WO2017186302A1

WO2017186302A1 - Performance measurement on a multipoint packet flow

Info

Publication number: WO2017186302A1
Application number: PCT/EP2016/059610
Authority: WO
Inventors: Mauro Cociglio
Original assignee: Telecom Italia S.P.A.
Priority date: 2016-04-29
Filing date: 2016-04-29
Publication date: 2017-11-02
Also published as: JP6740371B2; US20200067806A1; EP3449602A1; CN109196823B; KR20190002557A; US11218393B2; JP2019515551A; KR102565915B1; CN109196823A; EP3449602B1

Abstract

It is disclosed a method for performing a performance measurement on a multipoint packet flow transmitted in a packet-switched subnetwork. A monitoring network of measurement points is implemented in the subnetwork. All packets have a marking value, which is periodically switched between a first and second marking values. Each measurement point receiving a marked packet updates a couple of performance parameters, one per each marking value. Since not all the measurement points receive the same packets of the multipoint packet flow, in order to provide a performance measurement a cluster of measurement points is firstly identified in the monitoring network, which possesses the following property: each packet received at the input of the cluster is also received at the output of the cluster, if no packet loss occurs. The performance parameters provided by input and output measurement points of the identified cluster are then used for providing the performance measurements.

Description

PERFORMANCE MEASUREMENT

ON A MULTIPOINT PACKET FLOW

Technical field

The present invention relates to the field of communication networks. In particular, the present invention relates to a method for performing a performance measurement on a multipoint packet flow transmitted in a packet-switched communication network. Further, the present invention relates to nodes and computers for communication network configured to implement such a method, and to computer networks comprising such nodes and computers.

Background art

In a packet-switched communication network, packet flows are transmitted from source nodes to destination nodes through possible intermediate nodes. Exemplary packet-switched networks are IP (Internet Protocol) networks, Ethernet networks and MPLS (Multi- Protocol Label Switching) networks.

Packets not always reach their destination nodes, i.e. they may be lost during transmission through the network. Packet loss is due to different reasons. For instance, a node or link may fail, or packets may be discarded by a node due to a congestion of its ports. Besides, packets may be discarded by a node since they contain bit errors.

Moreover, each packet is transmitted at a transmission time by the source node and is received at a reception time by the destination node (if it is not lost). The time elapsing between transmission time and reception time is typically called "one-way delay". The one-way delay of a packet mainly depends on the number of possible intermediate nodes crossed by the packet from source to destination, the permanence time of the packet at each node and the propagation time along the links. Furthermore, packets may have different one-way delays. The difference between the one-way delays of two packets of a same packet flow is termed "interarrival jitter" (or, briefly, "jitter").

When a communication service (in particular, a real-time voice or data service such as call, conference call, video conference, etc.) is provided by means of a packet-switched network, a performance measurement in terms of packet loss, one-way delay and jitter on packet flows carrying the service provides an indication of the quality of service (QoS) perceived by the end users of the service. In addition, packet loss and high delay/jitter may require retransmission and then reduce the efficiency of the communication network. Therefore, measuring packet loss, one-way delay and/or jitter of packet flows in a communication network is of particular interest for network operators.

The ITU-T Recommendation Y.1731 (08/2015), Chapter 8 (pages 28-38) discloses OAM (Operation, Administration and Maintenance) functions for performance monitoring including measurement of frame loss ratio, frame delay and frame delay variation, which provides for injecting OAM frames in the packet flow to be measured. Besides, the known RTP (Real-time Trasport Protocol) and MPEG (Moving Picture Experts Group) protocol make use of sequence numbers inserted into packets in order to perform performance measurements.

WO 2010/072251 (in the name of the same Applicant) discloses a method for measuring packet loss of a packet flow transmitted from a transmitting node to a receiving node. Before transmitting each packet, the transmitting node marks it with either a first marking value or a second marking value. The marking value is switched with a marking period of e.g. 5 minutes, thereby dividing the packet flow in alternating blocks. Both the transmitting node and the receiving node implement a couple of counters, one for counting packets marked with the first marking value and the other for counting packets marked with the second marking value. By comparing the transmission-side and reception-side counters relating to packets marked by a same marking value, the packet loss during each marking period may be measured.

WO 201 1/079857 (in the name of the same Applicant) discloses a method for performing a time measurement on a packet flow which provides - in addition to the marking and counters described by WO 2010/072251 - to generate a transmission timestamp and a reception timestamp for one or more predefined sample packets of each marking period. Such timestamps are used for calculating one-way delay and/or jitter exhibited by the sample packets in each marking period.

WO 2013/174417 (in the name of the same Applicant) discloses a method for performing a time measurement on a packet flow which provides for calculating an average delay and/or an average jitter for each marking period based on cumulative transmission and reception timestamps generated for each marking period and on the counted number of packets transmitted and received during the same marking period.

Summary of the invention

The Applicant has noticed that the techniques provided by ITU-T Recommendation Y.1731 (08/2015) or by the known RTP and MPEG protocols allow monitoring the performance of a point-to-point packet flow, namely a packet flow made of packets having a number of header fields (also termed "identification fields" herein after) with the same values. This way, all the packets of the point-to-point packet flow are treated the same way by the nodes and accordingly follow a same path from a same source node to a same destination node, so that the measured packet flow is the same (namely, it comprises the same packets) at each measurement point implemented along the path. For instance, in TCP/IP networks all the packets of a point-to- point packet flow have the same value in the following identification fields: Source Address field, Destination Address field, Protocol field, Source Port field, Destination Port field and DSCP field.

However, in some cases, it may be of interest monitoring the performance of a so-called "multipoint packet flow". In the present description and in the claims, the expression "multipoint packet flow" will designate a packet flow comprising packets which are transmitted along two, or more, at least partially non overlapping end- to-end paths, so that different packets of the multipoint packet flow may be received at different measurement points implemented along those paths.

For instance, a multipoint packet flow may comprise two or more point-to-point packet flows with different source nodes and/or different destination nodes. Alternatively or in addition, a multipoint packet flow may comprise a point-to-point packet flow whose packets, e.g. due to a load balancing algorithm, are transmitted from a same source node to a same destination node via different intermediate nodes.

In order to measure the performance of a multipoint packet flow, the Applicant has noticed that the above described known techniques provided by ITU-T Recommendation Y.1731 (08/2015) or by the known RTP and MPEG protocols are not efficient from the computational point of view. Indeed, in order to provide a cumulative performance measurement of the multipoint packet flow, each point- to-point packet flow comprised in the multipoint packet flow shall be separately measured. In case of a multipoint packet flow comprising several point-to-point packet flows, this disadvantageously results in a high number of performance parameters (e.g. counters and/or timestamps) to be generated or updated, gathered and then merged to provide a cumulative performance measurement of the multipoint packet flow as a whole.

In view of the above, the Applicant has tackled the problem of providing a method for performing a performance measurement in a packet-switched communication network, which overcomes the aforesaid drawbacks.

In particular, the Applicant has tackled the problem of providing a method for performing a performance measurement in a packet- switched communication network, which allows to measure the performance of a multipoint packet flow in a more computationally efficient way.

In the following description and in the claims, the expression "performing a performance measurement on a packet flow" will designate an operation of measuring:

- a one-way delay, a two-way delay or a jitter induced on packets of said packet flow by transmission between the two measurement points; and/or

- a packet loss induced on packets of said packet flow by transmission between the two measurement points.

According to embodiments of the present invention, the above problem is solved by a method for performing a performance measurement on a multipoint packet flow which makes use of the marking technique described by WO 2010/072251 .

In particular, according to embodiments of the present invention, each packet of the multipoint packet flow is marked by either a first marking value V0 or a second marking value V1 , before being transmitted by the respective source node. The marking value is periodically switched between V0 and V1 with a certain marking period Tb so that, during each marking period, only packets of the multipoint packet flow with a certain marking value V0 or V1 are transmitted from the source node(s) to the destination node(s). Further, a monitoring network of measurement points is implemented in the subnetwork supporting transmission of the multipoint packet flow. Each measurement point provides a respective couple of performance parameters (e.g. counters), one relating to packets marked by V0 and the other relating to packets marked by V1 .

In order to provide a performance measurement on the multipoint packet flow PF using the couples of performance parameters provided by the measurement points of the monitoring network, at least one cluster of measurement points is preferably identified in the monitoring network, a cluster being defined as a set of measurement points of the monitoring network which exhibits the property that the ensemble of the packets received at the input measurement point(s) of the cluster is the same as the ensemble of the packets received at the output measurement point(s) of the cluster, if no packet loss occurs. In other words, if no packet loss occurs, each packet received by anyone of the cluster input measurement points is also received at one of the cluster output measurement points.

This way, for each identified cluster, the ensemble of the couples of performance parameters provided by the cluster input measurement point(s) is advantageously commensurable with the ensemble of the couples of performance parameters provided by the cluster output measurement point(s).

Hence, at the end of each marking period, the performance parameters relating to packets transmitted during the last marking period (and therefore having a same marking value) and provided by all the input and output measurement points of the cluster may be combined to provide a performance measurement of the multipoint packet flow during the last marking period.

Advantageously, such performance measurement of the multipoint packet flow is obtained without relying on separate performance measurements of the point-to-point packet flows forming the multipoint packet flow, which would require generating/updating, gathering and merging the performance parameters of each single point-to-point packet flow. The performance measurement method is accordingly very efficient from the computational point of view.

According to a first aspect, the present invention provides a method for performing a performance measurement on a multipoint packet flow transmitted in a subnetwork of a packet-switched communication network, the method comprising:

a) implementing in the subnetwork a monitoring network comprising a plurality of measurement points;

b) before transmitting each packet of the multipoint packet flow, marking the packet by setting a marking field thereof to a marking value equal to either a first marking value or a second marking value, the marking comprising periodically switching between the first marking value and the second marking value;

c) at each measurement point of the monitoring network, receiving at least one of the marked packets and, if the at least one of the marked packets is marked by the first marking value, updating a performance parameter;

d) providing the performance measurement on the multipoint packet flow, said providing comprising:

- identifying in the monitoring network a cluster of measurement points, the cluster of measurement points being a set of measurement points of the monitoring network selected so that, if no packet loss occurs, each packet received by at least one input measurement point of the cluster is also received at at least one output measurement point of the cluster; and

- providing the performance measurement on the multipoint packet flow using performance parameters provided by the at least one input measurement point and the at least one output measurement point of the cluster.

Preferably, identifying the cluster comprises:

- identifying in the monitoring network each couple of measurement points connected by a respective virtual link;

- grouping couples of measurement points having a same originating measurement point; and

- further grouping couples and/or groups of couples having at least one terminating measurement point in common, the further grouping providing the cluster.

Preferably, step d) also comprises:

- identifying at least two adjacent clusters and grouping the at least two clusters in a further cluster; and

- providing the performance measurement on the multipoint packet flow using performance parameters provided by at least one input measurement point and at least one output measurement point of the further cluster.

Preferably, the further cluster is the monitoring network as a whole.

According to some embodiments, at step b) the switching between the first marking value and the second marking value is performed for all the packets of the multipoint packet flow with a maximum mismatch of half a marking period Tb.

Preferably, step c) comprises identifying the at least one marked packet by checking whether at least one identification field of the at least one marked packet has a predefined value, the at least one identification field comprising a source address field and/or a destination address field.

Preferably, step c) performed by each measurement point comprises updating a counter indicative of a number of packets marked by the first marking value received at the measurement point during a marking period. Preferably, step c) performed by each measurement point also comprises updating an average timestamp indicative of an average time at which packets marked by the first marking value are received at the measurement point during a marking period.

Preferably, at step d) said providing the performance measurement comprises calculating a packet loss for the cluster as a difference between:

- a summation of counters provided by the at least one input measurement point of the cluster at the end of the marking period; and

- a summation of counters provided by the at least one output measurement point of the cluster at the end of the marking period.

Preferably, at step d) said providing the performance measurement comprises calculating an average one-way delay for the cluster as a difference between an average output timestamp and an average input timestamp, wherein:

- the average output timestamp is a weighted summation of average timestamps provided by the at least one output measurement point at the end of the marking period divided by the summation of counters provided by the at least one output measurement point, the weight for the average timestamp provided by each output measurement point being the counter provided by the same output measurement point; and

- the average input timestamp is a weighted summation of average timestamps provided by the at least one input measurement point at the end of the marking period divided by the summation of counters provided by the at least one input measurement point, the weight for the average timestamp provided by each input measurement point being the counter provided by the same input measurement point. Preferably, at step d) said providing said performance measurement comprises calculating a jitter based on: - timestannp provided by the at least one input measurement point for a first packet received during the marking period and a timestamp provided by the at least one input measurement point for a last packet received during the marking period;

- timestamp provided by the at least one output measurement point for a first packet received during the marking period and a timestamp provided by the at least one output measurement point for a last packet received during the marking period; and

number of packets received as a whole by the at least one input measurement point and number of packets received as a whole by the at least one output measurement point.

Preferably, at step d) said providing the performance measurement comprises calculating an average round-trip delay for the cluster by:

- calculating a first average one-way delay for the cluster in a first direction and a second average one-way delay for the cluster in a second direction opposite to the first direction; and

- calculating the average round-trip delay as a sum of the first average one-way delay and the second average one-way delay.

Preferably, step d) comprises providing at least one statistical performance measurement of the multipoint packet flow along an end-to-end path of the monitoring network by:

- identifying K cascaded clusters of the monitoring network which are crossed by the end-to-end path, K being equal to or higher than 2; and

- providing the at least one statistical performance measurement using performance measurements provided for said K cascaded clusters.

Preferably, providing the statistical performance measurement comprises providing a statistical packet loss measurement for the end-to-end path by: - calculating a cumulative packet loss probability of the K cascaded clusters as a function of packet losses probabilities of the K cascaded clusters; and

- calculating said statistical packet loss measurement for the end-to- end path as a function of said cumulative packet loss probability and a number of packets received at a measurement point terminating the end-to-end path.

Preferably, providing said statistical performance measurement comprises providing a statistical average one-way delay measurement for the end-to-end path by:

- calculating said statistical average one-way delay measurement for said end-to-end path as a summation of average one-way delays of said K cascaded clusters.

According to a second aspect, the present invention provides a packet-switched communication network comprising:

- a subnetwork configured to support transmission of a multipoint packet flow, each packet of the multipoint packet flow comprising a marking field set to a marking value equal to either a first marking value or a second marking value, the marking value being periodically switched between the first marking value and the second marking value;

- a monitoring network comprising a plurality of measurement points implemented in the subnetwork, each measurement point being configured to receive at least one of the marked packets and, if the at least one of the marked packets is marked by the first marking value, update a performance parameter; and

- a management server configured to provide a performance measurement on the multipoint packet flow, the management server being configured to:

- identify in the monitoring network a cluster of measurement points, the cluster of measurement points being a set of measurement points of the monitoring network selected so that, if no packet loss occurs, each packet received by at least one input measurement point of the cluster is also received at at least one output measurement point of the cluster; and

- provide the performance measurement on the multipoint packet flow using performance parameters gathered from the at least one input measurement point and the at least one output measurement point of the cluster.

According to a third aspect, the present invention provides a computer program product loadable in the memory of at least one computer and including software code portions for performing the steps of the method as set forth above, when the product is run on at least one computer.

Brief description of the drawings

The present invention will become clearer from the following detailed description, given by way of example and not of limitation, to be read with reference to the accompanying drawings, wherein:

- Figure 1 schematically shows an exemplary subnetwork supporting transmission of a multipoint packet flow;

- Figure 2 schematically show alternative subnetworks supporting transmission of other multipoint packet flows;

- Figure 3 shown the structure of the header of a packet formatted according to the known TCP over IPv4 protocol;

- Figure 4 shows a monitoring network of measurement points implemented in the subnetwork of Figure 1 ;

- Figure 5 is flow chart of the operation of each measurement point, according to an embodiment of the present invention;

- Figure 6 is a flow chart of a method for identifying a cluster of measurement points in a monitoring network of measurement points; - Figures 7a, 7b, 7c and 7d show the clusters of the monitoring network of Figure 4 identified by applying the method of Figure 6;

- Figure 8 shows the monitoring network of Figure 4 with exemplary values of counters and average timestamps;

- Figure 9 shows the monitoring network of Figure 4 with exemplary values of counters and timestamps of the first and last received packet;

- Figures 10a and 10b show the monitoring network of Figure 4 with exemplary values of counters and average timestamps for the multipoint packet flow of Figure 1 and for a counter-propagating multipoint packet flow, respectively; and

- Figures 1 1 a and 1 1 b show the cluster of Figure 7c with exemplary values of counters and average timestamps for the multipoint packet flow of Figure 1 and for a counter-propagating multipoint packet flow, respectively.

Detailed description of preferred embodiments of the invention

Figure 1 schematically shows an exemplary subnetwork SN supporting transmission of a multipoint packet flow PF, in which the method for performing a performance measurement according to embodiments of the present invention is implemented. The subnetwork network SN may be part of an IP network or any other type of packet-switched network (e.g. MPLS or Ethernet).

The subnetwork SN comprises a plurality of nodes reciprocally interconnected by links according to any known topology. In particular, the subnetwork SN comprises nodes N1 , N2, N5, N7, N8, N9 which, by way of non limiting example, are reciprocally interconnected according to a partially meshed topology. In particular, node N1 is connected to N2, N3 and N9, node N2 is connected to N4 and N5, node N3 is connected to N4 and N8, node N4 is connected to N7. Each node N1 , N2, N5, N7, N8, N9 preferably comprises at least one input interface and at least one output interface. The input and output interfaces of the nodes N1 , N2, N5, N7, N8, N9 are reciprocally interconnected by physical links. By way of non limiting example, as shown in Figure 1 :

- N1 comprises an input interface 1(1 1 ), a first output interface 0(1 1 ) connected to an input interface 1(21 ) of N2, a second output interface 0(12) connected to an input interface 1(31 ) of N3 and a third output interface 0(13) connected to an input interface 1(91 ) of N9;

- N2 comprises a first output interface 0(21 ) connected to an input interface 1(51 ) of N5 and a second output interface 0(22) connected to an input interface 1(41 ) of N4;

- N3 comprises a first output interface 0(31 ) connected to an input interface l(42) of N4 and a second output interface 0(32) connected to an input interface 1(81 ) of N8;

- N4 comprises an output interface 0(41 ) connected to an input interface 1(71 ) of N7;

- N5 comprises two output interfaces 0(51 ), 0(52); and

- N7, N8 and N9 comprise respective output interfaces 0(71 ), 0(81 ) and 0(91 ).

The subnetwork SN supports transmission of a multipoint packet flow PF.

The multipoint packet flow PF may comprise K point-to-point packet flows originated by N=1 source node and addressed to M>1 destination nodes. This is the case of the exemplary scenario depicted in Figure 1 , where the packet flow PF comprises K=5 point- to-point packet flows PF1 , PF2, ... PF5 having N=1 source node N1 and M=4 destination nodes N5, N7, N8 and N9. In particular, PF1 and PF2 are addressed to N5, PF3 is addressed to N7, PF4 is addressed to N8 and PF5 is addressed to N9. Alternatively, the multipoint packet flow PF may have N>1 source nodes and M=1 destination node. This is the case of the exemplary scenario depicted in Figure 2a, where nodes N1 , N2, N3 are three source nodes and N7 is the unique destination node.

Alternatively, the multipoint packet flow PF may have N>1 source nodes and M>1 destination nodes. This is the case of the exemplary scenario depicted in Figure 2b, where nodes N1 , N2, N3 are three source nodes, while N6 and N7 are two destination nodes. In this case, the multipoint packet flow PF may have few source nodes and several destination nodes, namely N«M. This is the case e.g. of multipoint packet flows carrying traffic of an OTT internet service in the downstream direction, namely from few OTT servers to multiple end users, or those carrying LTE (Long Term Evolution) traffic in the downstream direction, namely from few packet gateways to several eNodeBs. Alternatively, the multipoint packet flow PF may have several source nodes and few destination nodes, namely N»M. This is the case e.g. of multipoint packet flows carrying traffic of an OTT internet service in the upstream direction, namely from multiple end users to few OTT servers, or those carrying LTE traffic in the upstream direction, namely from several eNodeBs to few packet gateways.

Alternatively, the multipoint packet flow PF may have N=1 source node and M=1 destination node, but follow different paths through the subnetwork SN. This is the case of the exemplary scenario depicted in Figure 2c, where N1 is the unique source node, N5 is the unique destination node.

By referring again to the exemplary scenario of Figure 1 , the multipoint packet flow PF as a whole is injected in the subnetwork SN via the input interface 1(1 1 ) of the source node N1 and is then split. In particular, by way of non limiting example, PF1 , PF2 and first portion PF3' of the packet flow PF3 are transmitted by N1 to N2, while another portion PF3" of the packet flow PF3 together with PF4 are transmitted from N1 to N3 and PF5 is transmitted to N9. At the node N2, PF1 , PF2 and the first portion PF3' of the packet flow PF3 are further split, namely PF1 and PF2 are transmitted to N5, while PF3' is transmitted to N4. At the node N3, PF3" and PF4 are further split, namely PF3" is transmitted to N4, while PF4 is transmitted to N8. At the node N4, the portions PF3' and PF3" are joined and forwarded to N7. At the node N5, the packet flow PF1 and PF2 are further split, namely PF1 is transmitted to the output interface 0(51 ) and PF2 is transmitted to the output interface 0(52).

Preferably, each packet of the multipoint packet flow PF comprises a header and a payload. The payload comprises user data. Preferably, the header comprises information for routing the packet. The header format depends on the protocol according to which the packets are formatted. By way of non limiting example, Figure 3 shows the header Hi of a packet formatted according to the known TCP (Transmission Control Protocol) over IPv4 (Internet Protocol Version 4). The header Hi comprises 40 bytes divided into 20 bytes for the IP header and 20 bytes for the TCP header.

Preferably, the multipoint packet flow PF is defined (and hence distinguishable amongst all the traffic transmitted in the communication network CN) by one or more values of one or more of the above header fields, which are termed herein after "identification fields". In particular, in case of IPv4 protocol, the identification fields may comprise one or more of: Source Address, Destination Address, Protocol and DSCP of the IP header and Source Port and Destination Port of the TCP header. By suitably selecting the identification fields and their values, different types of multipoint packet flows may be defined.

For instance, the multipoint packet flow PF of the scenario shown in Figure 1 may be defined either by a single value of the Source Address field (namely, the IP address of host devices directly or indirectly connected to the source node N1 ) or by a set of values of the Destination Address field (namely, the IP addresses of host devices directly or indirectly connected to the destination nodes N5, N7, N8 and N9).

In general, defining a multipoint packet flow by a single value of the Source Address field results in a multipoint packet flow having N=1 source node (as PF in Figure 1 ), defining a multipoint packet flow by a single value of the Destination Address field results in a multipoint packet flow having M=1 destination node (see Figure 2a), defining a multipoint packet flow by multiple values of the Source Address field and/or multiple values of the Destination Address field results in a multipoint packet flow with N>1 source nodes and M>1 destination nodes (see Figure 2b), while defining a multipoint packet flow by a single value of the Source Address field and a single value of the Destination Address field results is a multipoint packet flow having N=1 source node and M=1 destination node (see Figure 2c).

Preferably, a monitoring network MN comprising a plurality of measurement points is implemented in the subnetwork SN. Each measurement point may be either embedded within a respective node, or implemented as a stand-alone machine connected to a respective node.

The monitoring network MN may comprise a measurement point at each one of the N source nodes and a measurement point at each one of the M destination nodes. For instance, with reference to Figure 1 , a measurement point A is implemented at the source node N1 (in particular on its input interface 1(1 1 )) and measurement points F, G, H, I, J are implemented at the destination nodes N5, N7, N8, N9 (in particular, at their interfaces 0(51 ), 0(52), 0(71 ), 1(81 ) and 1(91 ), respectively).

The monitoring network MN may also comprise one or more measurement points implemented at the intermediate nodes of the subnetwork SN. For instance, with reference to the exemplary scenario of Figure 1 :

a measurement point B is implemented at the input interface 1(21 ) of N2, where PF1 , PF2 and PF3' are split;

a measurement point C is implemented at the input interface 1(31 ) of N3, where PF3" and PF4 are split;

a measurement point D is implemented at the input interface 1(51 ) of N5, where PF1 and PF2 are split; and

- a measurement point E is implemented at the output interface 0(41 ) of N4, where PF3' and PF3" are joined.

It may be appreciated that this arrangement of the measurement points is purely exemplary. The performance measurement method of the present invention is indeed advantageously capable of providing consistent results with any arrangement of the measurement points.

Figure 4 shows the monitoring network MN comprising the measurement points A, B, ... J, the measurement points being reciprocally interconnected by logical links.

Preferably, the monitoring network MN is also provided with a management server, which is not shown in Figure 4 for simplicity. The management server may be either a stand-alone server connected to any of the nodes. Alternatively, the management server may be implemented at any of the nodes. The management server preferably cooperates with the measurement points of the monitoring network MN for gathering therefrom performance parameters, as it will be described in detail herein after.

According to preferred embodiments of the present invention, the packets of the multipoint packet flow PF are marked before they are injected in the subnetwork SN via the N source nodes. The marking may be implemented at the N source nodes themselves. In this case, the packets are preferably marked before they are received by the measurement points implemented at the N source nodes. Alternatively, the packets of the multipoint packet flow PF may be marked upstream the N source nodes.

More particularly, each packet of the multipoint packet flow PF preferably includes a marking field MF comprising at least one bit, whose value is set to one of two alternative marking values VO, V1 . The marking field MF is preferably comprised in the packet header Hi. The marking field MF may be for instance a field to which the protocol according to which the packet is formatted has not assigned a specific function yet. Alternatively, the marking field MF may be comprised in a field having other uses. For instance, in case of IP packets (see Figure 3), the marking field MF may comprise one bit of the 8-bit TOS (Type of Service) field or the RSV bit of the Flags field, and its two alternative marking values V1 and VO may be 1 and 0, respectively.

The value of the marking field MF is periodically switched between V1 and VO with a period Tb, which will be termed herein after "marking period". The marking period Tb may be set by the network operator, according to the desired time measurement rate (as it will be described in detail herein after, the marking period Tb is also the measurement period). For instance, the marking period Tb may be equal to 5 minutes.

The marking of all the K point-to-point packet flows of the multipoint packet flow PF is substantially synchronized, namely the marking value is changed substantially at the same time (namely, with a maximum mismatch of Tb/2) for all the K point-to-point packet flows of the multipoint packet flow PF. This way, the packets of the multipoint packet flow PF which are transmitted during a certain marking period are substantially all marked with a same marking value V1 or VO. Each measurement point of the monitoring network MN is preferably configured to provide at least one couple of performance parameters, one relating to packets marked by V1 and the other one relating to packets marked by V0, as it will be described in detail herein after with reference to the flow chart of Figure 5.

During each marking period, each measurement point X (X=A, B, ... J) receives all the traffic (or a copy thereof) received at the node input or output interface at which it is implemented (step 500).

The measurement point X then filters all the incoming traffic, in order to identify the packets of the multipoint packet flow PF (step 501 ). In order to perform the filtering step 501 , the measurement point X preferably reads the value(s) of the identification field(s) comprised in the header Hi of each received packet and checks whether it is (they are) equal to that one or those ones which define the multipoint packet flow PF as described above. Assuming that the packet flow PF shown in the exemplary scenario of Figure 1 is defined by the value of the IP address of the source node N1 , at step 501 each measurement point X preferably checks whether the Source Address of each incoming packet is equal to such value. Then, each measurement point X preferably reads the marking value V1 or V0 comprised in the marking field MF of each identified packet (step 502).

Then, depending on whether the packet is marked by V1 or V0, each measurement point X updates the value of the relevant performance parameter (step 503). Since - during each marking period - only packets with a same marking value are transmitted (e.g. V1 ), iterations of step 503 during that marking period result in the performance parameter relating to packets with that marking value (e.g. V1 ) being updated and the performance parameter relating to packets marked with the other marking value (e.g. V0) being fixed to the value reached at the end of the preceding marking period. Hence, upon expiration of each marking period (step 504), each measurement point X preferably sends to the management server the value which the performance parameter updated during the expired marking period reached at the end thereof (step 505). The management server will then use the currently fixed performance parameters gathered by the measurement points is order to provide a performance measurement of the multipoint packet flow PF.

In order to provide a performance measurement on the multipoint packet flow PF, at least one cluster of measurement points is preferably identified in the monitoring network MN. A cluster is preferably defined as a set of measurement points of the monitoring network MN which exhibits the property that the ensemble of the packets received at the input measurement point(s) of the cluster is the same as the ensemble of the packets received at the output measurement point(s) of the cluster, if no packet loss occurs.

Several clusters of different sizes may be identified in the monitoring network MN.

According to a preferred embodiment, first clusters of the monitoring network MN are identified by applying the algorithm shown in Figure 6.

Firstly, preferably, all the couples of measurement points (X→Y) connected by a virtual link are identified (step 600). In the exemplary monitoring network MN of Figure 4, the following couples are then identified at step 600: (A→B), (A→C), (A→J), (B→D), (B→E), (C→E), (C→l), (D→F), (D→G), (E→H).

Then, preferably, all the couples having the same originating measurement point are grouped (step 601 ). Hence, in the exemplary monitoring network MN of Figure 4, the couples (A→B), (A→C), (A→J) are grouped, the couples (B→D), (B→E) are grouped, the couples (C→E), (C→l) are grouped and the couples (D→F), (D→G) are grouped. Then, preferably, all the groups of couples having at least one terminating measurement point in common are preferably further grouped (step 602). Hence, in the exemplary monitoring network MN of Figure 4, the group of couples (B→D), (B→E) and the group of couples (C→E), (C→l) are further grouped, having the terminating measurement point E in common. The groups resulting from step 602 are elementary or minimum-sized clusters of the monitoring network MN.

Application of the above algorithm to the exemplary monitoring network MN of Figure 4 therefore allows identifying four clusters:

- cluster C1 (Figure 7a) comprising A as input measurement point and B, C and J as output measurement points;

- cluster C2 (Figure 7b) comprising B and C as input measurement points and D, E and I as output measurement points;

- cluster C3 (Figure 7c) comprising D as input measurement point and F, G as output measurement points; and

- cluster C4 (Figure 7d) comprising E as input measurement point and H as output measurement point.

Even if all the above clusters only comprise input measurement points and output measurement points, execution of the algorithm of Figure 6 may lead to identification of a cluster comprising also intermediate measurement points.

Further, two or more adjacent minimum-sized clusters identified as described above may be grouped into further larger clusters, which still exhibit the above cluster property. If the monitoring network MN comprises a measurement point at each one of the N source nodes and a measurement point at each one of the M destination nodes (as depicted in Figure 1 , by way of non limiting example), the largest cluster is the monitoring network MN itself, which is provided by grouping all its minimum-sized clusters identified as described above. Since, for each cluster, the ensemble of the packets received at the input measurement point(s) is the same as the ensemble of the packets received at the output measurement point(s) (if no packet loss occurs), then the ensemble of the couples of performance parameters provided by the input measurement point(s) of each cluster is advantageously commensurable with the ensemble of the couples of performance parameters provided by the output measurement point(s) of the same cluster.

Hence, at the end of each marking period, the performance parameters relating to packets transmitted during the last marking period (and therefore having a same marking value) and provided by all the input and output measurement points of a cluster may be combined to provide a performance measurement of the multipoint packet flow PF during the last marking period in the cluster.

Herein after, only the monitoring network MN as a whole and its minimum-sized cluster (also termed simply "clusters") identified e.g. by the algorithm of Figure 6 will be considered. However, every consideration relating to the monitoring network MN as a whole and its clusters is straightforwardly applicable to any intermediate-sized cluster obtained by grouping two or more minimum-sized clusters. A first type of performance measurements which may be provided for the multipoint packet flow PF according to embodiments of the present invention are packet loss measurements, namely measurements of the number of packets of the multipoint packet flow PF lost during transmission.

In order to enable packet loss measurements, the couple of performance parameters implemented by each measurement point X (X=A, B, ... J) preferably is a couple of counters C_xo, C_Xi (X=A, B ... J), one for counting packets marked by V0 and one for counting packet marked by V1 . In particular, during step 503 of the flow chart of Figure 5, upon reception of a packet of the multipoint packet flow PF marked by V0, the measurement point X preferably increases the counter C_xo by 1 whereas, upon reception of a packet of the multipoint packet flow PF marked by V1 , the measurement point X preferably increases the counter C_Xi by 1 . Hence, during marking periods wherein packets are marked by VO, the counter C_Xi at each measurement point X has a fixed value indicative of the number of packets marked by V1 received at the measurement point X during the preceding marking period, while the counter C_xo is increased upon reception of each packet marked by VO. Similarly, during marking periods wherein packets are marked by V1 , the counter C_xo at each measurement point X has a fixed value indicative of the number of packets marked by VO received at the measurement point X during the preceding marking period, while the counter C_Xi is increased upon reception of each packet marked by V1 . This latter case is depicted in Figure 4, which shows exemplary values of the counters C_xo (x=A, B, ... J) as provided by the measurement points A, B, C, ... J, during a marking period wherein packets of the multipoint packet flow PF are marked by V1 .

Then, an overall packet loss PL₀v of the multipoint packet flow PF may be calculated for that marking period as: PI_ov= C₀vⁱⁿ - Cov°^ut [1 ] wherein:

(i) Cov'ⁿ is the summation of the N counters provided by the N input measurement points of the monitoring network MN as a whole at the end of that marking period, and

(ii) Cov^out is the summation of the M counters provided by the M output measurement points of the monitoring network MN as a whole at the end of that marking period.

In the exemplary monitoring network MN of Figure 4, C₀vⁱⁿ= C_A and Cov^out = C_F+ C_G+ C_H+ Ci+ Cj, where C_x (X=A, F, G, I, J) is equal to Cxo for marking periods wherein packets were marked by V0 and to Cxi marking periods wherein packets were marked by V1 .

It may be appreciated that application of equation [1 ] to the exemplary values of C_xo set forth in Figure 4 results in PL_Ov=0, meaning that no packet loss occurred during the last marking period. Besides, a resulting PL₀v different from 0 would be indicative of a packet loss occurring in the last marking period.

The overall packet loss measurement PL₀v provides an overall indication of the performance of the multipoint packet flow PF in the subnetwork SN as a whole. However, if a packet loss occurs during a marking period, the measurement of PL₀v does not provide any indication of the physical link(s) or node(s) at which the packet loss occurred.

According to embodiments of the present invention, a cluster packet loss is also calculated for each cluster of the monitoring network MN. In particular, the cluster packet loss PL_Ci for a cluster Ci (i being the cluster index) and for a marking period is preferably calculated as:

Pl_ci= Ciⁱⁿ - Ci^0ut [2] wherein:

(i) Ci'ⁿ is the summation of the counters provided by the input measurement points of cluster Ci at the end of that marking period and

(ii) Ci^out is the summation of the counters provided by the output measurement points of cluster Ci at the end of that marking period.

Hence, in the exemplary monitoring network MN of Figure 4, at the end of a marking period the following cluster packet losses are provided by applying equation [2]:

- PLci= Ci'ⁿ - Ci°^ut =CA-(CB+CC⁺CJ);

- Pl_c2= C₂ ⁱⁿ - C₂ ^out = (CB+CC)-(CD+CE+C|); - Pl_c3= C₃'ⁿ - C₃ ^out = CD-(CF+C_g); and

pi _ in out _

- rLc₄- U₄ - U₄ - UE- H,

where C_x (X=A, B, ... J) is equal to C_xo for marking periods wherein packets were marked by V0 and to C_Xi marking periods wherein packets were marked by V1 .

Cluster packet loss measurements allow locating more precisely a packet loss occurring in the subnetwork SN supporting transmission of the multipoint packet flow PF.

Assuming for instance that, at the end of a marking period, the values of the counters C_x provided by the various measurement points are those shown in Figures 7a, 7b, 7c and 7d, the overall packet loss PL₀v calculated as described above by applying equation [1] results in PL₀v=1 , indicating that a packet loss occurred in the subnetwork SN. In order to locate more precisely the packet loss within the subnetwork SN, the cluster packet losses are calculated applying equation [2] to each cluster C1 , C2, C3, C4, which provide PLci=PLc3=PLc4=0 (meaning that no packet loss occurred in the last marking period in clusters C1 , C3 and C4) and PL_C2=1 , meaning that the packet loss occurred in cluster C2.

If the cluster Ci with cluster packet loss PL_Q different from 0 comprises a single virtual link, the above cluster packet loss measurement allows identifying exactly the physical link or node of the subnetwork SN on which the packet loss occurred.

If, instead, the cluster Ci comprises more than one virtual link, it is not possible to determine exactly on which virtual link(s) of the cluster Ci the packet loss occurred. In this case, a packet loss probability PLPQ of the cluster Ci may be calculated as the cluster packet loss PLci divided by C ⁿ (namely, the summation of the counters provided by the input measurement points of cluster Ci), namely:

By applying equation [3] to clusters C1 , C2, C3, C4 using the exemplary counter values shown in Figures 7a, 7b, 7c and 7d, it follows that PLPci=PLPc3=PLPc4=0, while PLP_C2=1/( 00+200)= 0.00333. This means that a packet transmitted though cluster C2 has a probability of 0.00333 of being lost, namely a packet every 300 is lost.

Under the assumption that each virtual link of a cluster Ci has a packet loss probability equal to the packet loss probability PLPQ of the cluster Ci, a statistical link packet loss may be calculated for each virtual link (X→Y) of the cluster Ci, by distributing the packet loss PLci of the cluster Ci on the various virtual links proportionally to the number of packets received at the measurement point terminating each virtual link.

More particularly, according to embodiments of the present invention, the statistical link packet loss PL_(Y) for a virtual link originating at the measurement point X and terminating at the measurement point Y of cluster Ci is calculated as follows:

PL(Y) ⁼ C(X_→Y) - Cy [4] where C_Y is the counter provided by the measurement point Y terminating the virtual link and C(x_→Y) is the number of packets statistically received at the measurement point X and transmitted over the virtual link (X→Y). However, C_Y equals the number C_(X→Y) of packets transmitted over the virtual link (X→Y) minus the packets statistically lost over the virtual link (X→Y), i.e. C_(X→Y) multiplied by PLPQ (the packet loss probability for the cluster Ci), namely:

C_Y = C(x_→Y) - C(x_→Y) PLPQ = C(x_→Y) (1 - PLPQ).

Accordingly, the number of packets C_(X→Y) transmitted over the virtual link (X→Y) is equal to C_Y divided by (1 -PLP_Ci). Hence, equation [4] may be rewritten as follows: PL(Y) C(X_→Y) -CY = C_Y— C_Y —— [4]

^{{ ) { }} ( - PLPci) 1 - PLPci

By applying the above equation [4] to each one of the virtual links of cluster C2 using the exemplary counter values shown in Figures 7a, 7b, 7c and 7d, the following statistical link packet losses are obtained:

50 50

- PL₍ ^{D⁾)= ₁ L 50 = — 50 = 50.167-50 = 0.167;

(1 - 0.00333)

300

It may be appreciated that the summation of the statistical link packet losses of all the virtual links of the cluster C2 equals the cluster packet loss PL_C2-

Moreover, according to embodiments of the present invention, a statistical end-to-end packet loss PL_(W→z₎ may be calculated for each end-to-end path (W→Z) originating at a measurement point W implemented at one of the N source nodes and terminated at a measurement point Z implemented at one of the M destination nodes.

To this purpose, the cluster(s) crossed by each end-to-end path (W→Z) in the monitoring network MN are preferably identified. With reference to the exemplary monitoring network of Figure 4:

- path A→F : C1→C2→C3;

- path A→G : C1→C2→C3;

- path A→H: C1→C2→C4;

- path A→l: C1→C2; and - path A→J: C1 .

Then, a packet loss probability is preferably calculated for the cascade of clusters crossed by each end-to-end path. In particular, the packet loss probability of the cascade of K clusters is preferably calculated as:

PLP(K clusters) = 1-n(1 - PLP_Ck) [5] k=1

where PLP_Ck is the packet loss probability of the k^th cluster.

Hence, by applying equation [5] to the exemplary counter values shown in Figures 7a, 7b, 7c and 7d, the following end-to-end packet loss probabilities are provided:

- PLP(A_→F)= PLPci,C2,C3 = 1 -[(1 -0)(1 -0.00333)(1 -0)]=0.00333;

-

PLPci,C2,C3 = 1 -[(1 -0)(1 -0.00333)(1 -0)]=0.00333;

- PLP(A→H)⁼ PLPci,C2,C4 ⁼ 1 -[(1 -0)(1 -0.00333)(1 -0)]=0.00333;

- PLP_(A→i)= PLPci,c2 = 1 -[(1 -0)(1 -0.00333)]=0.00333;

- PLP_(A→J)= PLPci = 1 -[(1 -0)]=0.

Then, the statistical end-to-end packet loss PL_(W→z) is preferably calculated for the end-to-end path (W→Z) as:

PL_(W→z) = C(w→z) -Cz = ^Cp - C_z = C_z · p. p^→Z) . [^β] where C_z is the counter provided by the measurement point Z terminating the end-to-end path and C(w→z) is the number of packets statistically received at the measurement point W and transmitted over the end-to-end path (W→Z), which is equal to C_z divided by (1 - PLP(w→z))-

By applying equation [6] to each one of the end-to-end paths of the exemplary monitoring network MN of Figure 4 using the exemplary counter values shown in Figures 7a, 7b, 7c and 7d, the following statistical end-to-end packet losses are obtained: 40

- PL (A→F)^■ 40 = 40.134-40 = 0.134:

(1 - -0.00333)

10

— 10 = 10.033 - 10 = 0.033

( -0.00333)

199

- PL_(A→H) = . . - 199 = 199.666 - 199 = 0.666;

( -0.00333)

50

- PL(A→i) = — 50 = 50.167-50 = 0.167; and

^{{ }} (1 -0.00333)

It shall be noticed that the summation of the statistical end-to-end packet losses for the various end-to-end links calculated as described above equals the overall packet loss PL₀v=1■

Besides packet loss measurements, also time measurements may be advantageously performed on the multipoint packet flow PF.

In order to enable this type of performance measurements, each measurement point X (X=A, B, ... J) preferably implements not only a couple of counters C_xo, C_Xi (X=A, B ... J), but also a couple of average timestamps T_xo, Τχι (X=A, B ... J), one relating to packets marked by V0 and one relating to packets marked by V1 .

In particular, during step 503, upon reception of a packet of the multipoint packet flow PF marked by V0, the measurement point X preferably increases the counter C_xo by 1 , generates a timestamp indicating the time at which the packet has been received and uses this timestamp for updating the average timestamps T_xo, which is the summation of all the timestamps generated for packets marked by V0 already received in current marking period divided by C_xo. Similarly, upon reception of a packet of the multipoint packet flow PF marked by V1 , the measurement point X preferably increases the counter C_Xi by 1 , generates a timestamp indicating the time at which the packet has been received and uses this timestamp for updating the average timestamps T_x-i, which is the summation of all the timestamps generated for packets marked by V1 already received in current marking period divided by C_x-|.

Hence, during marking periods wherein packets are marked by V0, the counter C_Xi and the average timestamp T_X at each measurement point X have fixed values, while the counter C_xo and the average timestamp T_xo are updated upon reception of each packet marked by V0. Similarly, during marking periods wherein packets are marked by V1 , the counter C_xo and the average timestamp T_xo at each measurement point X have fixed values, while the counter C_Xi and the average timestamp T_X are updated upon reception of each packet marked by V1. This latter case is depicted in Figure 8, which shows exemplary values of the counters C_x and average timestamps T_x (x=A, B, ... J) as provided by the measurement points A, B, C, ... J in the exemplary monitoring network MN of Figure 4, at the end of a marking period. The unit of measurement for the average timestamps is milliseconds.

Under the assumption that all the measurement points have substantially synchronized clocks, a first performance measurement which may be done is an overall average one-way delay OWD₀v for the multipoint packet flow PF from the N source nodes to the M destination nodes in a certain marking period. The overall average one-way delay OWD₀v for a marking period is preferably calculated as OWDov= Tov^out - [7]

wherein:

(i) C_k and T_k (i=1 , ... M) are the counters and average timestamps provided by the M output measurement points of the monitoring network MN as a whole at the end of the last marking period; and

(ii) C_j and T_j (i=1 , ... N) are the counters and average timestamps provided by the N input measurement points of the monitoring network MN as a whole at the end of the last marking period.

By applying equation [7] to the exemplary counter and average timestamp values shown in Figure 8, the following overall average one-way delay OWDov is obtained:

OWDov= ^F^F C_AT_A

In case of packet loss PL_Ov≠0 during a marking period, the error ERR on the measurement of the overall average one-way delay OWDov using equation [7] during that marking period is equal to:

,_ ₊ (T_b/4)- PL_0V

ERR= ± ^{' b'} ' > ^uv [8]

Gov

where T_b is the marking period and Cov'ⁿ is the summation of the N counters provided by the N input measurement points of the monitoring network MN as a whole at the end of that marking period.

Besides, an average one-way delay may also be calculated for each cluster of the monitoring network MN.

In particular, the cluster average one-way delay OWD_Q of the multipoint packet flow PF for cluster Ci may be calculated for a marking period as: [9]

wherein:

(i) C_k and T_k are the counters and average timestamps provided by the output measurement points of the cluster Ci at the end of the last marking period; and

(ii) Cj and Tj are the counters and average timestamps provided by the input measurement points of the cluster Ci at the end of the last marking period.

By applying equation [9] to the exemplary counter and average timestamps values shown in Figure 8, the following cluster one-way delays are obtained:

OWDci = I ci - I ci

_ C_DT_D + C_ET_E + C,T, CBT_b + C_c

OWDc2= I C2 - I C2

C_D + C_E + C, + C_C

_ C_HT_H C_ET_E

OWDc4= I C4 - I C4

C_H C_E

The calculation of the cluster average one-way delays advantageously provides a statistical indication of how the overall average one-way delay OWDov is distributed in the various clusters of the monitoring network MN. For instance, with the exemplary values of Figure 8, the calculation of the cluster average one-way delays indicates that the overall average one-way delay OWDov of 75.15 ms is mainly accumulated in clusters C3 and C4.

It may be appreciated that, if a cluster Ci comprises a single virtual link (as cluster C4), the above cluster average one-way delay measurement basically is the average one-way delay accumulated by the packets of the multipoint packet flow PF on the corresponding physical link.

If, instead, the cluster Ci comprises more than one virtual link, it is not possible to determine exactly the average-one way delay accumulated by the packets on each virtual link of the cluster Ci. However, under the assumption that each virtual link of a cluster Ci has one-way delay equal to the cluster average one-way delay OWDQ, a statistical end-to-end average one-way delay OWD_(W→z) may be calculated for each end-to-end path (W→Z) originating at a measurement point W implemented at one of the N source nodes and terminated at a measurement point Z implemented at one of the M destination nodes, starting from the cluster average one-way delay OWDQ calculated as described above.

In particular, the cluster(s) crossed by each end-to-end path (W→Z) in the monitoring network MN are preferably identified. With reference to the exemplary monitoring network of Figure 4:

- path A→F : C1→C2→C3;

- path A→G : C1→C2→C3;

- path A→H: C1→C2→C4;

- path A→l: C1→C2; and

- path A→J: C1 .

Then, an end-to-end average one-way delay OWD₍w→z) is preferably calculated for each end-to-end path, as a summation of the cluster average one-way delays of the clusters crossed by each end-to-end path. With reference to the exemplary values of counters and average timestamps set forth in Figure 8, the following end-to- end average one-way delays are therefore calculated:

- OWD(_A→F)= OWDc3+OWDc2+OWDci = 46+20+19.7=85.7;

- OWD_(A→G)= OWD_C3 ⁺OWDc2⁺OWDci = 46+20+19.7=85.7;

- OWD_(A→H)= OWD_C4+OWDc₂+OWDci = 50+20+19.7=89.7;

- OWD_(A→i)= OWDc2+OWDci = 20+19.7=35.7; and

Under the assumption of substantially synchronized measurement points, also average jitter measurements may be performed on the multipoint packet flow PF.

To this purpose, in addition to counters CX₀, CX-i , each measurement point also preferably stores a timestamp Tx(fj_rst) of the first packet received during a marking period and a timestamp Tx(i_ast) of the last packet received during the same marking period. Exemplary values of counters and timestamps T_X(f_iret), T_X(i_ast) in the 5 exemplary monitoring network MN of Figure 4 are set forth in Figure 9.

According to advantageous variants of the present invention, an overall average one-way jitter OWJov for a marking period is preferably calculated according to the following equation: i n M Π.1

I U L ' UJ

where:

(i) T₀ut(first) is the timestamp with lowest value amongst the timestamps T_X(f_iret) provided by the output measurement points of the monitoring network MN as a whole,

5 (ii) Tout(iast) is the timestamp with highest value amongst the timestamps T_X(i_ast) provided by the output measurement points of the monitoring network MN as a whole,

(iii) Thirst) is the timestamp with lowest value amongst the timestamps T_X(f_iret) provided by the input measurement points of 0 the monitoring network MN as a whole, and

(iv) Tin(iast) is the timestamp with highest value amongst the timestamps T_X(i_ast) provided by the input measurement points of the monitoring network MN as a whole.

Besides, as described above, C₀v'ⁿ and C₀v^out are the number of 5 packets received as a whole by the N input measurement points and by the M output measurement points, respectively, of the monitoring network MN as a whole during a same marking period.

Hence, by applying equation [10] to the exemplary values of counters and timestamps set forth in Figure 9,

T₀ut(iast)-TF(iast)-99, Tj_n(first)-TA(first)-0 and Tj_n(iast)= A(iast)^_1■ The overall average one-way jitter OWJov is then [(99-25)/329]-[(1 - 0)/329]=0.221 .

According to advantageous variants, a cluster average one-way jitter OWJc for each cluster may be calculated. The calculation is similar to the calculation of the cluster average one-way delay described above. Hence, a detailed description will not be repeated.

According to other embodiments of the present invention, an average round-trip delay measurement may be performed. This is particularly advantageous if the clocks of the various measurement points are not reciprocally synchronized.

To this purpose, besides monitoring the multipoint packet flow PF as described above, the measurement points of the monitoring network MN shall also monitor a counter-propagating packet flow PF', having M source nodes corresponding to the M destination nodes of the multipoint packet flow PF and N destination nodes corresponding to the N source nodes of the multipoint packet flow PF. For instance, with reference to the exemplary scenario shown in Figure 1 , the counter-propagating packet flow PF' preferably has N5, N7, N8 and N9 as source nodes and N1 as destination nodes.

It shall be noticed that, in general, the intermediate nodes crossed by the counter-propagating multipoint packet flow PF' may be different from those crossed by the multipoint packet flow PF, e.g. due to an asymmetric routing or due to the fact that, in at least one marking period, no traffic might be present at one or more of the nodes of the subnetwork SN. Herein after, however, for simplicity it is assumed that the intermediate nodes are the same for both PF and PF'. Under this assumption, the monitoring network MN and its clusters are the same for both PF and PF', although the measurement points acting as input measurement points for PF will act as output measurement points for PF'm, and vice versa. Hence, at the identification step 501 of the flow chart of Figure 5, each measurement point X (X=A, B, ... J) shall identify not only packets of the multipoint packet flow PF, but also packets of the counter-propagating multipoint packet flow PF'. For instance, since the multipoint packet flow PF has a single source node N1 , its identification may be done by checking whether the Source Address field of incoming packets comprises the IP address of a host device directly or indirectly connected to N1 (e.g. 192.23.45.67). On the other hand, since the counter-propagating multipoint packet flow PF' has a single destination node N1 , its identification may be done by checking whether the Destination Address field of incoming packets comprises the IP address of a host device connected to N1 (e.g. 192.23.45.67).

Then, preferably, at step 503 of the flow chart of Figure 5, each measurement point X=A, B, ... J preferably updates performance parameters relating also to the counter-propagating multipoint packet flow PF'.

In particular, in order to enable an average round-trip delay measurement, each measurement point preferably updates a counter C'x counting the packets of PF' received during a marking period and an average timestamp T'_x indicating the average time at which the packets of PF' are received during the marking period.

Figures 10a and 10b show exemplary values of counters Cx and average timestamps T_x generated by the measurement points X=A, B, ... J of the exemplary monitoring network MN for the multipoint packet flow PF (Figure 10a) and exemplary values of counters C'x and average timestamps T'_x generated by the measurement points X=A, B, ... J of the exemplary monitoring network MN for the counter-propagating multipoint packet flow PF (Figure 10b).

Preferably, an overall average round-trip delay RTD₀v is calculated as: RTDov = OWDov+ OVVD'ov [ 1 ] where OWD₀v is the overall average one-way delay of the packet flow PF and OWDOv is the overall average one-way delay of the packet flow PF'.

However, since the clocks of the various measurement points are not reciprocally synchronized, the overall average one-way delays may not be calculated using the above equation [7]. Reference is made e.g. to Figures 1 1 a and 1 1 b, which show exemplary values of counters and average timestamps generated by the measurement points D, F and G of the cluster C3 for PF (Figure 1 1 a) and for PF' (Figure 1 1 b), in an exemplary situation wherein the clock of G is delayed by 100 ms relative to the clocks of D and F, which are instead substantially synchronized. Such exemplary values clearly indicate that the one-way delay of the cluster C3 in both propagation directions is 1 ms. However, by applying equations [9] above for calculating the average one-way delays in cluster C3 for PF and PF', the results are:

OWDc₃= CpTp +^ C^ ^ + IOS ^ S^ ^^

C + C_Q C_Q 3 3

OWD'o3= go To CV T_{F +}C_G T_G 105.1 ₊ 5.2 _{= 32}

The average round-trip delay of cluster C3 calculated by summing up OWDc3 and OWD'_C3 calculated according to equation [9] would be RTDc3=OWDc3+OWD'c3=35.333. This result is however clearly wrong, since it should be equal to 2 ms (1 ms per each propagation direction).

Hence, preferably, instead of using OWD₀v and OWDOv in equation [7], further overall average one-way delays OWD_ov and OWD'_ov are preferably used in equation [1 1 ] for calculating the overall round-trip delay RTD₀v, which are calculated according to the following equations:

wherein:

(i) C_k and T_k (i=1 , ... M) are the counters and average timestamps provided by the M output measurement points of the monitoring network MN as a whole for PF;

(ii) Cj and Tj (i=1 , ... N) are the counters and average timestamps provided by the N input measurement points of the monitoring network MN as a whole for PF;

(iii) C'_k and T'_k (i=1 , ... M) are the counters and average timestamps provided by the M input measurement points of the monitoring network MN as a whole for PF';

(iv) C'j and T'j (i=1 , ... N) are the counters and average timestamps provided by the M output measurement points of the monitoring network MN as a whole for PF'.

The overall round-trip delay RTDov is finally calculated by applying equation [11] to the values of OWD_ov and OWD'_ov provided by equations [12a] and [12b].

Also average round-trip delay measurement for each single cluster may be performed. Preferably, the average round-trip delay RTD_Q for cluster Ci is calculated according to the following equation:

RTDQ = OWD_Ci + OWD'ci [13] where OWD_Ci is the average one-way delay of the packet flow PF in cluster Ci and OWD'_Ci is the average one-way delay of the packet flow PF' in cluster Ci, which are preferably calculated according to the following equations:

OWD_Ci = 1

[14a]

∑c_k ∑_Cj ∑e_k ∑σ,

k=1 j=1 k=1 j=1

wherein:

(i) C_k and T_k are the counters and average timestamps provided by the cluster output measurement points for PF;

(ii) C_j and T_j are the counters and average timestamps provided by the cluster input measurement points for PF;

(iii) C'_k and T'_k are the counters and average timestamps provided by the cluster input measurement points for PF';

(iv) Cj and T'j (i=1 , ... N) are the counters and average timestamps provided by the cluster output measurement points for

PF'.

It may be appreciated that application of equations [14a] and [14b] to the exemplary values of Figures 1 1 a and 1 1 b provide OWD_C3 =51 and OWD'_C3 =-49. The average round-trip delay RTD_C3 of cluster C3 is as per equation [13] is therefore RTD_C3= OWD_Ci + OWD'_Ci =2 ms, which is a correct result.

Claims

A method for performing a performance measurement on a multipoint packet flow (PF) transmitted in a subnetwork (SN) of a packet-switched communication network, said method comprising:

a) implementing in said subnetwork (SN) a monitoring network (MN) comprising a plurality of measurement points (A, B ...

J);

b) before transmitting each packet of said multipoint packet flow (PF), marking said packet by setting a marking field thereof to a marking value equal to either a first marking value or a second marking value, said marking comprising periodically switching between said first marking value and said second marking value;

c) at each measurement point (A, B, ... J) of said monitoring network (MN), receiving at least one of said marked packets and, if said at least one of said marked packets is marked by said first marking value, updating a performance parameter; d) providing said performance measurement on said multipoint packet flow (PF), said providing comprising:

identifying in said monitoring network (MN) a cluster of measurement points (MN, C1 , C2, C3, C4), said cluster of measurement points (MN, C1 , C2, C3, C4) being a set of measurement points of said monitoring network (MN) selected so that, if no packet loss occurs, each packet received by at least one input measurement point of said cluster is also received at at least one output measurement point of said cluster; and

providing said performance measurement on said multipoint packet flow (PF) using performance parameters provided by said at least one input measurement point and said at least one output measurement point of said cluster (C1 , C2, C3, C4).

The method according to claim 1 , wherein said identifying said cluster (C1 , C2, C3, C4) comprises:

(i) identifying in said monitoring network (MN) each couple of measurement points (X→Y) connected by a respective virtual link;

(ii) grouping couples of measurement points (X→Y) having a same originating measurement point; and

(iii) further grouping couples and/or group of couples having at least one terminating measurement point in common, said further grouping providing said cluster (C1 , C2, C3, C4).

The method according to claim 2, wherein step d) also comprises:

identifying at least two adjacent clusters (C1 , C2, C3, C4) and grouping said at least two clusters (C1 , C2, C3, C4) in a further cluster (MN); and

providing said performance measurement on said multipoint packet flow (PF) using performance parameters provided by at least one input measurement point and at least one output measurement point of said further cluster (MN).

The method according to claim 2 or 3, wherein said further cluster is said monitoring network (MN) as a whole.

The method according to any of the preceding claims, wherein at step b) said switching between said first marking value and said second marking value is performed for all the packets of said multipoint packet flow (PF) with a maximum mismatch of half a marking period Tb.

The method according to any of the preceding claims, wherein step c) comprises identifying said at least one marked packet by checking whether at least one identification field of said at least one marked packet has a predefined value, said at least one identification field comprising a source address field and/or a destination address field.

7. The method according to any of the preceding claims, wherein step c) performed by each measurement point (A, B, ... J) comprises updating a counter indicative of a number of packets marked by said first marking value received at said measurement point (A, B, ... J) during a marking period.

8. The method according to claim 7, wherein step c) performed by each measurement point (A, B, ... J) comprises updating an average timestamp indicative of an average time at which packets marked by said first marking value are received at said measurement point (A, B, ... J) during a marking period.

9. The method according to claim 6, wherein at step d) said providing said performance measurement comprises calculating a packet loss (PL_Ci) for said cluster (C1 , C2, C3, C4) as a difference between:

- a summation (Ci'ⁿ) of counters provided by said at least one input measurement point of said cluster (C1 , C2, C3, C4) at the end of said marking period; and

- a summation (C °^ut) of counters provided by said at least one output measurement point of said cluster (C1 , C2, C3, C4) at the end of said marking period.

10. The method according to claim 9, wherein at step d) said providing said performance measurement comprises calculating an average one-way delay (OWD_Q) for said cluster (C1 , C2, C3, C4) as a difference between an average output timestamp (Ta^out) and an average input timestamp (Τ_α'^η), wherein: - said average output timestamp (Τ_α ) is a weighted summation of average timestamps provided by said at least one output measurement point at the end of said marking period divided by said summation (C °^ut) of counters provided by said at least one output measurement point, the weight for the average timestamp provided by each output measurement point being the counter provided by the same output measurement point; and

- said average input timestamp (Τα^ιη) is a weighted summation of average timestamps provided by said at least one input measurement point at the end of said marking period divided by said summation (Ci'ⁿ) of counters provided by said at least one input measurement point, the weight for the average timestamp provided by each input measurement point being the counter provided by the same input measurement point.

1 1 . The method according to claim 9, wherein at step d) said providing said performance measurement comprises calculating a jitter based on:

- timestamp provided by said at least one input measurement point for a first packet received during said marking period and a timestamp provided by said at least one input measurement point for a last packet received during said marking period;

- timestamp provided by said at least one output measurement point for a first packet received during said marking period and a timestamp provided by said at least one output measurement point for a last packet received during said marking period; and

- number of packets received as a whole by said at least one input measurement point and number of packets received as a whole by said at least one output measurement point.

12. The method according to claim 9, wherein at step d) said providing said performance measurement comprises calculating an average round-trip delay (RTD_Q) for said cluster (C1 , C2, C3, C4) by:

- calculating a first average one-way delay (OWD_Ci ) for said cluster (C1 , C2, C3, C4) in a first direction and a second average one-way delay (OWD'_Ci ) for said cluster (C1 , C2, C3, C4) in a second direction opposite to said first direction; and

- calculating said average round-trip delay (RTD_Q) as a sum of said first average one-way delay ( OWD_Ci ) and said second average one-way delay ( OWD'_Ci ).

13. The method according to any of the preceding claims, wherein said step d) comprises providing at least one statistical performance measurement of said multipoint packet flow (PF) along an end-to-end path (W→Z) of said monitoring network (MN) by:

- identifying K cascaded clusters of said monitoring network

(MN) which are crossed by said end-to-end path (W→Z), K being equal to or higher than 2; and

providing said at least one statistical performance measurement using performance measurements provided for said K cascaded clusters.

14. The method according to claim 13, wherein said providing said statistical performance measurement comprises providing a statistical packet loss measurement (PL₍w→z)) for said end-to-end path (W→Z) by:

- calculating a cumulative packet loss probability (PLP_(K dusters)) of said K cascaded clusters as a function of packet losses probabilities of said K cascaded clusters; and

calculating said statistical packet loss measurement (PL(w→z)) for said end-to-end path (W→Z) as a function of said cumulative packet loss probability (PLP_(K dusters)) and a number of packets received at a measurement point terminating said end-to-end path (W→Z).

The method according to claim 13, wherein said providing said statistical performance measurement comprises providing a statistical average one-way delay measurement (OWD(w→z)) for said end-to-end path (W→Z) by:

calculating said statistical average one-way delay measurement (OWD₍w→z)) for said end-to-end path (W→Z) as a summation of average one-way delays of said K cascaded clusters.

A packet-switched communication network (CN) comprising: a subnetwork (SN) configured to support transmission of a multipoint packet flow (PF), each packet of said multipoint packet flow (PF) comprising a marking field set to a marking value equal to either a first marking value or a second marking value, said marking value being periodically switched between said first marking value and said second marking value;

a monitoring network (MN) comprising a plurality of measurement points (A, B, ... J) implemented in said subnetwork (SN), each measurement point (A, B, ... J) being configured to receive at least one of said marked packets and, if said at least one of said marked packets is marked by said first marking value, update a performance parameter; and

a management server configured to provide a performance measurement on said multipoint packet flow (PF), said management server (MN) being configured to:

identify in said monitoring network (MN) a cluster of measurement points (C1 , C2, C3, C4), said cluster of measurement points (C1 , C2, C3, C4) being a set of measurement points of said monitoring network (MN) selected so that, if no packet loss occurs, each packet received by at least one input measurement point of said cluster is also received at at least one output measurement point of said cluster; and

provide said performance measurement on said multipoint packet flow (PF) using performance parameters gathered from said at least one input measurement point and said at least one output measurement point of said cluster (C1 , C2, C3, C4).

A computer program product loadable in the memory of at least one computer and including software code portions for performing the steps of the method of any of claims 1 to 15, when the product is run on at least one computer.