WO2011124742A2

WO2011124742A2 - Method for estimating the parameters of a control element such as a token bucket

Info

Publication number: WO2011124742A2
Application number: PCT/ES2011/070239
Authority: WO
Inventors: Javier Aracil Rico; Javier Ramos De Santiago; Pedro María SANTIAGO DEL RÍO; Jorge LÓPEZ DE VERGARA MÉNDEZ; Luis DE PEDRO SÁNCHEZ; Sergio LÓPEZ BUEDO; Ivan GONZÁLEZ MARTÍNEZ; Francisco Javier GÓMEZ ARRIBAS
Original assignee: Universidad Autónoma de Madrid
Priority date: 2010-04-09
Filing date: 2011-04-08
Publication date: 2011-10-13
Also published as: ES2372213A1; WO2011124742A3; ES2372213B1

Abstract

The invention relates to a method and a program for estimating the parameters of a control element such as a token bucket for controlling or regulating data packet transmission (QoS) in communication networks. Said method enables the non-intrusive estimation of the parameters for each service controlled by a token-bucket regulator. Said method can also discriminate interferences with said service, and can be carried out in the presence of processes in systems that have shared resources. Said method enables the acquisition of the parameters of a control element such as a token-bucket in such a way that service level agreements (SLA) between a company and a telecommunications operator can be checked. The invention also relates to a computer program that can carry out said method.

Description

METHOD FOR ESTIMATING THE PARAMETERS OF A TOKEN-B U CKET TYPE CONTROL ELEMENT

DESCRIPTION

Technical Field of the Invention The invention is in the field of Quality of Service (QoS) technologies for the control or regulation of the transmission of data packets in communication networks. More specifically, it refers to a non-intrusive method that allows estimating those parameters for each service controlled by a token-bucket regulator. In addition, this method is able to discriminate interference through said service allowing it to be executed in the presence of processes in those systems with shared resources. Through this method it is possible to obtain the parameters of a control element of the token-bucket type, so that it allows verifying service level agreements (SLA) between a company and a telecommunications operator. State of the art

QoS technologies are especially important for certain applications - such as video or voice transmission - since they try to guarantee the transmission of a certain amount of data in a given time. Among the technologies is the token-bucket algorithm that is an admission control system of a network. That is, an algorithm whose objective is to limit the transmission rate at which a type of service or a user injects traffic into a network.

The operation of a token-bucket algorithm can be found in the state of the art. In particular, the algorithm supports the entry of information packets and generates tokens, particular instances associated with information packets, at a constant rate of token generation (CIR). These tokens are capable of transporting information packets from the entrance. These tokens accumulate in a "cube" of size B _r . In this algorithm a token is removed when a data packet is transmitted. The system can consume all the tokens accumulated in the cube if a sufficient volume of data is reached so that it is possible to send bursts of data (bursts) at a maximum transmission rate (PIR) when tokens are available unlike other algorithms Like the leaky bucket.

Thus, before the entry of a variable flow of data, if there are accumulated tokens, they are used to transmit packets at the maximum transmission rate; and, if they have been consumed, packets can only be transmitted as tokens are generated, that is, at the maximum token generation rate.

This same token-bucket algorithm can also be applied to limit at different speeds depending on the type of service. For example, give a speed for real-time applications (teleconference, video on demand, etc.) and a different one for file transfers. This is done by adding a queue and a cube for each type of service, thus being able to give a different transmission rate to each service.

The determination of the parameters CIR, PIR and size of the cube is considered as a technical problem, since this cannot be done using generic bandwidth measurement tools, such as SpeedTest (www, speedtest.net). These tools are not able to measure a limited link by the token-bucket algorithm because they result in a bandwidth that reflects an average of the CIR and the PIR, weighted according to the size of the cube. Therefore, they would not offer a reliable result when verifying an SLA contract, and in no case would they be able to extract the CIR, PIR and B _r parameters that limit the link, neither in the case of a single queue, nor, much less, in the case of several queues depending on the type of service.

Other methods such as those described by Rosario G. Garroppo, Stefano Giordano, Michele Pagano, "Estimation of token-bucket parameters for aggregated VoIP sources" International Journal of Communication Systems Vol. 15 n. 10 pp 851-866 (2002), propose an analytical method for estimating the CIR and B _r parameters of a token-bucket filter. However, the main problem of the proposed method is the need to know the stochastic model of the multiplexed traffic of the entrance, which is not viable in a real environment, where the distribution of traffic is not known and is very dependent on the context of the concrete network

In other words, there are no known methods to estimate these token-bucket parameters in a non-intrusive way, and it is also necessary to estimate the CIR, PIR, B _r parameters in the presence of interfering traffic. Interfering traffic is present in many environments where the measure is of maximum industrial interest. The presence of an interfering traffic can distort the measure being an obstacle to measure a link in use, not empty, which is the environment of greatest interest on many occasions.

In addition, when the estimation of these parameters is carried out by means of a dedicated or shared device, (ie in which processes are executed at the same time as the measurement process), it occurs that the estimation may be contaminated due to contaminating processes in the same thread of execution (thread) and therefore not be representative for a correct traffic control. It is therefore necessary to provide a method capable of estimating those key parameters and select those representative estimates of the measure, discarding the rest. The applicant does not know of solutions that allow the measurement or estimation of the parameters of the token-bucket algorithm of a commercial router as does the method presented, both for the case of a single queue, as for the case of several queues according to type of service. Description of the invention

The object of the invention is to provide a method for estimating the parameters of a token-bucket control element, comprised in a communications network. This method allows to estimate the token-bucket parameters with the accuracy and precision required to verify service level agreements (SLA). Said method is defined by independent claim 1. In a second inventive aspect, a program for implementing this method according to independent claim 15 is provided.

A first inventive aspect comprises a method for estimating the token-bucket parameters where, among other stages, a change is detected between a first (maximum) transmission speed and a second speed, and allows the CIR, PIR parameters and the cube size to be determined. in operational mode This method comprises the following steps:

- A train of N probe packages marked for identification is generated and injected into the input line by means of the emission means. In a first step a train of N packages is generated that we will call "probe packages". These probe packages are marked for identification with a package identifier. Thus, reading the subsequent package of these packages allows identifying a read package as a package belonging to the probe packet train. The probe packets are injected into the input line, which may or may not have interfering packets, preferably using a service, using emission means. In an exemplary embodiment, these packages have a similar size B.

When the broadcast media supplies the probe packets to the input line, these packets are inserted into the data flow regulated by the token-bucket. The token-bucket control element receives the packets and transmits them on the output line together with two other packets.

- the moments of passage of each of the probe packets are read through the probe means of the output line.

In a second step, probe means allow the instants of passage time tL, ... t _N to be read. These moments reflect the time that each probe pack reaches the probe means in the output line.

- the transmission speed is determined from the moments of passage of the probe packets,

In a third step, from the different times t, ... t _N , the data transmission rate is determined by the probe packets, and, for a train of several probe packets, a transmission rate of data differentiating between the probe packets that arrive at the beginning of the probe packet stream, and the last probe packets that arrive at the end of the probe packet stream. The packet transmission is regulated in operational mode by the size of the cube (B _r ) and the CIR and PIR parameters.

- from the transmission speed of the probe packets it is established whether there is a change in speed between the probe packets at the beginning and the probe packets at the end of the probe packet stream. The packet flow in the output line is regulated by the token-bucket, and the data transmission rate is limited to a PIR speed. The change of speeds from the PIR to the CIR allows determining the parameters of interest according to the method of the invention. - If there is a change in speed:

• the speed of the first probe packets determines the maximum transmission rate at the output (PIR),

• the speed of the second probe packets determines the token generation rate (CIR); and, 'the volume of data corresponding to packets with the maximum transmission rate determines the size of the cube in operating mode (Br); and if there is no change in speed, the measure is considered invalid.

If there is a change in the speed of these packets that corresponds to a deceleration from an observed speed before the speed change, the method allows the PIR to be determined from the speed of the first probe packets, before the change in speed, and the CIR from the speed of the second probe packets, after the change of packet speed. The first, higher speed, is due to the fact that the train of probe packets has found accumulated tokens, and the second speed is due to the fact that the accumulated tokens have been used up and can only be transmitted as tokens are generated. The cube size parameter (B _r ) in operating mode is determined from the volume of data transmitted at maximum speed. By choosing these conditions in the evolution of packet speed, a method is provided that seeks to detect the conditions in which the token-bucket cube delivers accumulated tokens. This criterion makes the estimation of the speeds of the first and second probe packets reflect the parameters of a token-bucket when measuring the time associated with the arrival of the probe packet train. Therefore, if this change in speed does not occur, the measure is considered invalid.

Thus, this method, not described in the state of the art, allows, among other advantages, to provide a method with which service level agreements can be verified non-intrusively. As the embodiment example will be detailed, the invention according to independent claim 1 may also comprise rejection criteria of the estimated measures. The various embodiments resulting from the combination of dependent claims 2 to 14 are incorporated by reference to this description. In a second inventive aspect a computer program is provided which, when executed on a computer or any means to process information, can perform the method described in the first inventive aspect.

Brief description of the figures

These and other features and advantages of the invention may be more clearly apparent from the detailed description that follows in a preferred embodiment, given only by way of illustrative and non-limiting example, with reference to the accompanying figures:

Fig. 1. Schematic representation of a token-bucket regulator; Fig. 2. Schematic representation of a token-bucket regulator regulating access to a network;

Fig. 3. Schematic representation of a program according to the invention executed on a computer;

Fig. 4. Representation of an interference process between interfering packets and a train of probe packets;

Fig. 5. Schematic representation of a set of token-bucket regulators for a set of services; Y

Fig. 6. Schematic representation of an implementation of a computer program.

Embodiments of the Invention

An embodiment of the method of the invention is described below. A diagram of part of the set of elements that configures the example of the embodiment of the invention is shown in Figure 1. An emission means (4), - these emission means (4) can be any device from which a data stream such as a computer is emitted -, are connected to a communications network regulated by a control element of the type token-bucket or token-bucket (3). In these transmission means (4) N packets (1.1—1. N) probe are generated for transporting information, which form a train (1) of probe packets. In an exemplary embodiment of the invention the size of the probe packet (1) is substantially similar to the maximum network transmission unit (MTU).

Each identifier (1.1—1. N) probe includes an identifier, which can be integrated into the structure of each packet (1.1—1N) probe, depending on the communications protocol that is implemented. As an example, in a packet (1.1) probe with IP structure, the identifier can go in the header or in another part of the structure of this packet (1.1) probe. This identifier allows each packet (1.1-1 .N) probe to be identified as belonging to this train (1) of probe packets when each packet (1.1-1N) probe is transmitted through the token-bucket (3). In this example, the number of probe packets (1.1-lN) in the train (1) of probe packets is 100.

This example can be generalized to an end-to-end channel between the emission means (4) and the probe means. In this end-to-end channel a series of links and nodes are included to transmit the information through a network such as the Internet (6) Figure 1 shows an example of embodiment with several additional input lines (2.1-2.3) for the same service leading to the same token-bucket regulator (3). In the following an example of the token-bucket regulator (3) is described as the one shown schematically in figure 2. In this scheme an input line (2) ends in a token-bucket type control element (3 ). The probe packets (1.1-lN) arrive through the input line (2), and are transmitted through the token-bucket regulator (3) depending on the token generation rate (CIR), the size of the cube and (B _r ) with a maximum transmission speed (PIR).

From this token-bucket (3) starts an output line where the probe means (not shown in the figures) are located. In one example the probe means can be a dedicated device (s), or not dedicated (s) like a computer.

The probe means read the instants of passage time ordered as t _í <t ₂ <··· <t _N of each of the packets (1.1-lN) probe.

In this example the transmission speed information (or data transmission rate, data rate) between and ^'-th package (l .j) probe and the next packet B train probe (1) of probe packets can be defined as v¡ «-, j = Ι, ..., Ν - l,

^■ j + 1 ^LL ji where B is the size of each probe package (1.1-lN) in the train (1) of probe packages.

Thus, it is possible to monitor the evolution of the information transmission rate of the probe packets (1.1-lN) arriving at the probe medium and determine if there is a change in the information transmission rate. When the information transmission rate is represented as a function of time, said change is observed as a deceleration in the information transmission rate. This change (deceleration) corresponds to the situation in which there are no tokens in the cube. For the first r-1 packets (1.1-lr-1) probe (before the change) this information transmission rate corresponds to the maximum transmission rate (PIR), since the cube contains accumulated tokens available for use. Once the tokens are consumed in the cube, the information transmission rate of the second packet group (lr-l) probe determines the token generation rate (CIR). The volume of data until the change determines the size of the cube in operating mode (B _r ). If this change in the behavior of the information transmission rate is not detected, it is considered that the measurement is not valid, so it must be repeated if you want to determine or estimate the token-bucket parameters (3). In an exemplary embodiment, the change in the rate of information transmission is established when the time between the arrival of two consecutive probe packets t _{j + 1} - t _j to the probe means is greater than T _Nom «0.1 B / CIR _nom , where CIR _nom is included in the service level agreement. The PIR parameter of the token-bucket (3) can be determined according to

PIR «min, i = 1, r— 1. The CIR parameter of the token-bucket (3) can be determined according to

The parameter? _{r dd} token-bucket (3) can be determined according to

The bandwidth (i? J) can be estimated as

Ρ / Λ + (N - r) x C / Λ

BW *

N

Figure 3 shows a situation in which the transmission of this train (1) of packets (1.1-l.N) probe is interfered with by an interfering packet (7.1) before arrival at the token-bucket (3). Said interfering packet (7.1) corresponds to a packet issued by other applications or services (8.1-S.s) that are running in the issuing device. This interfering packet (7.1) is sandwiched between the packets (1.1, 1.2) train probe (1) of probe packets and causes interference to the extent that they distort the token-bucket parameter estimation (3).

In an exemplary embodiment, it is estimated or determined if, when injecting in the input line (2), the probe packages (1.1-1N) there is at least one interfering package (7.1) in the N-1 gaps left between the packages (1.1 - lN) probe. That is, if there are no two packages (l.k, l.k + l) contiguous probe, k = 1, ... N— 1,

In an exemplary embodiment, the number of packets emitted by the transmission means (4) is counted from when the first probe packet (1.1) of the train (1) of probe packets is received in the probe means until it is received in The probe means the last packet (lN) probe. If the number of packets accounted for N + m is much greater than N, the reading for interfering packets (7.1-lm) is discarded. In an implementation of the embodiment example, if N + m exceeds a threshold value M of rejection, the measure is rejected.

Another variant of embodiment is provided with a safer criterion where a probability (P) is estimated that evaluates as an event that all gaps between packets (1.1-1) probe have at least one interfering packet (7.1). When the calculated probability is above an acceptance threshold value T, it is considered that there are no two adjacent probe packets (l.k, l.k + l). The range of T values between 0.01 and 0.05 provides a safer rejection level. Therefore, this criterion allows us to rule out those situations in which the estimate moves away from the real value since in these cases the measurement between the information transmission rate of two packets does not correspond to the information transmission rate between two packets (lk , l.k + l) contiguous probe.

The calculation of the probability of finding at least two packets (l .k, l.k + l) adjacent probe contemplates the case of N packets (1.1-lN) probe and m interfering packets (7.1- Ί .ni) that can be distributed in the N-1 gaps between the N packets (1.1-lN) probe. The number of interfering packets (7.1-7. Ni) that can be located in each hole is n¿, i = 1, ..., N— 1 „and of course m = E -Γι ¹ n¿ must be verified . To get an estimate of the probability, it is considered that each of the interfering packets (7.1-l.ni) has the same probability of occupying any of the gaps (10) between the packets (1.1-lN) probe, and that this probability It is independent for each packet (7.1-l.ni) interfering. Under this hypothesis, the probability density function that models the occupation of the gaps corresponds to a multinomial distribution for N and ra given by the expression P _mul (n, ...,

To find the probability that there is at least one interfering packet (7.1) (n¿> 1, i = 1, ..., N— 1) for all gaps, the next minimum P _muí (min¿ _{must be} calculated) _{= 1 w - 1} n¿> 1).

This probability P _mu i (m _{i = 1 W} -i ¿≥ 1) ^is I ^at 1 ^{UQ is} compared with a threshold value T. As shown in Figure 4 this situation does not correspond to a real situation because the The distribution of these interfering packets (7.1-lm) does not have a uniform probability distribution of intercalating in any gap. The starting hypothesis considers, however, that each interfering packet (7.1-lm) can occupy any gap. This is not true since an interfering packet (7.1) cannot occupy tl gaps (9.1-9tl) between t packets (1.1- \ .t) probe that have already passed. The result of applying this hypothesis is a safer rejection criterion.

In another variant embodiment, the transmission means (4) are implemented in a computer - or in any programmable medium - containing storage means - such as a memory - and a data processing unit. This allows the method to be developed in devices shared with other applications (8), as shown in Figure 3. In this way, the parameters of the probe means are not taken into account, but of the emission means (4), which allow discriminate the measure either based on the fraction of time used by the CPU to carry out the issuance process against the total time used, or based on the fraction of free memory during the process realization versus the total available memory, or a combination thereof. The threshold value for the CPU load level is CPU _{\ oaá} . The threshold value for media load level of storage is ΜΕΜ \ _0Ά &. If any one of these fractions exceeds a threshold value, CPU \ _or à or MEM \ _or ¿, the measurement is rejected.

This allows to avoid those situations in which other applications or services (8.1-S.s) provide slow down the emission of the probe packets (1.1- \ .N), distorting the measurement of the speed of the packets transmitted to the probe means.

In an additional variant, the criteria on the CPU fraction and the previous memory fraction can be combined to reject these measures.

Finally the same embodiments and variants of the methods can be used in any combination. In the example of figure 1, several additional input lines (5.1- 5.3) for the same service can be seen leading to the same tofen-bucket regulator (3).

In the example of figure 5, the set of parameters is illustrated when the speed is limited for each service (C1R 1, ..., CIR t; B _r l, ..., B _r t; PIR) and there are two input lines (2.5.1) regulated by token-bucket (3). An exemplary embodiment is also provided for a computer program (11). A variant embodiment is shown in Figure 6 where a first stage (11.1) is included in 1 that the emission and probe means (4) are connected. In a second stage (11.2) a burst of packets (1.1-lN) probe is sent through an end-to-end channel regulated by a token-bucket (3). The computer program (1 1) obtains by means of an implementation of any of the combinations of the examples of the embodiments of the method described above the parameters on the speed change, the interfering traffic and the fraction of CPU used. After this, a series of stages are carried out where the measurement is rejected based on the different conditions expressed in the examples of the embodiments of the method previous. In this example, an example is included which comprises: a step (11.3) that verifies if the number of interfering packets (7.1-7.m) is greater than a threshold; a step (11.4) that verifies if the fraction that reflects the level of CPU usage is greater than a threshold; In one stage (11.5) repeat the stages (1 1.2-11.44) on the number of queues for each service. In a last stage (11.6) the computer program (11) shows the result of the parameter estimation.

Claims

1.- Method for estimating the parameters of a token-bucket control element (3) comprising:

- a token-bucket control element (3);

- a starting line;

- at least one input line (2);

- transmission means (4) for the injection of data in the input line (2); Y,

- probe means for reading data on the output line;

wherein the token-bucket control element (3) is characterized by:

- a token generation rate (CIR);

- a maximum output transmission speed (PIR); Y,

- a cube size (B _r ), where the method comprises the following steps:

- a train (1) of N probe packages marked for identification is generated and injected into the input line (2) by means of the transmission means (4),

- the moments of passage of each of the packets (1.1-1 .N) probe are read through the probe means of the output line,

- the transmission speed is determined from the instants of the packet time (1.1-l.N),

- from the transmission speed of the packets (1.1-lN) probe it is established whether there is a change in speed between the packets (1.1-lr-1) probe of the beginning and the packets (lr-lN) probe of the end of the train (1) probe packets,

- If there is a change in speed: • the speed of the first packets (1.1-lr-1) probe determines the maximum transmission rate at the output (PIR),

• the speed of the second packets (l.r- l.N) probe determines the token generation rate (CIR); Y,

· The volume of data that corresponds to the packets with the maximum transmission rate determines the size of the cube in operational mode

(B _r );

and if there is no change in speed, the measure is considered invalid.

2 - Method according to claim 1 characterized in that in the injection in the inlet line (2) of the N packages (1.1-lN) probe it is estimated or determined if all the gaps between packages (1.1-lN) probe have at least one interfering package (7), in which case the invalid measure is considered.

3. - Method according to claim 2 characterized in that it is estimated if there are no pairs of packages (lk, l.k + l) contiguous probe as follows: · the probability P of all the gaps between packages (1.1- lN) probe have at least one interfering packet (7.1),

• a threshold value of acceptance of the mean is determined, if the probability P is greater than T then the measure is considered invalid.

4. Method according to claim 3 characterized in that the probability P that all the N-1 gaps between packets (1.1-l.N) probe have at least one packet

(7.1) interfering is established as follows:

• In the injection into the line (2) of the N packets (1.1-lN) probe, the number of interfering packets (7) m injected through the same means (4) of emission and that are intercalated between the packets (l. ll.N) probe,

• given the hollow Nl, the z ^' -th slot may be occupied by n packets (7) interfering with i = 1, ..., N— 1, where m = E -Ti ¹ ^ so that it is considered that each of the interfering packets (7) has the same probability of falling into any of the gaps between packets (l. ll.N) probe,

• it is considered that each of the interfering packets (7) may fall into a particular gap regardless of whether another interfering packet has done so,

· The density function multinomial probability Pmui ⁽ⁿ i> ⁿ 2> is determined - ⁿ Ni and from this the probability P = P _mu is determined (min _{i = 1 w -. 1} nDo> 1).

5. - Method according to claim 2, characterized in that in the injection in the line (2) of the Npacket (1.1-lN) probe, it is determined: · the number m of interfering packets (7) injected through the same means (4) emission and that are interleaved between the packets (l. Ll.N) probe,

• a rejection threshold M value, the estimate that all gaps between packets (1.1-l.N) probe have at least one interfering packet (7.1) is established when N + m is greater than M.

6. - Method according to claim 1 characterized in that the transmission means (4) are implemented in a computer containing storage means and a data processing unit.

7. - Method according to claim 6, characterized in that if the load level of the data processing unit exceeds a threshold value CPU _{\ oaá} then the measurement is rejected.

8. - Method according to claim 6 characterized in that if the load level of the storage media exceeds a threshold value ΜΕΜ _{\ 0Ά &} then the measurement is rejected.

9. - Method according to claim 1 wherein the speed change is established when the time between two consecutive packets (l .k, l .k + l) is greater than 7N _0m .

10. - Method according to claim 1 wherein the PIR parameter is estimated as the minimum of the speed between packets (1.1-l.r-1) probe before the speed change.

1 1. - The method according to claim 1 wherein the CIR parameter is estimated as the minimum of the speed between packets (l.r- l.N) probe after the speed change.

12. - Method according to claim 1 wherein the size of the cube in operating mode

B is estimated as the product between the factor

transmitted before the speed change.

13. - A method according to claim 1 wherein the packet size (1.1-l. N) probe substantially maximum network transmission unit (MTU).

14. - Method according to claim 1 wherein the number of probe packets (1.1-lN) the train (1) of probe packets is 100. - Computer program for estimating the parameters of a control element of the token-bucket type (3) in a computer according to a set of instructions that implement in a computer a method according to claim 1.