FR3017507A1 - METHOD FOR ANALYZING DATA PACKET STREAMS - Google Patents

Abstract

The subject of the invention is a method for analyzing a stream of data transmitted in the form of packets and called the analysis stream, each packet of said stream comprising a counter whose value is representative of its transmission order, said method comprising the steps of: extracting (203) bit counter values associated with packets present in a sliding observation window of a length corresponding to an odd number W of consecutive packets of the analysis stream; determining (207) the median counter value in the observation window; determining (206) at least one indicator characterizing the flow, this indicator being deduced from at least one of the median values obtained at a given moment of measurement.

Description

The invention relates to a packet flow analysis method and applies to the field of telecommunications networks.

The transmission of binary data over telecommunications networks assumes that the transmissions are configured to meet QoS constraints specific to the content concerned. For example, in the case of video type data, the transmission must be able to be carried out in real time for certain applications such as for example the broadcasting of content in "multicast". For this, communication protocols adapted to the real time must be implemented. A well-known transmission protocol is the Real-time Transport Protocol (RTP). This protocol is widely used especially in the context of video-on-demand or Voice over IP (VolP) services. As regards the real-time transmission of a digital video, several packet-type streams may be transmitted, for example a video stream and audio stream. The real-time constraints related to this type of flow are not compatible with the use of Automatic Repeat-reQuest (ARQ) mechanisms aimed at solving the problem of packet loss on reception by specific requests for retransmission of non-packet data. received. Thus, alternative techniques are used. For example, it is possible to use network redundancy. In other words, a given packet stream is transmitted multiple times using separate telecommunications networks. In reception, the incident flows can be processed differently. A first technique is to select the incident data stream having the best performance in terms of quality of service. For example, the stream for which the packet loss rate is minimum may be selected. In another example, the redundant flows are adapted to allow recombination of the receive packets and thus obtain a recombined stream for improving the FEC (Forward Error Correction) correction rate.

However, it is difficult in these techniques to analyze the received data streams and in particular to measure their packet loss rate, to determine if packets have been received several times to determine the extent to which received packets have been received. have been disordered in relation to their order of transmission.

The characterization of the performance of packet flow transmission such as video streams is particularly useful especially for companies broadcasting content. Indeed, these companies may use different telecommunication operators or use different telecommunications networks belonging to the same operator to transmit data streams redundantly or not. It is therefore particularly important to have tools to measure the transmission performance of flows using one or more telecommunications networks. This allows, for example, the broadcaster to optimize its content distribution strategy by selecting the telecommunication networks best suited to its activity. An object of the invention is in particular to overcome the aforementioned drawbacks. To this end, the subject of the invention is a method of analyzing a stream of data transmitted in the form of packets and called the analysis stream, each packet of said stream comprising a counter whose value is representative of its order of transmission, said method comprising the steps of: extracting binary counter values associated with packets present in a sliding observation window 30 of a length corresponding to an odd number W of consecutive packets of the analysis stream; determining the median counter value in the observation window; determining at least one indicator characterizing the flow, this indicator being deduced from at least one of the median values obtained at a given moment of measurement. According to one aspect of the invention, an indicator is deduced from the comparison of at least two median values obtained at two consecutive measurement instants. An indicator is for example used to detect a loss of n consecutive packets, said loss of n consecutive packets being detected when a jump equal to n is identified between two successive median values, n being an integer greater than or equal to one . An indicator is for example implemented so as to detect that a packet is received several times, a packet being considered received M times when 15 M identical median values are identified successively. An indicator is implemented by identifying the position of the median value in a list Li comprising counter values present in the measurement window, a packet being considered to have been received according to its instant of emission when the position of the The median value is positioned in the center of the list Li, a disorder being detected and measured when there is a difference between the position found and the central position of the list Li. An indicator is for example implemented so as to compare the central value of a list Lc with the central value of a list Li, said list Lc corresponding to the list Li whose counter values have been classified in ascending or descending order, a packet being considered disordered if these two Central values are not identical. According to one aspect of the invention, a specific counter representative of packet disorder is determined by incrementing an integer variable each time a disorder is detected at a given measurement instant. According to one aspect of the invention, the data streams are RTP streams.

The bit counters associated with the analysis flows correspond, for example, to the "sequence number" fields present in RTP packet headers. The analysis flow corresponds for example to the multiplexing or recombination of at least two received data streams respectively resulting from the emission of two identical data streams. The invention also relates to a computer program comprising instructions for the execution of the method described above, when the program is executed by a processor. The subject of the invention is also a system for analyzing a stream of data transmitted in the form of packets and called the analysis stream, each packet of the stream comprising a counter whose value is representative of their transmission order. said system comprising means for: extracting binary counter values associated with the packets present in a sliding observation window of a length corresponding to an odd number W of consecutive packets of the analysis stream; Determining the median counter value in the observation window; determining at least one indicator characterizing the flow, said indicator being derived from at least one of these median values. According to one aspect of the invention, the system is adapted to implement an incremental implementation. According to one aspect of the invention, the system comprises an extraction module making it possible to obtain the counter values of the packets of an analysis flow, a calculation module having the function of comparing several counter values so as to obtain determining the location in a memory of the acquired new counter value so as to generate an ordered list Lc and a sequencer controlling the read and write operations in the memory for the determination of the median values. According to one aspect of the invention, the system comprises a register 5 for storing the last acquired counter value, this register being adapted to inform a sequencer of the reception of a new value and to transmit said value to a calculation module . Other characteristics and advantages of the invention will become apparent from the following description given by way of nonlimiting illustration, with reference to the appended drawings in which: FIG. 1 schematically illustrates the principle of extraction a median sequence number in an observation window; FIG. 2 is a diagram showing in a simplified manner the steps of a packet type flow analysis method; FIG. 3 gives a simple example illustrating the principle of the flow analysis method according to the invention; FIGS. 4a to 4b illustrate how the extraction of a medial sequence number in an observation window can be used to measure the disorder of the packets in reception; FIG. 5 gives an example of packet type flow analysis circuit. Figure 1 schematically illustrates the principle of extracting a median sequence number in an observation window. The example of the transmission of data streams composed of RTP packets is used in the following description. However, the invention can be implemented for other transport protocols based on the transmission of data packets, a data packet being associated with an incremented counter following the transmission of each packet. This counter is for example included in a header specific to the transport format considered. The invention described hereinafter is therefore applicable to other transport protocols. In the case of the RTP protocol, an RTP header is added to the packets. This header includes a sixteen-bit coded counter called Sequence Number. For a given RTP packet stream, this counter is incremented upon transmission of each new packet. Transmission of an RTP stream over a telecommunications network usually implies that some packets may be lost. In addition, some RTP packets may be received in a disordered manner. The invention is based on the use of a sliding observation window 103. This corresponds, for example, to FIFO data management (Fist In, First Out). The packets of the analysis stream 104 enter one after the other in this FIFO 103 to a depth corresponding to an integer number W of packets. At each flow analysis period, a new packet 105 of the analysis stream enters the analysis window that can be materialized by the FIFO 103 and a packet 106 comes out. The number W of packets is chosen odd. The value of W is furthermore chosen as a function of the network, the flux and in particular taking into account the cumulation of the time jitter. By way of example, W may be chosen so that it corresponds to a duration of the order of a few tens of milliseconds. In one embodiment, the sequence numbers contained in the packet RTP headers are extracted, and it is these sequence numbers that fill the FIFO. For each analysis period, the FIFO includes an offset sequence of sequence numbers. The invention is then based on an essential treatment that is performed on these number sequences.

This processing corresponds to the determination of the median sequence number present in the series of sequence numbers present in the FIFO. Obtaining these median sequence numbers from one measurement window to the other will allow finely analyzing the characteristics of the stream 104. The manner in which a medial sequence number in a list of W elements can be determined is part of general knowledge of the person skilled in the art. By way of example, the sequence numbers present in the FIFO at a given measurement instant can be classified in ascending or descending order of 1 0 so as to generate an ordered list of sequence numbers. The median sequence number can be estimated in this case by selecting the number present in the central position of the reordered list. The use of an observation window is implemented in particular because it makes it possible to detect the lost and disordered packets simultaneously. In addition, for packets that have been received, it is possible to determine whether they have been received several times. Fig. 2 is a diagram showing in a simplified manner the steps of a packet type flow analysis method. A first step 201 aims to receive N packet type data stream. The invention applies for N greater than or equal to one. A second optional step recombines or multiplexes multiple incident data streams into a single analysis stream. In a third step 203, the values of the bit counters associated with the packets of the analysis stream present in a sliding measurement window called the observation window are extracted. This acquisition corresponds, for example, to the extraction of the "sequence number" fields present in the RTP packet headers. As previously indicated, the length of the sliding window is equal to W elements, W being chosen odd. In the remainder of the description, W = 2xS + 1 where S denotes the amplitude of the disorder brought by the network. A step 207 is then implemented to determine the median value of the sequence numbers present in the observation window.

By way of example, the determination of the median value at a given measurement instant can be determined by generating a generated list Lc corresponding to the values of the sequence numbers acquired in the measurement window and sorted in ascending or descending order. then identifying the stored sequence number in central position of the list Lc. Then, a flow analysis step 206 is implemented in order, for example, to identify if packets of the original stream are missing in reception and / or if packets of this stream have been received several times. For this, the successive median values identified in step 207 are compared in pairs and indicators can be determined. Examples of indicators are given later in the description. The method according to the invention has been described for an analysis flow resulting from the transmission of a packet stream through a given telecommunication network. However, to improve the quality of service associated with a given stream, it is possible to transmit several identical streams. Thus, transmission redundancy can be introduced. In addition, each of the duplicate flows can borrow a different telecommunications network. Thus, if a first network has failures for a given duration, the transmission of a redundant stream on a second network can compensate for the losses and delays resulting from failures. The method according to the invention as well as the devices implementing it can be applied in this type of situation. In the example of FIG. 1, an exemplary transmission of two redundant data streams 101, 102 is shown in the box 110. The method is therefore applicable for a number N greater than or equal to one of redundant data streams. If N equals one, the analysis flow corresponds to the result of the transmission of a single transmission stream. In Example 110, the incident data stream 101 results from the transmission of an original data stream over a first telecommunications network and the incident data stream 102 results from the transmission of the same original data stream. on a second telecommunications network. The streams 101, 102 are combined or multiplexed 100 so as to generate a single data stream corresponding to a succession of packets 104. This single stream is in this case used as an analysis flow for the application of the method described above.

As mentioned above, when several streams are used to generate a single flow of analysis, several techniques are possible. The first technique that can be implemented is to recombine the N incident flows 101, 102. In this description, the recombination corresponds to the generation of a single stream from at least two so-called incident packet streams so as to avoid that the single stream has duplicate packets. One way of implementing this recombination is to use, for example, a memory with N write pointers, ie an incident flux pointer and a single read pointer. When recombination is used, each received packet uses its sequence number as the write base address, which writing then decrees the corresponding memory area as nonempty. If the same packet is received several times, either from the N streams or because it has been duplicated on some streams, it will be rewritten from the same base address as many times as it has been duplicated. from N tributaries or within a single flow. The single read pointer is incremented and reads the base addresses corresponding to non-empty memory areas with a delay with respect to the writes chosen according to the cumulative jitter of the networks. Reading a packet declares the corresponding memory area empty. Thus, the possible missing packets of a stream are compensated by the N-1 identical packets of the other streams except if a packet is absent on the N streams simultaneously. In addition, redundant packets received on multiple incident streams are deleted. When recombination is used, there are no duplicate packets and the packets are reordered.

The second technique that can be used as an alternative to recombination is multiplexing. All the packets received on the two streams are grouped into a single stream. Unlike recombination, packets are not reordered using their sequence number and packets that have been received multiple times will appear as duplicated in the multiplexed stream. In the case where several redundant flows are received, the method of characterizing a packet stream can be implemented on a recombined or multiplexed stream, but also on a given incident stream. It is thus possible to analyze the performance of a given telecommunications network and / or analyze the performance resulting from the use of a plurality of telecommunications networks transmitting redundant packet flows. This means that a device implementing the method according to the invention can be positioned downstream of the device performing the recombination or the multiplexing but also upstream. Upstream, a device is associated with a given incident flow. For these devices, the analysis flow then corresponds to the incident flow associated with them. FIG. 3 gives a simple example illustrating the principle of the flow analysis method according to the invention. A sequence 300 of packets corresponding to a portion of an analysis stream is represented. The numbers indicated within each packet correspond to the packet counter, for example the sequence number in the case of RTP packets. In this example, the observation window is of length W = 5. Lc reordered lists are shown for different clock ticks and the median values 301 associated with them, said median values corresponding to the central values of these lists. It appears that the values 302 and 304 are identical. The two-by-two comparison of the median values makes it possible to detect this type of event. In this case, the result of the comparison is an indication that a packet having a sequence number equal to four has been received twice.

Thus, an indicator may be implemented to detect in sequence 301 median values the sequence numbers following each other and having the same value in order to measure and detect the packets of the analysis stream that have been received several times. Then, the sequence numbers five 305, seven 306, eight 307 and eight 308 are produced as median values of the Lc listings obtained for different observation windows. An indicator may also be implemented to detect lost packets identified by discontinuities in the median value sequence 301 produced. A discontinuity corresponds to a jump greater than one between two successive counter values in the sequence 301 of median values. In the example of FIG. 3, a discontinuity appears between the median values 305 and 306 respectively corresponding to the sequence numbers five and seven. The loss of the packet having the sequence number six has thus been detected. In addition, two median values 307, 308 are identical and equal to eight.

Thus, it has been detected that the packet having a sequence number equal to eight has been received in duplicate. Figures 4a and 4b illustrate how the extraction of a medial sequence number in an observation window can be used to measure the disorder of the packets in reception. In these examples, the extraction window is of length W = 7. In these examples, the received packets were mixed during the transmission. In the remainder of the description, the list Li corresponds to the sequence numbers present in the observation window without these having been reordered.

In a first alternative (FIG. 4a), an indicator is implemented to identify the position of the median value 402 in the list Li. If the position of the median value is the central position of this list, the packet is considered as having received according to its time of issue. In the case where the position of the median value 402 is different from the central position of the list Li, the difference 404 between the position found and the central position is determined and corresponds to a measure of the disorder of the packets in reception. In a second alternative (FIG. 4b), another indicator of the disorder of the packets is implemented by comparing the central value 406 of the list Lc with the central value 405 of the list Li, the list Lc corresponding to the list Li whose items were ranked in ascending or descending order. If these two values are identical, it means that the packet has not been mixed in the stream. Otherwise, it means that this package has been disordered in the scan stream. Thus, a specific counter representative of packet disorder can be determined by incrementing an integer variable at each measurement period when a disorder has been detected in the measurement window. FIG. 5 gives an example of a packet type flow analysis circuit. Although simple in principle, the implementation of ranking functions can be complex. If the algorithm described is implemented entirely with a hardware description language such as VHDL (VHSIC Hardware Description Language) or Verilog, a hybrid implementation (based on software and hardware) is also suitable, especially in the context of of an incremental embodiment. For example, two approaches can be considered. In a first approach, a CPU (Central Processing Unit) processor is used to implement the steps of the method described above. An FPGA (Field-Programmable Gate Array) circuit can then implement a buffer storing consecutive sequence number values at the rate of their arrival and waiting for the host CPU to read them for processing.

In a second approach, an embedded microcontroller software implements steps 201-205 and transmits the result of processing associated with a host CPU processor for it to implement the analysis 206 and the generation of indicators.

In both of these approaches, the performance requirements of the embedded microcontroller or host computer can be high. In the case corresponding to a rate of 1Gb / s, a complete cycle of treatment lasts two microseconds including the restitution of the previous context that is to say the previous state of the FIFO, the calculation of the new median value and the 1 0 update estimators. In one embodiment, a packet type flow analysis device according to the invention can be implemented using an incremental implementation. This is possible because: a minimum time of the order of a few microseconds between two RTP packets can be exploited, which usually represents a large number of clock ticks and makes it possible to envisage a series of processing and not of parallel; - the sliding analysis window moves only one position between 2 consecutive RTP packets. Consequently, the update of the order number of each sample already present in the memory can be deduced from a simple test from the previous sequence number by simple comparison of the new incident sequence number and the sequence numbers already present. in memory.

In the example of FIG. 5, the flow analysis system comprises an extraction module 505 making it possible to obtain the sequence number values of the packets of the analysis stream 506. A register 500 makes it possible to memorize the last acquired sequence number value. This register furthermore makes it possible to inform a sequencer 501 of the reception of a new value and to transmit said value to a calculation module 503. The function of the calculation module is to compare several sequence number values so as to determining the location in a memory 504 of the new acquired sequence number value. This memory 504 is used to store W sequence number values classified for example in ascending order. In other words, the memory 504 is updated to store a list Li representative of the current measurement period. Sequencer 501 controls 507, 508 the read and write operations in memory 504 and determines the median counter value at each measurement period. These median values 502 are transmitted to a module determining estimators characterizing the flow thus analyzed. The memory 504 preferably dual port is for example implemented with an FPGA circuit. The fast calculation unit 503 can be implemented by a dedicated circuit. Consider that upon receipt of a new sequence number value Tmp (o) .SeqNum in the register 500, the memory 504 contains the last W received sequence numbers MEM (j) .SeqNum and their number. order MEMÙ) .Order assumed to be increasing, that is to say that the most anterior data item has the order number 0 while the most recent data item has the order number W resulting from the previous processing step 20 . To reactualize the contents of the memory, it will be enough then: - to compare the new value of received sequence number Tmp (o) .SeqNum with all the values of sequence number MEMOE1) .SeqNum in memory, and to memorize in the passage the Sequence number corresponding to the searched sequence number (Median = Memum) .SeqNum such as Memum) .Order = S), thus possibly measuring the disorder (Shuffle_est = jm- S); - to decrement MEMù - 1) .Order. if MEMù - 1) .SeqNum <Tmp (o) .SeqNum 30 - to shift all MEMa-i) .SeqNum and MEMù - 1) .Order one position up in the memory; in order to store MEM (o) .SeqNum and its MEM (o) .Order in position 0 of the memory at the end of the cycle.

Claims (15)

CLAIMS1- A method of analyzing a stream of data transmitted in packet form and called analysis stream, each packet of said stream comprising a counter whose value is representative of its transmission order, said method comprising the following steps: extraction (203) of bit counter values associated with the packets present in a sliding observation window of a length corresponding to an odd number W of consecutive packets of the analysis stream; determining (207) the median counter value in the observation window; determination (206) of at least one indicator characterizing the flow, this indicator being deduced from at least one of the median values obtained at a given instant of measurement. 2. The method of claim 1 wherein an indicator is derived from the comparison of at least two median values obtained at two consecutive measurement times. The method of claim 2 wherein an indicator is implemented to detect a loss of n consecutive packets, said loss of n consecutive packets being detected when a jump equal to n is identified between two successive median values, n being a integer greater than or equal to one. 4- Method according to one of claims 2 or 3 wherein an indicator is implemented to detect that a packet is received several times, a packet being considered received M times when M identical median values are identified successively. 5- Method according to one of claims 2 to 4 wherein an indicator is implemented by identifying the position (402) of the median value in a list Li comprising counter values present in the measurement window, a packet being considered to have been received according to its moment of emission when the position of the median value is positioned in the center of the list Li, a disorder being detected and measured when there is a difference (404) between the position found (402) and the central position of the Li list. 6. Method according to one of claims 2 to 4 wherein an indicator is implemented so as to compare the central value (406) of a list Lc with the central value (405) of a list Li, said list. Lc corresponding to the list Li whose counter values have been classified in ascending or descending order, a packet being considered disordered if these two central values are not identical. 7- Method according to one of claims 5 or 6 wherein a specific counter representative of the disorder of the packets is determined by incrementing an integer variable each time a disorder is detected at a given measurement instant. 8. Method according to one of the preceding claims wherein the data streams are RTP streams. The method of claim 8 wherein the bit counters associated with the analysis streams correspond to the sequence number fields present in RTP packet headers. 10- Method according to one of the preceding claims wherein the analysis stream corresponds to the multiplexing or recombination of at least two received data streams resulting respectively from the emission of two identical data streams. 11- computer program comprising instructions for the execution of the method according to any one of claims 1 to 10, when the program is executed by a processor A system for analyzing a stream of data transmitted in the form of packets and called the analysis stream, each packet of the stream comprising a counter whose value is representative of their transmission order, said system comprising means for: extracting binary counter values associated with the packets present in a sliding observation window of a length corresponding to an odd number W of consecutive packets of the analysis stream; determine the median counter value in the observation window; determining at least one indicator characterizing the flow, said indicator being derived from at least one of these median values. 13- System according to claim 12 adapted to implement an implementation of incremental type. 14- System according to claim 13 comprising an extraction module (505) for obtaining the counter values of the packets of an analysis stream (506), a calculation module (503) having the function of comparing several counter values to determine the location in a memory (504) of the newly acquired counter value to generate an ordered list Lc and a sequencer (501) controlling (507, 508) the read and write operations in the memory (504) for determining the median values. 15- System according to claim 14 comprising a register (500) for storing the last acquired counter value, this register being adapted to inform a sequencer (501) of the receipt of a new value and to transmit said value to a module of calculation (503).