RFC 8889
This document is obsolete. Please refer to RFC 9342.

Internet Engineering Task Force (IETF)                  G. Fioccola, Ed.
Request for Comments: 8889                           Huawei Technologies
Category: Experimental                                       M. Cociglio
ISSN: 2070-1721                                           Telecom Italia
                                                                A. Sapio
                                                       Intel Corporation
                                                                R. Sisto
                                                   Politecnico di Torino
                                                             August 2020

Multipoint Alternate-Marking Method for Passive and Hybrid Performance


   The Alternate-Marking method, as presented in RFC 8321, can only be
   applied to point-to-point flows, because it assumes that all the
   packets of the flow measured on one node are measured again by a
   single second node.  This document generalizes and expands this
   methodology to measure any kind of unicast flow whose packets can
   follow several different paths in the network -- in wider terms, a
   multipoint-to-multipoint network.  For this reason, the technique
   here described is called "Multipoint Alternate Marking".

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 candidates 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

Copyright Notice

   Copyright (c) 2020 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
   (https://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.  Terminology
     2.1.  Correlation with RFC 5644
   3.  Flow Classification
   4.  Multipoint Performance Measurement
     4.1.  Monitoring Network
   5.  Multipoint Packet Loss
   6.  Network Clustering
     6.1.  Algorithm for Clusters Partition
   7.  Timing Aspects
   8.  Multipoint Delay and Delay Variation
     8.1.  Delay Measurements on a Multipoint-Paths Basis
       8.1.1.  Single-Marking Measurement
     8.2.  Delay Measurements on a Single-Packet Basis
       8.2.1.  Single- and Double-Marking Measurement
       8.2.2.  Hashing Selection Method
   9.  A Closed-Loop Performance-Management Approach
   10. Examples of Application
   11. Security Considerations
   12. IANA Considerations
   13. References
     13.1.  Normative References
     13.2.  Informative References

   Authors' Addresses

1.  Introduction

   The Alternate-Marking method, as described in RFC 8321 [RFC8321], is
   applicable to a point-to-point path.  The extension proposed in this
   document applies to the most general case of multipoint-to-multipoint
   path and enables flexible and adaptive performance measurements in a
   managed network.

   The Alternate-Marking methodology described in RFC 8321 [RFC8321]
   allows the synchronization of the measurements in different points by
   dividing the packet flow into batches.  So it is possible to get
   coherent counters and show what is happening in every marking period
   for each monitored flow.  The monitoring parameters are the packet
   counter and timestamps of a flow for each marking period.  Note that
   additional details about the applicability of the Alternate-Marking
   methodology are described in RFC 8321 [RFC8321] while implementation
   details can be found in the paper "AM-PM: Efficient Network Telemetry
   using Alternate Marking" [IEEE-Network-PNPM].

   There are some applications of the Alternate-Marking method where
   there are a lot of monitored flows and nodes.  Multipoint Alternate
   Marking aims to reduce these values and makes the performance
   monitoring more flexible in case a detailed analysis is not needed.
   For instance, by considering n measurement points and m monitored
   flows, the order of magnitude of the packet counters for each time
   interval is n*m*2 (1 per color).  The number of measurement points
   and monitored flows may vary and depends on the portion of the
   network we are monitoring (core network, metro network, access
   network) and the granularity (for each service, each customer).  So
   if both n and m are high values, the packet counters increase a lot,
   and Multipoint Alternate Marking offers a tool to control these

   The approach presented in this document is applied only to unicast
   flows and not to multicast.  Broadcast, Unknown Unicast, and
   Multicast (BUM) traffic is not considered here, because traffic
   replication is not covered by the Multipoint Alternate-Marking
   method.  Furthermore, it can be applicable to anycast flows, and
   Equal-Cost Multipath (ECMP) paths can also be easily monitored with
   this technique.

   In short, RFC 8321 [RFC8321] applies to point-to-point unicast flows
   and BUM traffic, while this document and its Clustered Alternate-
   Marking method is valid for multipoint-to-multipoint unicast flows,
   anycast, and ECMP flows.

   Therefore,the Alternate-Marking method can be extended to any kind of
   multipoint-to-multipoint paths, and the network-clustering approach
   presented in this document is the formalization of how to implement
   this property and allow a flexible and optimized performance
   measurement support for network management in every situation.

   Without network clustering, it is possible to apply Alternate Marking
   only for all the network or per single flow.  Instead, with network
   clustering, it is possible to use the partition of the network into
   clusters at different levels in order to perform the needed degree of
   detail.  In some circumstances, it is possible to monitor a
   multipoint network by analyzing the network clustering, without
   examining in depth.  In case of problems (packet loss is measured or
   the delay is too high), the filtering criteria could be specified
   more in order to perform a detailed analysis by using a different
   combination of clusters up to a per-flow measurement as described in
   RFC 8321 [RFC8321].

   This approach fits very well with the Closed-Loop Network and
   Software-Defined Network (SDN) paradigm, where the SDN orchestrator
   and the SDN controllers are the brains of the network and can manage
   flow control to the switches and routers and, in the same way, can
   calibrate the performance measurements depending on the desired
   accuracy.  An SDN controller application can orchestrate how
   accurately the network performance monitoring is set up by applying
   the Multipoint Alternate Marking as described in this document.

   It is important to underline that, as an extension of RFC 8321
   [RFC8321], this is a methodology document, so the mechanism that can
   be used to transmit the counters and the timestamps is out of scope
   here, and the implementation is open.  Several options are possible
   -- e.g., see "Enhanced Alternate Marking Method"

   Note that the fragmented packets case can be managed with the
   Alternate-Marking methodology only if fragmentation happens outside
   the portion of the network that is monitored.  This is always true
   for both RFC 8321 [RFC8321] and Multipoint Alternate Marking, as
   explained here.

2.  Terminology

   The definitions of the basic terms are identical to those found in
   Alternate Marking [RFC8321].  It is to be remembered that RFC 8321
   [RFC8321] is valid for point-to-point unicast flows and BUM traffic.

   The important new terms that need to be explained are listed below:

   Multipoint Alternate Marking:  Extension to RFC 8321 [RFC8321], valid
      for multipoint-to-multipoint unicast flows, anycast, and ECMP
      flows.  It can also be referred to as Clustered Alternate Marking.

   Flow definition:  The concept of flow is generalized in this
      document.  The identification fields are selected without any
      constraints and, in general, the flow can be a multipoint-to-
      multipoint flow, as a result of aggregate point-to-point flows.

   Monitoring network:  Identified with the nodes of the network that
      are the measurement points (MPs) and the links that are the
      connections between MPs.  The monitoring network graph depends on
      the flow definition, so it can represent a specific flow or the
      entire network topology as aggregate of all the flows.

   Cluster:  Smallest identifiable subnetwork of the entire monitoring
      network graph that still satisfies the condition that the number
      of packets that go in is the same as the number that go out.

   Multipoint metrics:  Packet loss, delay, and delay variation are
      extended to the case of multipoint flows.  It is possible to
      compute these metrics on the basis of multipoint paths in order to
      associate the measurements to a cluster, a combination of
      clusters, or the entire monitored network.  For delay and delay
      variation, it is also possible to define the metrics on a single-
      packet basis, and it means that the multipoint path is used to
      easily couple packets between input and output nodes of a
      multipoint path.

   The next section highlights the correlation with the terms used in
   RFC 5644 [RFC5644].

2.1.  Correlation with RFC 5644

   RFC 5644 [RFC5644] is limited to active measurements using a single
   source packet or stream.  Its scope is also limited to observations
   of corresponding packets along the path (spatial metric) and at one
   or more destinations (one-to-group) along the path.

   Instead, the scope of this memo is to define multiparty metrics for
   passive and hybrid measurements in a group-to-group topology with
   multiple sources and destinations.

   RFC 5644 [RFC5644] introduces metric names that can be reused here
   but have to be extended and rephrased to be applied to the Alternate-
   Marking schema:

   a.  the multiparty metrics are not only one-to-group metrics but can
       be also group-to-group metrics;

   b.  the spatial metrics, used for measuring the performance of
       segments of a source to destination path, are applied here to
       group-to-group segments (called clusters).

3.  Flow Classification

   A unicast flow is identified by all the packets having a set of
   common characteristics.  This definition is inspired by RFC 7011

   As an example, by considering a flow as all the packets sharing the
   same source IP address or the same destination IP address, it is easy
   to understand that the resulting pattern will not be a point-to-point
   connection, but a point-to-multipoint or multipoint-to-point

   In general, a flow can be defined by a set of selection rules used to
   match a subset of the packets processed by the network device.  These
   rules specify a set of Layer 3 and Layer 4 header fields
   (identification fields) and the relative values that must be found in
   matching packets.

   The choice of the identification fields directly affects the type of
   paths that the flow would follow in the network.  In fact, it is
   possible to relate a set of identification fields with the pattern of
   the resulting graphs, as listed in Figure 1.

   A TCP 5-tuple usually identifies flows following either a single path
   or a point-to-point multipath (in the case of load balancing).  On
   the contrary, a single source address selects aggregate flows
   following a point-to-multipoint, while a multipoint-to-point can be
   the result of a matching on a single destination address.  In the
   case where a selection rule and its reverse are used for
   bidirectional measurements, they can correspond to a point-to-
   multipoint in one direction and a multipoint-to-point in the opposite

   So the flows to be monitored are selected into the monitoring points
   using packet selection rules, which can also change the pattern of
   the monitored network.

   Note that, more generally, the flow can be defined at different
   levels based on the potential encapsulation, and additional
   conditions that are not in the packet header can also be included as
   part of matching criteria.

   The Alternate-Marking method is applicable only to a single path (and
   partially to a one-to-one multipath), so the extension proposed in
   this document is suitable also for the most general case of
   multipoint-to-multipoint, which embraces all the other patterns of
   Figure 1.

          point-to-point single path
              +------+      +------+      +------+
          ---<>  R1  <>----<>  R2  <>----<>  R3  <>---
              +------+      +------+      +------+

          point-to-point multipath
                          <>  R2  <>
                         / +------+ \
                        /            \
              +------+ /              \ +------+
          ---<>  R1  <>                <>  R4  <>---
              +------+ \              / +------+
                        \            /
                         \ +------+ /
                          <>  R3  <>

                                     <>  R4  <>---
                                    / +------+
                          +------+ /
                         <>  R2  <>
                        / +------+ \
              +------+ /            \ +------+
          ---<>  R1  <>              <>  R5  <>---
              +------+ \              +------+
                        \ +------+
                         <>  R3  <>
                          +------+ \
                                    \ +------+
                                     <>  R6  <>---

          ---<>  R1  <>
              +------+ \
                        \ +------+
                        <>  R4  <>
                        / +------+ \
              +------+ /            \ +------+
          ---<>  R2  <>              <>  R6  <>---
              +------+              / +------+
                          +------+ /
                         <>  R5  <>
                        / +------+
              +------+ /
          ---<>  R3  <>

              +------+                +------+
          ---<>  R1  <>              <>  R6  <>---
              +------+ \            / +------+
                        \ +------+ /
                         <>  R4  <>
                          +------+ \
              +------+              \ +------+
          ---<>  R2  <>             <>  R7  <>---
              +------+ \            / +------+
                        \ +------+ /
                         <>  R5  <>
                        / +------+ \
              +------+ /            \ +------+
          ---<>  R3  <>              <>  R8  <>---
              +------+                +------+

                       Figure 1: Flow Classification

   The case of unicast flow is considered in Figure 1.  The anycast flow
   is also in scope, because there is no replication and only a single
   node from the anycast group receives the traffic, so it can be viewed
   as a special case of unicast flow.  Furthermore, an ECMP flow is in
   scope by definition, since it is a point-to-multipoint unicast flow.

4.  Multipoint Performance Measurement

   By using the Alternate-Marking method, only point-to-point paths can
   be monitored.  To have an IP (TCP/UDP) flow that follows a point-to-
   point path, we have to define, with a specific value, 5
   identification fields (IP Source, IP Destination, Transport Protocol,
   Source Port, Destination Port).

   Multipoint Alternate Marking enables the performance measurement for
   multipoint flows selected by identification fields without any
   constraints (even the entire network production traffic).  It is also
   possible to use multiple marking points for the same monitored flow.

4.1.  Monitoring Network

   The monitoring network is deduced from the production network by
   identifying the nodes of the graph that are the measurement points,
   and the links that are the connections between measurement points.

   There are some techniques that can help with the building of the
   monitoring network (as an example, see [ROUTE-ASSESSMENT]).  In
   general, there are different options: the monitoring network can be
   obtained by considering all the possible paths for the traffic or
   periodically checking the traffic (e.g. daily, weekly, monthly) and
   updating the graph as appropriate, but this is up to the Network
   Management System (NMS) configuration.

   So a graph model of the monitoring network can be built according to
   the Alternate-Marking method: the monitored interfaces and links are
   identified.  Only the measurement points and links where the traffic
   has flowed have to be represented in the graph.

   Figure 2 shows a simple example of a monitoring network graph:

                                                   <>  R6  <>---
                                                  / +------+
                           +------+     +------+ /
                          <>  R2  <>---<>  R4  <>
                         / +------+ \   +------+ \
                        /            \            \ +------+
              +------+ /   +------+   \ +------+   <>  R7  <>---
          ---<>  R1  <>---<>  R3  <>---<>  R5  <>   +------+
              +------+ \   +------+ \   +------+ \
                        \            \            \ +------+
                         \            \            <>  R8  <>---
                          \            \            +------+
                           \            \
                            \            \ +------+
                             \            <>  R9  <>---
                              \            +------+
                                \ +------+
                                 <>  R10 <>---

                     Figure 2: Monitoring Network Graph

   Each monitoring point is characterized by the packet counter that
   refers only to a marking period of the monitored flow.

   The same is also applicable for the delay, but it will be described
   in the following sections.

5.  Multipoint Packet Loss

   Since all the packets of the considered flow leaving the network have
   previously entered the network, the number of packets counted by all
   the input nodes is always greater than, or equal to, the number of
   packets counted by all the output nodes.  Noninitial fragments are
   not considered here.

   The assumption is the use of the Alternate-Marking method.  In the
   case of no packet loss occurring in the marking period, if all the
   input and output points of the network domain to be monitored are
   measurement points, the sum of the number of packets on all the
   ingress interfaces equals the number on egress interfaces for the
   monitored flow.  In this circumstance, if no packet loss occurs, the
   intermediate measurement points only have the task of splitting the

   It is possible to define the Network Packet Loss of one monitored
   flow for a single period.  In a packet network, the number of lost
   packets is the number of packets counted by the input nodes minus the
   number of packets counted by the output nodes.  This is true for
   every packet flow in each marking period.

   The monitored network packet loss with n input nodes and m output
   nodes is given by:

   PL = (PI1 + PI2 +...+ PIn) - (PO1 + PO2 +...+ POm)


   PL is the network packet loss (number of lost packets)

   PIi is the number of packets flowed through the i-th input node in
   this period

   POj is the number of packets flowed through the j-th output node in
   this period

   The equation is applied on a per-time-interval basis and a per-flow

      The reference interval is the Alternate-Marking period, as defined
      in RFC 8321 [RFC8321].

      The flow definition is generalized here.  Indeed, as described
      before, a multipoint packet flow is considered, and the
      identification fields can be selected without any constraints.

6.  Network Clustering

   The previous equation can determine the number of packets lost
   globally in the monitored network, exploiting only the data provided
   by the counters in the input and output nodes.

   In addition, it is also possible to leverage the data provided by the
   other counters in the network to converge on the smallest
   identifiable subnetworks where the losses occur.  These subnetworks
   are named "clusters".

   A cluster graph is a subnetwork of the entire monitoring network
   graph that still satisfies the packet loss equation (introduced in
   the previous section), where PL in this case is the number of packets
   lost in the cluster.  As for the entire monitoring network graph, the
   cluster is defined on a per-flow basis.

   For this reason, a cluster should contain all the arcs emanating from
   its input nodes and all the arcs terminating at its output nodes.
   This ensures that we can count all the packets (and only those)
   exiting an input node again at the output node, whatever path they

   In a completely monitored unidirectional network (a network where
   every network interface is monitored), each network device
   corresponds to a cluster, and each physical link corresponds to two
   clusters (one for each device).

   Clusters can have different sizes depending on the flow-filtering
   criteria adopted.

   Moreover, sometimes clusters can be optionally simplified.  For
   example, when two monitored interfaces are divided by a single router
   (one is the input interface, the other is the output interface, and
   the router has only these two interfaces), instead of counting
   exactly twice, upon entering and leaving, it is possible to consider
   a single measurement point.  In this case, we do not care about the
   internal packet loss of the router.

   It is worth highlighting that it might also be convenient to define
   clusters based on the topological information so that they are
   applicable to all the possible flows in the monitored network.

6.1.  Algorithm for Clusters Partition

   A simple algorithm can be applied in order to split our monitoring
   network into clusters.  This can be done for each direction
   separately.  The clusters partition is based on the monitoring
   network graph, which can be valid for a specific flow or can also be
   general and valid for the entire network topology.

   It is a two-step algorithm:

   1.  Group the links where there is the same starting node;

   2.  Join the grouped links with at least one ending node in common.

   Considering that the links are unidirectional, the first step implies
   listing all the links as connections between two nodes and grouping
   the different links if they have the same starting node.  Note that
   it is possible to start from any link, and the procedure will work.
   Following this classification, the second step implies eventually
   joining the groups classified in the first step by looking at the
   ending nodes.  If different groups have at least one common ending
   node, they are put together and belong to the same set.  After the
   application of the two steps of the algorithm, each one of the
   composed sets of links, together with the endpoint nodes, constitutes
   a cluster.

   In our monitoring network graph example, it is possible to identify
   the clusters partition by applying this two-step algorithm.

   The first step identifies the following groups:

   1.  Group 1: (R1-R2), (R1-R3), (R1-R10)

   2.  Group 2: (R2-R4), (R2-R5)

   3.  Group 3: (R3-R5), (R3-R9)

   4.  Group 4: (R4-R6), (R4-R7)

   5.  Group 5: (R5-R8)

   And then, the second step builds the clusters partition (in
   particular, we can underline that Groups 2 and 3 connect together,
   since R5 is in common):

   1.  Cluster 1: (R1-R2), (R1-R3), (R1-R10)

   2.  Cluster 2: (R2-R4), (R2-R5), (R3-R5), (R3-R9)

   3.  Cluster 3: (R4-R6), (R4-R7)

   4.  Cluster 4: (R5-R8)

   The flow direction here considered is from left to right.  For the
   opposite direction, the same reasoning can be applied, and in this
   example, you get the same clusters partition.

   In the end, the following 4 clusters are obtained:

          Cluster 1
                          <>  R2  <>---
                         / +------+
              +------+ /   +------+
          ---<>  R1  <>---<>  R3  <>---
              +------+ \   +------+
                                \ +------+
                                 <>  R10 <>---

          Cluster 2
              +------+     +------+
          ---<>  R2  <>---<>  R4  <>---
              +------+ \   +------+
              +------+   \ +------+
          ---<>  R3  <>---<>  R5  <>---
              +------+ \   +------+
                            \ +------+
                             <>  R9  <>---

          Cluster 3
                         <>  R6  <>---
                        / +------+
              +------+ /
          ---<>  R4  <>
              +------+ \
                        \ +------+
                         <>  R7  <>---

          Cluster 4
          ---<>  R5  <>
              +------+ \
                        \ +------+
                         <>  R8  <>---

                         Figure 3: Clusters Example

   There are clusters with more than two nodes as well as two-node
   clusters.  In the two-node clusters, the loss is on the link (Cluster
   4).  In more-than-two-node clusters, the loss is on the cluster, but
   we cannot know in which link (Cluster 1, 2, or 3).

   In this way, the calculation of packet loss can be made on a cluster
   basis.  Note that the packet counters for each marking period permit
   calculating the packet rate on a cluster basis, so Committed
   Information Rate (CIR) and Excess Information Rate (EIR) could also
   be deduced on a cluster basis.

   Obviously, by combining some clusters in a new connected subnetwork
   (called a "super cluster"), the packet-loss rule is still true.

   In this way, in a very large network, there is no need to configure
   detailed filter criteria to inspect the traffic.  You can check a
   multipoint network and, in case of problems, go deep with a step-by-
   step cluster analysis, but only for the cluster or combination of
   clusters where the problem happens.

   In summary, once a flow is defined, the algorithm to build the
   clusters partition is based on topological information; therefore, it
   considers all the possible links and nodes crossed by the given flow,
   even if there is no traffic.  So, if the flow does not enter or
   traverse all the nodes, the counters have a nonzero value for the
   involved nodes and a zero value for the other nodes without traffic;
   but in the end, all the formulas are still valid.

   The algorithm described above is an iterative clustering algorithm,
   but it is also possible to apply a recursive clustering algorithm by
   using the node-node adjacency matrix representation

   The complete and mathematical analysis of the possible algorithms for
   clusters partition, including the considerations in terms of
   efficiency and a comparison between the different methods, is in the
   paper [IEEE-ACM-ToN-MPNPM].

7.  Timing Aspects

   It is important to consider the timing aspects, since out-of-order
   packets happen and have to be handled as well, as described in RFC
   8321 [RFC8321].  However, in a multisource situation, an additional
   issue has to be considered.  With multipoint path, the egress nodes
   will receive alternate marked packets in random order from different
   ingress nodes, and this must not affect the measurement.

   So, if we analyze a multipoint-to-multipoint path with more than one
   marking node, it is important to recognize the reference measurement
   interval.  In general, the measurement interval for describing the
   results is the interval of the marking node that is more aligned with
   the start of the measurement, as reported in Figure 4.

   Note that the mark switching approach based on a fixed timer is
   considered in this document.

           time -> start         stop
           T(R1)   |-------------|
           T(R2)     |-------------|
           T(R3)        |------------|

                       Figure 4: Measurement Interval

   In Figure 4, it is assumed that the node with the earliest clock (R1)
   identifies the right starting and ending times of the measurement,
   but it is just an assumption, and other possibilities could occur.
   So, in this case, T(R1) is the measurement interval, and its
   recognition is essential in order to make comparisons with other
   active/passive/hybrid Packet Loss metrics.

   When we expand to multipoint-to-multipoint flows, we have to consider
   that all source nodes mark the traffic, and this adds more

   Regarding the timing aspects of the methodology, RFC 8321 [RFC8321]
   already describes two contributions that are taken into account: the
   clock error between network devices and the network delay between
   measurement points.

   But we should now consider an additional contribution.  Since all
   source nodes mark the traffic, the source measurement intervals can
   be of different lengths and with different offsets, and this mismatch
   m can be added to d, as shown in Figure 5.

                |                   L                    |
                |         L/2                L/2         |
                |<=><===>|                      |<===><=>|
                  m   d  |                      |  d   m
                       available counting interval

               Figure 5: Timing Aspects for Multipoint Paths

   So the misalignment between the marking source routers gives an
   additional constraint, and the value of m is added to d (which
   already includes clock error and network delay).

   Thus, three different possible contributions are considered: clock
   error between network devices, network delay between measurement
   points, and the misalignment between the marking source routers.

   In the end, the condition that must be satisfied to enable the method
   to function properly is that the available counting interval must be
   > 0, and that means:

   L - 2m - 2d > 0.

   This formula needs to be verified for each measurement point on the
   multipoint path, where m is misalignment between the marking source
   routers, while d, already introduced in RFC 8321 [RFC8321], takes
   into account clock error and network delay between network nodes.
   Therefore, the mismatch between measurement intervals must satisfy
   this condition.

   Note that the timing considerations are valid for both packet loss
   and delay measurements.

8.  Multipoint Delay and Delay Variation

   The same line of reasoning can be applied to delay and delay
   variation.  Similarly to the delay measurements defined in RFC 8321
   [RFC8321], the marking batches anchor the samples to a particular
   period, and this is the time reference that can be used.  It is
   important to highlight that both delay and delay-variation
   measurements make sense in a multipoint path.  The delay variation is
   calculated by considering the same packets selected for measuring the

   In general, it is possible to perform delay and delay-variation
   measurements on the basis of multipoint paths or single packets:

   *  Delay measurements on the basis of multipoint paths mean that the
      delay value is representative of an entire multipoint path (e.g.,
      the whole multipoint network, a cluster, or a combination of

   *  Delay measurements on a single-packet basis mean that you can use
      a multipoint path just to easily couple packets between input and
      output nodes of a multipoint path, as described in the following

8.1.  Delay Measurements on a Multipoint-Paths Basis

8.1.1.  Single-Marking Measurement

   Mean delay and mean delay-variation measurements can also be
   generalized to the case of multipoint flows.  It is possible to
   compute the average one-way delay of packets in one block, a cluster,
   or the entire monitored network.

   The average latency can be measured as the difference between the
   weighted averages of the mean timestamps of the sets of output and
   input nodes.  This means that, in the calculation, it is possible to
   weigh the timestamps by considering the number of packets for each

8.2.  Delay Measurements on a Single-Packet Basis

8.2.1.  Single- and Double-Marking Measurement

   Delay and delay-variation measurements relative to only one picked
   packet per period (both single and double marked) can be performed in
   the multipoint scenario, with some limitations:

      Single marking based on the first/last packet of the interval
      would not work, because it would not be possible to agree on the
      first packet of the interval.

      Double marking or multiplexed marking would work, but each
      measurement would only give information about the delay of a
      single path.  However, by repeating the measurement multiple
      times, it is possible to get information about all the paths in
      the multipoint flow.  This can be done in the case of a point-to-
      multipoint path, but it is more difficult to achieve in the case
      of a multipoint-to-multipoint path because of the multiple source

   If we would perform a delay measurement for more than one picked
   packet in the same marking period, and especially if we want to get
   delay measurements on a multipoint-to-multipoint basis, neither the
   single- nor the double-marking method is useful in the multipoint
   scenario, since they would not be representative of the entire flow.
   The packets can follow different paths with various delays, and in
   general it can be very difficult to recognize marked packets in a
   multipoint-to-multipoint path, especially in the case when there is
   more than one per period.

   A desirable option is to monitor simultaneously all the paths of a
   multipoint path in the same marking period; for this purpose, hashing
   can be used, as reported in the next section.

8.2.2.  Hashing Selection Method

   RFCs 5474 [RFC5474] and 5475 [RFC5475] introduce sampling and
   filtering techniques for IP packet selection.

   The hash-based selection methodologies for delay measurement can work
   in a multipoint-to-multipoint path and can be used either coupled to
   mean delay or stand-alone.

   [ALTERNATE-MARKING] introduces how to use the hash method (RFCs 5474
   [RFC5474] and 5475 [RFC5475]) combined with the Alternate-Marking
   method for point-to-point flows.  It is also called Mixed Hashed
   Marking: the coupling of a marking method and hashing technique is
   very useful, because the marking batches anchor the samples selected
   with hashing, and this simplifies the correlation of the hashing
   packets along the path.

   It is possible to use a basic-hash or a dynamic-hash method.  One of
   the challenges of the basic approach is that the frequency of the
   sampled packets may vary considerably.  For this reason, the dynamic
   approach has been introduced for point-to-point flows in order to
   have the desired and almost fixed number of samples for each
   measurement period.  Using the hash-based sampling, the number of
   samples may vary a lot because it depends on the packet rate that is
   variable.  The dynamic approach helps to have an almost fixed number
   of samples for each marking period, and this is a better option for
   making regular measurements over time.  In the hash-based sampling,
   Alternate Marking is used to create periods, so that hash-based
   samples are divided into batches, which allows anchoring the selected
   samples to their period.  Moreover, in the dynamic hash-based
   sampling, by dynamically adapting the length of the hash value, the
   number of samples is bounded in each marking period.  This can be
   realized by choosing the maximum number of samples (NMAX) to be
   caught in a marking period.  The algorithm starts with only a few
   hash bits, which permits selecting a greater percentage of packets
   (e.g., with 0 bits of hash all the packets are sampled, with 1 bit of
   hash half of the packets are sampled, and so on).  When the number of
   selected packets reaches NMAX, a hashing bit is added.  As a
   consequence, the sampling proceeds at half of the original rate, and
   also the packets already selected that do not match the new hash are
   discarded.  This step can be repeated iteratively.  It is assumed
   that each sample includes the timestamp (used for delay measurement)
   and the hash value, allowing the management system to match the
   samples received from the two measurement points.  The dynamic
   process statistically converges at the end of a marking period, and
   the final number of selected samples is between NMAX/2 and NMAX.
   Therefore, the dynamic approach paces the sampling rate, allowing to
   bound the number of sampled packets per sampling period.

   In a multipoint environment, the behavior is similar to a point-to-
   point flow.  In particular, in the context of a multipoint-to-
   multipoint flow, the dynamic hash could be the solution for
   performing delay measurements on specific packets and overcoming the
   single- and double-marking limitations.

   The management system receives the samples, including the timestamps
   and the hash value, from all the MPs, and this happens for both
   point-to-point and multipoint-to-multipoint flows.  Then, the longest
   hash used by the MPs is deduced and applied to couple timestamps from
   either the same packets of 2 MPs of a point-to-point path, or the
   input and output MPs of a cluster (or a super cluster or the entire
   network).  But some considerations are needed: if there isn't packet
   loss, the set of input samples is always equal to the set of output
   samples.  In the case of packet loss, the set of output samples can
   be a subset of input samples, but the method still works because, at
   the end, it is easy to couple the input and output timestamps of each
   caught packet using the hash (in particular, the "unused part of the
   hash" that should be different for each packet).

   Therefore, the basic hash is logically similar to the double-marking
   method, and in the case of a point-to-point path, double-marking and
   basic-hash selection are equivalent.  The dynamic approach scales the
   number of measurements per interval.  It would seem that double
   marking would also work well if we reduced the interval length, but
   this can be done only for a point-to-point path and not for a
   multipoint path, where we cannot couple the picked packets in a
   multipoint path.  So, in general, if we want to get delay
   measurements on the basis of a multipoint-to-multipoint path, and
   want to select more than one packet per period, double marking cannot
   be used because we could not be able to couple the picked packets
   between input and output nodes.  On the other hand, we can do that by
   using hashing selection.

9.  A Closed-Loop Performance-Management Approach

   The Multipoint Alternate-Marking framework that is introduced in this
   document adds flexibility to Performance Management (PM), because it
   can reduce the order of magnitude of the packet counters.  This
   allows an SDN orchestrator to supervise, control, and manage PM in
   large networks.

   The monitoring network can be considered as a whole or split into
   clusters that are the smallest subnetworks (group-to-group segments),
   maintaining the packet-loss property for each subnetwork.  The
   clusters can also be combined in new, connected subnetworks at
   different levels, depending on the detail we want to achieve.

   An SDN controller or a Network Management System (NMS) can calibrate
   performance measurements, since they are aware of the network
   topology.  They can start without examining in depth.  In case of
   necessity (packet loss is measured or the delay is too high), the
   filtering criteria could be immediately reconfigured in order to
   perform a partition of the network by using clusters and/or different
   combinations of clusters.  In this way, the problem can be localized
   in a specific cluster or a single combination of clusters, and a more
   detailed analysis can be performed step by step by successive
   approximation up to a point-to-point flow detailed analysis.  This is
   the so-called "closed loop".

   This approach can be called "network zooming" and can be performed in
   two different ways:

   1) change the traffic filter and select more detailed flows;

   2) activate new measurement points by defining more specified

   The network-zooming approach implies that some filters or rules are
   changed and that therefore there is a transient time to wait once the
   new network configuration takes effect.  This time can be determined
   by the Network Orchestrator/Controller, based on the network

   For example, if the network zooming identifies the performance
   problem for the traffic coming from a specific source, we need to
   recognize the marked signal from this specific source node and its
   relative path.  For this purpose, we can activate all the available
   measurement points and better specify the flow filter criteria (i.e.,
   5-tuple).  As an alternative, it can be enough to select packets from
   the specific source for delay measurements; in this case, it is
   possible to apply the hashing technique, as mentioned in the previous

   [IFIT-FRAMEWORK] defines an architecture where the centralized Data
   Collector and Network Management can apply the intelligent and
   flexible Alternate-Marking algorithm as previously described.

   As for RFC 8321 [RFC8321], it is possible to classify the traffic and
   mark a portion of the total traffic.  For each period, the packet
   rate and bandwidth are calculated from the number of packets.  In
   this way, the network orchestrator becomes aware if the traffic rate
   surpasses limits.  In addition, more precision can be obtained by
   reducing the marking period; indeed, some implementations use a
   marking period of 1 sec or less.

   In addition, an SDN controller could also collect the measurement

   It is important to mention that the Multipoint Alternate Marking
   framework also helps Traffic Visualization.  Indeed, this methodology
   is very useful for identifying which path or cluster is crossed by
   the flow.

10.  Examples of Application

   There are application fields where it may be useful to take into
   consideration the Multipoint Alternate Marking:

   VPN:  The IP traffic is selected on an IP-source basis in both
      directions.  At the endpoint WAN interface, all the output traffic
      is counted in a single flow.  The input traffic is composed of all
      the other flows aggregated for source address.  So, by considering
      n endpoints, the monitored flows are n (each flow with 1 ingress
      point and (n-1) egress points) instead of n*(n-1) flows (each
      flow, with 1 ingress point and 1 egress point).

   Mobile Backhaul:  LTE traffic is selected, in the Up direction, by
      the EnodeB source address and, in the Down direction, by the
      EnodeB destination address, because the packets are sent from the
      Mobile Packet Core to the EnodeB.  So the monitored flow is only
      one per EnodeB in both directions.

   Over The Top (OTT) services:  The traffic is selected, in the Down
      direction, by the source addresses of the packets sent by OTT
      servers.  In the opposite direction (Up), it is selected by the
      destination IP addresses of the same servers.  So the monitoring
      is based on a single flow per OTT server in both directions.

   Enterprise SD-WAN:  SD-WAN allows connecting remote branch offices to
      data centers and building higher-performance WANs.  A centralized
      controller is used to set policies and prioritize traffic.  The
      SD-WAN takes into account these policies and the availability of
      network bandwidth to route traffic.  This helps ensure that
      application performance meets Service Level Agreements (SLAs).
      This methodology can also help the path selection for the WAN
      connection based on per-cluster and per-flow performance.

   Note that the preceding list is just an example and is not
   exhaustive.  More applications are possible.

11.  Security Considerations

   This document specifies a method of performing measurements that does
   not directly affect Internet security or applications that run on the
   Internet.  However, implementation of this method must be mindful of
   security and privacy concerns, as explained in RFC 8321 [RFC8321].

12.  IANA Considerations

   This document has no IANA actions.

13.  References

13.1.  Normative References

   [RFC5474]  Duffield, N., Ed., Chiou, D., Claise, B., Greenberg, A.,
              Grossglauser, M., and J. Rexford, "A Framework for Packet
              Selection and Reporting", RFC 5474, DOI 10.17487/RFC5474,
              March 2009, <https://www.rfc-editor.org/info/rfc5474>.

   [RFC5475]  Zseby, T., Molina, M., Duffield, N., Niccolini, S., and F.
              Raspall, "Sampling and Filtering Techniques for IP Packet
              Selection", RFC 5475, DOI 10.17487/RFC5475, March 2009,

   [RFC5644]  Stephan, E., Liang, L., and A. Morton, "IP Performance
              Metrics (IPPM): Spatial and Multicast", RFC 5644,
              DOI 10.17487/RFC5644, October 2009,

   [RFC8321]  Fioccola, G., Ed., Capello, A., Cociglio, M., Castaldelli,
              L., Chen, M., Zheng, L., Mirsky, G., and T. Mizrahi,
              "Alternate-Marking Method for Passive and Hybrid
              Performance Monitoring", RFC 8321, DOI 10.17487/RFC8321,
              January 2018, <https://www.rfc-editor.org/info/rfc8321>.

13.2.  Informative References

              Mizrahi, T., Arad, C., Fioccola, G., Cociglio, M., Chen,
              M., Zheng, L., and G. Mirsky, "Compact Alternate Marking
              Methods for Passive and Hybrid Performance Monitoring",
              Work in Progress, Internet-Draft, draft-mizrahi-ippm-
              compact-alternate-marking-05, 6 July 2019,

              Zhou, T., Fioccola, G., Lee, S., Cociglio, M., and W. Li,
              "Enhanced Alternate Marking Method", Work in Progress,
              Internet-Draft, draft-zhou-ippm-enhanced-alternate-
              marking-05, 13 July 2020, <https://tools.ietf.org/html/

              Cociglio, M., Fioccola, G., Marchetto, G., Sapio, A., and
              R. Sisto, "Multipoint Passive Monitoring in Packet
              Networks", IEEE/ACM Transactions on Networking vol. 27,
              no. 6, pp. 2377-2390, DOI 10.1109/TNET.2019.2950157,
              December 2019,

              Mizrahi, T., Navon, G., Fioccola, G., Cociglio, M., Chen,
              M., and G. Mirsky, "AM-PM: Efficient Network Telemetry
              using Alternate Marking", IEEE Network vol. 33, no. 4,
              pp. 155-161, DOI 10.1109/MNET.2019.1800152, July 2019,

              Song, H., Qin, F., Chen, H., Jin, J., and J. Shin, "In-
              situ Flow Information Telemetry", Work in Progress,
              Internet-Draft, draft-song-opsawg-ifit-framework-12, 14
              April 2020, <https://tools.ietf.org/html/draft-song-

   [RFC7011]  Claise, B., Ed., Trammell, B., Ed., and P. Aitken,
              "Specification of the IP Flow Information Export (IPFIX)
              Protocol for the Exchange of Flow Information", STD 77,
              RFC 7011, DOI 10.17487/RFC7011, September 2013,

              Alvarez-Hamelin, J., Morton, A., Fabini, J., Pignataro,
              C., and R. Geib, "Advanced Unidirectional Route Assessment
              (AURA)", Work in Progress, Internet-Draft, draft-ietf-
              ippm-route-10, 13 August 2020,


   The authors would like to thank Al Morton, Tal Mizrahi, and Rachel
   Huang for the precious contributions.

Authors' Addresses

   Giuseppe Fioccola (editor)
   Huawei Technologies
   Riesstrasse, 25
   80992 Munich

   Email: giuseppe.fioccola@huawei.com

   Mauro Cociglio
   Telecom Italia
   Via Reiss Romoli, 274
   10148 Torino

   Email: mauro.cociglio@telecomitalia.it

   Amedeo Sapio
   Intel Corporation
   4750 Patrick Henry Dr.
   Santa Clara, CA 95054
   United States of America

   Email: amedeo.sapio@intel.com

   Riccardo Sisto
   Politecnico di Torino
   Corso Duca degli Abruzzi, 24
   10129 Torino

   Email: riccardo.sisto@polito.it