combinations.

Shpiner, et al.               Experimental                     [Page 11]

RFC 8039             Multipath Time Synchronization        December 2016

   o  The server responds to the client's NTP messages using the IP
      address combination from the received NTP packet.

   o  Using the {source IP, destination IP} address pair in the received
      packets, the client identifies the path and performs its
      computations for each of the paths accordingly.

   o  The client combines the information from all paths.

5.4.  Using Traceroute for Path Discovery

   The approach described thus far uses multiple IP addresses in a
   single clock to create multiple paths.  However, although each
   two-way path is defined by a different {master IP, slave IP} address
   pair, some of the IP address pairs may traverse exactly the same
   network path, making them redundant.

   Traceroute-based path discovery can be used for filtering only the IP
   addresses that obtain diverse paths.  'Paris traceroute' [PARIS] and
   'TraceFlow' [TRACEFLOW] are examples of tools that discover the paths
   between two points in the network.  It should be noted that this
   filtering approach is effective only if the Traceroute implementation
   uses the same IP addresses and UDP ports as the synchronization
   protocol packets.  Since some Traceroute implementations vary the UDP
   ports, they may not be effective in identifying and filtering
   redundant paths in synchronization protocols.

   Traceroute-based filtering can be implemented by both master and
   slave nodes, or it can be restricted to run only on slave nodes to
   reduce the overhead on the master.  For networks that guarantee that
   the path of the timing packets in the forward and reverse directions
   are the same, path discovery should only be performed at the slave.

   Since network routes change over time, path discovery and redundant
   path filtering should be performed periodically.  Two {master IP,
   slave IP} address pairs that produce two diverse paths may be
   rerouted to use the same paths.  Thus, the set of addresses that are
   used by each clock should be reassessed regularly.

Shpiner, et al.               Experimental                     [Page 12]

RFC 8039             Multipath Time Synchronization        December 2016

5.5.  Using Unicast Discovery for MPPTP

   As presented above, MPPTP uses Announce messages and the BMC
   algorithm to discover the master.  The unicast discovery option of
   PTP can be used as an alternative.

   When using unicast discovery, the MPPTP slave ports maintain a list
   of the IP addresses of the master.  The slave port uses unicast
   negotiation to request unicast service from the master as follows:

   o  In single-ended MPPTP, the slave uses unicast negotiation from
      each of its identities to the master's (only) identity.

   o  In dual-ended MPPTP, the slave uses unicast negotiation from its
      IP addresses, each to a corresponding master IP address, to
      request unicast synchronization messages.

   Afterwards, the message exchange continues as described in
   Sections 5.2.1 and 5.3.1.

   The unicast discovery option can be used in networks that do not
   support multicast or in networks in which the master clocks are known
   in advance.  In particular, unicast discovery avoids multicasting
   Announce messages.

6.  Combining Algorithm

   Previous sections discussed the methods of creating the multiple
   paths and obtaining the time information required by the slave
   algorithm.  Once the time information is received through each of the
   paths, the slave should use a combining algorithm, which consolidates
   the information from the different paths into a single clock.
   Various methods have been suggested for combining information from
   different paths or from different clocks, e.g., [NTP] [SLAVEDIV]
   [HIGH-AVAI] [KALMAN].  The choice of the combining algorithm is local
   to the slave and does not affect interoperability.  Hence, this
   document does not define a specific method to be used by the slave.
   The combining algorithm should be chosen carefully based on the
   system properties, as different combining algorithms provide
   different advantages.  For example, some combining algorithms (e.g.,
   [NTP] [DELAY-ATT]) are intended to be robust in the face of security
   attacks, while other combining algorithms (e.g., [KALMAN]) are more
   resilient to random delay variation.

Shpiner, et al.               Experimental                     [Page 13]

RFC 8039             Multipath Time Synchronization        December 2016

7.  Security Considerations

   The security aspects of time synchronization protocols are discussed
   in detail in [RFC7384].  The methods described in this document
   propose to run a time synchronization protocol through redundant
   paths and thus allow the detection and mitigation of
   man-in-the-middle attacks, as described in [DELAY-ATT].
   Specifically, multipath synchronization can mitigate the following
   threats (as per [RFC7384]):

   o  Packet manipulation (Section 3.2.1 of [RFC7384]).

   o  Packet interception and removal (Section 3.2.5 of [RFC7384]).

   o  Packet delay manipulation (Section 3.2.6 of [RFC7384]).

   It should be noted that when using multiple paths, these paths may
   partially overlap, and thus an attack that takes place in a common
   segment of these paths is not mitigated by the redundancy.  Moreover,
   an on-path attacker may in some cases have access to more than one
   router or may be able to migrate from one router to another.
   Therefore, when using multiple paths, it is important for the paths
   to be as diverse and as independent as possible, making the
   redundancy scheme more tolerant to on-path attacks.

   It should be noted that the multipath approach requires the master
   (or NTP server) to dedicate more resources to each slave (client)
   than the conventional single-path approach.  Hence, well-known
   Distributed Denial-of-Service (DDoS) attacks may potentially be
   amplified when the multipath approach is enabled.

8.  Scope of the Experiment

   This memo is published as an Experimental RFC.  The purpose of the
   experimental period is to allow the community to analyze and to
   verify the methods defined in this document.  An experimental
   evaluation of some of these methods has been published in [MULTI].
   It is expected that the experimental period will allow the methods to
   be further investigated and verified by the community.  The duration
   of the experiment is expected to be no less than two years from the
   publication of this document.

Shpiner, et al.               Experimental                     [Page 14]

RFC 8039             Multipath Time Synchronization        December 2016

9.  References

9.1.  Normative References

   [IEEE1588] IEEE Instrumentation and Measurement Society, "IEEE
              Standard for a Precision Clock Synchronization Protocol
              for Networked Measurement and Control Systems", IEEE
              Std 1588-2008, DOI 10.1109/IEEESTD.2008.4579760.

   [NTP]      Mills, D., Martin, J., Ed., Burbank, J., and W. Kasch,
              "Network Time Protocol Version 4: Protocol and Algorithms
              Specification", RFC 5905, DOI 10.17487/RFC5905, June 2010,

9.2.  Informative References

              He, Y., Faloutsos, M., Krishnamurthy, S., and B. Huffaker,
              "On routing asymmetry in the Internet", IEEE GLOBECOM,
              DOI 10.1109/GLOCOM.2005.1577769, December 2005.

              Pathak, A., Pucha, H., Zhang, Y., Hu, C., and Z. Mao, "A
              measurement study of internet delay asymmetry",
              International Conference on Passive and Active Network
              Measurement 2008, DOI 10.1007/978-3-540-79232-1_19,
              April 2008.

              Mizrahi, T., "A Game Theoretic Analysis of Delay Attacks
              against Time Synchronization Protocols", IEEE
              International Symposium on Precision Clock Synchronization
              for Measurement, Control and Communication (ISPCS),
              DOI 10.1109/ISPCS.2012.6336612, September 2012.

              Ferrari, P., Flammini, A., Rinaldi, S., and G. Prytz,
              "High availability IEEE 1588 nodes over IEEE 802.1 aq
              Shortest Path Bridging networks", IEEE International
              Symposium on Precision Clock Synchronization for
              Measurement, Control and Communication (ISPCS),
              DOI 10.1109/ISPCS.2013.6644760, September 2013.

   [KALMAN]   Giorgi, G. and C. Narduzzi, "Kalman filtering for
              multi-path network synchronization", IEEE International
              Symposium on Precision Clock Synchronization for
              Measurement, Control and Communication (ISPCS),
              DOI 10.1109/ISPCS.2014.6948693, September 2014.

Shpiner, et al.               Experimental                     [Page 15]

RFC 8039             Multipath Time Synchronization        December 2016

   [MIF]      Blanchet, M. and P. Seite, "Multiple Interfaces and
              Provisioning Domains Problem Statement", RFC 6418,
              DOI 10.17487/RFC6418, November 2011,

   [MPTCP]    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,

   [MULTI]    Shpiner, A., Revah, Y., and T. Mizrahi, "Multi-path Time
              Protocols", IEEE International Symposium on Precision
              Clock Synchronization for Measurement, Control and
              Communication (ISPCS), DOI 10.1109/ISPCS.2013.6644754,
              September 2013.

   [PARIS]    Augustin, B., Friedman, T., and R. Teixeira, "Measuring
              Load-balanced Paths in the Internet", 7th ACM SIGCOMM
              conference on Internet measurement (IMC '07),
              DOI 10.1145/1298306.1298329, October 2007.

   [PARIS2]   Augustin, B., Friedman, T., and R. Teixeira, "Measuring
              Multipath Routing in the Internet", IEEE/ACM Transactions
              on Networking, 19(3), pp. 830-840,
              DOI 10.1109/TNET.2010.2096232, June 2011.

   [RFC7384]  Mizrahi, T., "Security Requirements of Time Protocols in
              Packet Switched Networks", RFC 7384, DOI 10.17487/RFC7384,
              October 2014, <http://www.rfc-editor.org/info/rfc7384>.

   [SLAVEDIV] Mizrahi, T., "Slave Diversity: Using Multiple Paths to
              Improve the Accuracy of Clock Synchronization Protocols",
              IEEE International Symposium on Precision Clock
              Synchronization for Measurement, Control and Communication
              (ISPCS), DOI 10.1109/ISPCS.2012.6336621, September 2012.

              Narasimhan, J., Venkataswami, B., Groves, R., and P.
              Hoose, "Traceflow", Work in Progress,
              draft-janapath-intarea-traceflow-00, January 2012.

Shpiner, et al.               Experimental                     [Page 16]

RFC 8039             Multipath Time Synchronization        December 2016


   The authors would like to thank Yoram Revah for his contribution to
   this work.  The authors also gratefully acknowledge the useful
   comments provided by Peter Meyer, Doug Arnold, Joe Abley, Zhen Cao,
   Watson Ladd, and Mirja Kuehlewind, as well as other comments received
   from the TICTOC working group participants.

Authors' Addresses

   Alex Shpiner
   Mellanox Technologies, Ltd.
   Hakidma 26
   Ofer Industrial Park
   Yokneam, 2069200

   Email: alexshp@mellanox.com

   Richard Tse
   8555 Baxter Place
   Burnaby, BC  V5A 4V7

   Email: Richard.Tse@microsemi.com

   Craig Schelp

   Email: craig.schelp@oracle.com

   Tal Mizrahi
   6 Hamada St.
   Yokneam, 2066721

   Email: talmi@marvell.com

Shpiner, et al.               Experimental                     [Page 17]