CN1581848A - Flow regulating method for ensuring integrated flow fairness of guaranted repeat business - Google Patents

Abstract

A traffic regulation method based on congestion charging to ensure the fairness of "AF service" converged flow in a differentiated service network, which is to apply the flow control method based on congestion charging used in the Internet to ensure the fairness of user ratios to differentiated services In the network: first establish the traffic regulation mathematical model of the AF aggregation flow based on congestion charging in the differentiated services network, and send detection packets between the edge routers of the differentiated services network to obtain the congestion fee of the current network, and then according to the congestion fee information Calculate the optimal sending rate of each AF aggregation flow according to the principle of net utility maximization, and adjust the token bucket parameters of each AF aggregation flow according to the optimized sending rate, so as to ensure that each AF aggregation flow obtains a bandwidth sharing share approximately proportional to its target rate , to achieve proportional fairness of shared resources among AF converged flows. This method solves the defect that the existing traffic adjustment algorithm in the differentiated services network cannot guarantee the fairness of the AF converged flow in proportion to the target rate.

Description

A traffic regulation method to ensure the fairness of "guaranteed forwarding service" aggregation flow

technical field

The invention relates to a traffic regulation method for guaranteeing the fairness of "assured forwarding (AF) service" convergence flow. Specifically, it relates to a traffic regulation method for ensuring the fairness of "AF service" convergence flow based on congestion charging in a differentiated service network, and belongs to the technical field of traffic control and congestion control in the Internet.

Background technique

According to different billing methods, Internet billing is divided into two categories: static billing and dynamic billing. Currently, the most commonly used billing method on the Internet is static flat billing. Flat billing collects usage fees from users by timing. The rate is fixed and depends on the user's access rate. The rate of dynamic billing is dynamically changed according to network conditions. Congestion pricing is a typical dynamic charging method. The charging price is dynamically adjusted according to the current network congestion status and fed back to users in real time. Congestion charging indirectly informs users of the network congestion status through feedback fees, and the more congested the network is, the higher the price will be. In order for the user to adjust the sending rate according to the changing price, when the cost exceeds the user's expectation, the user can reduce the total cost by reducing the sending rate, thus effectively supporting the congestion control and flow management of the network. At this time, billing is no longer simply a way to recover network investment costs, but also a technical means to control traffic.

The traffic control technology based on congestion charging usually establishes a market model first, abstracting network resources (mainly considering bandwidth resources) as commodities in the market, and abstracting service sources (users) as commodity buyers, commodity buyers ( Network users) need to pay a certain fee when purchasing goods (that is, consuming network bandwidth resources). Firstly, a utility function model is established to represent the user's satisfaction when consuming resources, that is, the user's bandwidth demand is expressed in the form of a utility function. Driven by the maximization of benefits, the sending rate of user business flows will be continuously adjusted according to the feedback cost, and finally reach the equilibrium state of system optimization. Research shows that traffic optimization control technology based on congestion charging can guarantee the proportional fairness of resource sharing among users.

The meaning of ensuring the proportional fairness of the AF service aggregation flow in the differentiated service (DiffServ) network is: when the network is underloaded, all AF service aggregation flows should share the remaining bandwidth share proportional to the target rate, that is, the target rate is the same , the remaining bandwidth share is also the same; when the network is overloaded, all AF service aggregation flows should experience a throughput drop proportional to the target rate, that is, the same target rate, the same throughput drop. Therefore, proportional fairness is essentially adjusting the bandwidth of each AF service aggregation flow in proportion to the target rate.

Differentiated Services (DiffServ) is a system framework for supporting Quality of Service (QoS) in the Internet proposed by the Internet Engineering Task Force IETF. Compared with the resource reservation of Integrated Services (IntServ), DiffServ adopts a priority-based strategy to ensure Quality of service for users. In the DiffServ system, the concept of domain is introduced, and the set of routers supporting the DiffServ function of the same rule is called a DiffServ domain, and the DiffServ domain is composed of border routers and intermediate routers. In DiffServ networks, a series of complex mechanisms related to QoS guarantees are pushed to the edge of the DiffServ domain. Edge routers need to complete traffic conditioning (Traffic Conditioning) functions, including mechanisms such as metering, shaping and marking, while core routers only need to complete corresponding packet scheduling and discarding policies.

In the DiffServ domain, all packets entering the domain are divided into several different levels at the edge of the domain, and are identified by the Differentiated Services Code Point (DSCP) field in the packet, and the network provides information to the packet according to the DSCP field of each packet. required services. At present, in the DiffServ system, three types of service levels are defined: accelerated forwarding service (EF): requires low delay and delay jitter, generally corresponding to real-time services; guaranteed forwarding service (AF): requires low Packet loss rate, and there is a certain demand on bandwidth; default best-effort forwarding (BE) service: traditional best-effort IP packet forwarding. In the DiffServ system, EF services have the highest priority, followed by AF services and BE services.

In the differentiated services architecture, the traffic control function acts on the edge of the DS domain to adjust the incoming or outgoing traffic. It can be regarded as a traffic regulator in the edge router, including mechanisms such as metering, marking, and shaping. The circuit structure of the flow regulator is shown in Figure 1. In the edge router, the data packet first passes through the classifier to determine the type of service, so as to determine which type of traffic specification the traffic conditioner adopts for regulation. The meter measures the flow characteristic of the service flow according to the flow specification, and judges whether the data packet meets the flow specification. Then according to the judgment result, the marker, the shaper and the packet discarder will respectively adopt corresponding policies to process the data packet. According to the judgment result of the meter, the marker can set a DSCP value corresponding to its service type for the data packet. A shaper adjusts traffic flow to conform to traffic specifications by delaying one or more packets. Because the buffer in the shaper usually has a limited capacity, if the delayed traffic exceeds the buffer capacity, the packet dropper will discard some data packets according to a certain strategy. Generally speaking, packets that are inconsistent with the flow specification are mostly discarded.

Because the metering and marking functions of BE and EF services are relatively simple, most marking mechanisms are mainly for Assured Forwarding (AF) services. The current marking mechanism can be divided into three types: algorithm based on token bucket, algorithm based on rate and algorithm based on policy. The representative algorithm based on the token bucket is the IN/OUT fair marker (Fair Marker) proposed by Hyogon Kim in 1999 and the three-color marker (TCM) proposed by J.Heinanen and its extended algorithm: single-rate three-color marker (srTCM) and two-rate three-color marker (trTCM). The rate-based representative algorithm is the Time Sliding Window Tricolor Marker (TSWTCM) proposed by W. Fang in 2000. The representative algorithm based on policy is TCP Friendly Traffic Marker (TFTM) proposed by F.Azeem in 1999. Among them, the two-rate three-color marker (trTCM) is currently the most commonly used traffic adjustment algorithm, which adjusts the sending rate of each AF service flow by adjusting the peak rate PIR parameter of the token bucket. The following briefly introduces the trTCM algorithm:

trTCM has four parameters: peak rate PIR (Peak Information Rate), peak burst size PBS (Peak Burst Size), target rate CIR (Committed Information Rate) and committed burst size CBS (Committed Burst Size). The unit of PIR and CIR is Byte/s, and the unit of PBS and CBS is Byte. If the packet exceeds the PIR, it will be marked as red, if it exceeds the CIR but not exceed the PIR, it will be marked as yellow, otherwise it will be marked as green. In a DiffServ network, red, yellow, and green are mapped to different DSCP values.

trTCM has two token buckets P and C, whose token generation rates are PIR and CIR respectively. The capacity of bucket P is PBS, and the capacity of bucket C is CBS. At the beginning, the token buckets P and C are both full, that is, the number of tokens Tp(0)=PBS, Tc(0)=CBS. Tp increases by 1 PIR byte per second until Tp=PBS; Tc increases by 1 CIR byte per second until Tc=CBS.

When a packet of size B bits arrives, the trTCM in Color-Blind mode operates as follows:

If Tp(t)-B<0, the group is marked in red, otherwise:

If Tc(t)-B<0, the packet is marked yellow, and Tp is reduced by B bits;

If Tc(t)-B≥0, the packet is marked green and both Tp and Tc are decreased by B bits.

A large number of studies have shown that the traffic adjustment mechanism in the current DiffServ network system is difficult to ensure the fairness among AF service aggregation flows. There are response flow and non-response flow in the AF service aggregation flow. The response flow is an application flow with end-to-end congestion control similar to the Transmission Control Protocol TCP (Transmission Control Protocol); the non-response flow does not use end-to-end congestion control Application flow, typically such as User Datagram Protocol UDP (User Data Protocol) flow, especially the application flow that will not reduce the sending rate even after packet loss. Obviously, non-response flows will take up more resources, which is unfair to response flows; similarly, response flows with different attributes will also cause unfair consequences when competing for network resources. Therefore, the fairness between the AF service aggregation flows is mainly reflected in two aspects: one is the fairness between the response flow and the non-response flow, which is caused by the different processing methods of the two for the congestion signal; The fairness problem between response flows, there is unfair resource allocation between response flows with different attributes, these attributes can be summarized as: (1) loopback delay, (2) the number of microflows in the aggregated flow, (3 ) the size of the target rate, (4) the packet length.

The three marking methods currently used: token bucket-based algorithm, rate-based algorithm, and policy-based algorithm cannot guarantee the proportional fairness of resource sharing between AF services, and token bucket-based algorithms are more commonly used. Recently, many researchers have conducted a lot of research on the defect that the fairness of AF services in DiffServ cannot be guaranteed, and obtained some improved algorithms. For example, the DCCS algorithm introduced in the article "Direct Congestion Control Scheme (DCCS) for Differentiated Services IP Networks" (see IEEE GLOBECOM'2001) is a representative traffic adjustment algorithm based on TCP-friendly control. Its characteristics are: first calculate the maximum throughput that can be achieved by the TCP aggregation flow in the AF aggregation flow, and limit other non-responsive flows (such as UDP flow) according to this throughput, so as to achieve a certain goal of fairness guarantee. This kind of DiffServ traffic adjustment algorithm based on TCP friendly control is very simple, and can improve the fairness of AF business aggregation flow to a certain extent, but this method has a characteristic: by suppressing strong competition (such as non-response flow, number of connections There are relatively many TCP flows, etc.) to protect flows with relatively weak competition ability, so as to ensure fairness. This method is feasible in the traditional Internet network, but in the DiffServ network where the target rate is introduced, there is a problem: the fairness proportional to the target rate cannot be guaranteed.

The Internet optimization flow control technology based on congestion charging can provide proportional fairness among different users. Therefore, if the flow control technology based on congestion charging is applied to the flow adjustment mechanism of DiffServ to provide resource allocation proportional to the target rate for the AF service aggregation flow, it should be able to achieve the goal of improving the fairness of the AF aggregation flow in the DiffServ network.

According to the applicant's search, no mechanism has been found in the previous literature to use congestion charging for flow control and congestion control in DiffServ networks; and the existing DiffServ flow adjustment mechanism cannot guarantee the proportional fairness of AF converged flows , that is to ensure that the AF aggregation flow obtains a bandwidth share proportional to its target rate.

Contents of the invention

The purpose of the present invention is to provide a traffic regulation method based on congestion charging to ensure the fairness of "AF service" converged flow in the differentiated services network, which solves the problem that the flow regulation algorithm in the existing differentiated services network cannot guarantee the converged flow of the AF service The defect of realizing the fairness proportional to the target rate realizes the proportional fairness of the shared bandwidth of the AF service aggregation flow, and improves the overall throughput of the system at the same time.

The object of the present invention is achieved in this way: a traffic regulation method based on congestion charging to ensure the fairness of "AF service" convergence flow in a differentiated service network, characterized in that: the method will be used in the Internet to ensure the proportion of users The fair flow control method based on congestion charging is applied to the differentiated service network: first establish the traffic regulation mathematical model of the AF aggregation flow based on congestion charging in the differentiated service network, and send the probe packet in the edge router of the differentiated service network The congestion cost of the current network is obtained, and then the optimal sending rate of each AF aggregation flow is calculated according to the congestion expense information and the principle of net utility maximization, and the token bucket parameters of each AF aggregation flow are adjusted according to the optimized sending rate to ensure that each The AF aggregation flow obtains a bandwidth sharing share approximately proportional to its target rate, and realizes the proportional fairness of resource sharing among AF aggregation flows.

Including the following steps:

(1) According to the service level agreement between the user and the network, establish a mathematical model based on congestion charging to realize AF aggregation flow flow regulation;

(2) The detection packet of each AF aggregation flow is sent by the ingress edge router, and stamped with a time stamp; when the detection packet reaches the corresponding egress router, the egress router immediately returns the detection packet to the ingress router, and the ingress router returns the detection packet to the ingress router according to the delay Calculate congestion charges;

(3) The edge router calculates the optimal transmission rate of each AF aggregation flow under the current congestion fee according to the congestion fee information and the principle of maximizing the user's net utility, that is, the transmission rate that increases the user's net utility;

(4) Adjust the token generation rate PIR of the token bucket P of each AF aggregated flow according to the optimized sending rate; then return to step (2) to perform the loop operation of flow adjustment.

Described step (1) further comprises following operation steps:

(11) Define the set of AF converged flow users S={s ₁ , s ₂ ...}, where s is each AF converged flow user; and define the following mathematical model for each AF converged flow user s:

The utility function of user s is U _s =w _s logx _s ; where x _s is the current sending rate of user s; w _s is the coefficient of the utility function of user s, and the value of w _s is set as the target rate of user s. The target rate is determined according to the service level agreement SLA;

The net utility function of user s is J _s (x _s )=U _s (x _s )-x _s p ^s ; where p ^s is the fee that user s needs to pay for consuming unit bandwidth, that is, x _s p ^s is user s is the total fee to be paid for its AF convergence flow;

(12) Set the token generation rate CIR of the token bucket C to the target rate of the AF convergence flow, and set the token generation rate PIR of the token bucket P to twice the CIR.

Described step (2) further comprises following operation steps:

(21) According to the priority characteristic of each AF aggregation flow, that is, the DSCP value, a detection packet with only the header is generated. The DSCP value of the detection packet is the same as the DSCP value of the AF aggregation flow, and is sent to each pair of ingress routers and egress routers. The detection packet is sent multiple times according to the time period T;

(22) Measure the one-way queuing delay q _s of each AF aggregation flow user s passing through the DiffServ domain: q _s =[ _rs -min(rs ₎ ]/2, where _rs is the detection packet between the ingress router and The round-trip delay between egress routers, min( _rs ) is the minimum value of _rs , which is the approximate value of the sum of propagation delay and processing delay when there is no queuing;

(23) According to the one-way queuing delay q _s value, calculate the unit bandwidth congestion charge p ^s =γ·q _s for the AF converged flow user s passing through the DiffServ domain, which is a linear function of the one-way queuing delay q _s , and the coefficient γ The experimental experience value of γ is 0.01, and the larger the γ, the greater the impact of network congestion charges on users, but it will not affect the fairness among AF converged flows; then calculate the congestion charge that user s of AF converged flows needs to pay: x _s ·p ^s =γ·x _s ·q _s , where x _s is the current sending rate of user s.

Described step (3) further comprises following operation steps:

(31) Utilize the metering function of the flow regulator to measure and calculate the current sending rate x _s of the AF convergence flow user s;

(32) Calculate the net utility function value J _s (x _s )=U _s (x _s )-x _s p ^s of the AF converged flow user s under the current sending rate x _s and congestion cost, and its first derivative: ${\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)=\frac{\&PartialD;J_{\mathrm{the\; s}}}{\&PartialD;x_{\mathrm{the\; s}}}=\frac{w_{\mathrm{the\; s}}}{x_{\mathrm{the\; s}}}-p^{\mathrm{the\; s}};$

；如果 ${\stackrel{\&CenterDot;}{J}}_s\left(x_s\right)\&le;0$ ，说明净效用函数值J_s(x_s)会随着x_s的减小而增大，则用户减小发送速率： $x_s^{\mathrm{opt}}=x_s-\mathrm{\&beta;}\&CenterDot;\left|{\stackrel{\&CenterDot;}{J}}_s\left(x_s\right)\right|$ ，式中x_s为用户s的当前发送速率，减小幅度为

：上述两式中的α和β均为调整系数，且0＜α≤β，即增加发送速率时增加的幅度小，而在减小发送速率时减小的幅度大，以便网络在拥塞时候能够迅速缓解拥塞，不至于引起网络状态的震荡，有利于保持网络稳定。(33) According to the principle of net utility maximization, calculate the optimal transmission rate x _s ^opt of the AF convergence flow under the current congestion cost, if ${\stackrel{\&Center\; Dot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)>0$ , indicating that the net utility function value J _s (x _s ) will increase as x _s increases, so the user increases the sending rate: $x_{\mathrm{the\; s}}^{\mathrm{opt}}=x_{\mathrm{the\; s}}+\mathrm{\&alpha;}\&Center\; Dot;{\stackrel{\&Center\; Dot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)$ , where x _s is the current sending rate of user s, and the increase range is

;if ${\stackrel{\&Center\; Dot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)\&le;0$ , indicating that the net utility function value J _s (x _s ) will increase as x _s decreases, and the user reduces the sending rate: $x_{\mathrm{the\; s}}^{\mathrm{opt}}=x_{\mathrm{the\; s}}-\mathrm{\&beta;}\&CenterDot;\left|{\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)\right|$ , where x _s is the current sending rate of user s, and the reduction range is

: α and β in the above two formulas are adjustment coefficients, and 0<α≤β, that is, the increase range is small when the transmission rate is increased, and the decrease range is large when the transmission rate is reduced, so that the network can Quickly relieve congestion without causing fluctuations in network status, which is conducive to maintaining network stability.

The step (4) includes the following steps: in each rate adjustment period, the token generation rate PIR of the token bucket P of each AF aggregated flow is adjusted to its current optimal sending rate x _opt , namely $\mathrm{PIR}=x_{\mathrm{the\; s}}^{\mathrm{opt}}$ , and compare the values of PIR and CIR to ensure that PIR is not less than the value of CIR, that is, PIR=Max(CIR, PIR); then return to step (2), and enter the next loop operation of flow adjustment period.

The principle of maximizing the net utility is to seek the maximum possible value of the so-called net utility value obtained by subtracting the user's utility from the fee that the user needs to pay for this utility.

The invention is a traffic regulation method for ensuring the fairness of "AF service" convergence flow based on congestion charging in a differentiated service network, which applies the flow control method based on congestion charging in the Internet to the differentiated service network. Compared with the traditional traffic regulation algorithm (represented by trTCM) and the improved algorithm (represented by DCCS algorithm), the method of the present invention solves the defects of the traffic regulation algorithms in the existing various DiffServ networks well, and can Guarantee the proportional fairness of the AF aggregation flow; and improve the fairness of the bandwidth shared by the AF aggregation flow according to the target rate, and at the same time improve the overall throughput of the system, so that network users can obtain higher cost performance.

The traffic regulation method of the present invention belongs to the algorithm based on the token bucket, which utilizes a standard two-rate three-color marker (trTCM) to adjust the sending rate of each AF aggregation flow by adjusting the PIR parameter of the token bucket. Due to the dynamic adjustment of congestion charging, the more congested the network is, the higher the congestion fee will be. In order to achieve the goal of maximizing the net utility, each AF aggregation flow will reduce the flow, that is, reduce its sending rate; and if the network bandwidth margin is higher The larger the congestion cost, the lower the congestion cost. In order to achieve the goal of maximizing the net utility, each AF aggregation flow will increase the flow, that is, increase its sending rate; A bandwidth sharing share that is approximately proportional to its target rate, so as to achieve the goal of ensuring proportional fairness between each AF aggregation flow.

Description of drawings

Fig. 1 is a logical structural diagram of a traffic conditioner in a differentiated services network.

Fig. 2 is an overall flowchart of the operation steps of the flow regulating method of the present invention.

Fig. 3 is a flow chart of the operation steps of measuring the congestion charge of the AF converged flow in the flow regulation method of the present invention.

Fig. 4 is a flow chart of the operation steps of calculating the optimal transmission rate of the AF aggregated flow in the flow adjustment method of the present invention.

Detailed ways

The working mechanism and specific operation steps of the method of the present invention will be described in detail below in conjunction with the accompanying drawings.

The present invention is a traffic adjustment method based on congestion charging to guarantee the fairness of "AF service" convergence flow in DiffServ network. Its working mechanism is to abstract DiffServ network resources into commodities in the market, and The AF aggregated flow of resources is regarded as users in the market, and the form of logarithmic utility function is used to express the satisfaction degree obtained by users consuming bandwidth resources. Users who consume more resources will get higher satisfaction, and at the same time, each user must pay the corresponding fee for the network resources it consumes. The network here adopts a congestion charging strategy, that is, the cost per unit resource is related to the degree of network congestion. The more congested the network is, the higher the cost will be. When the network is not congested, the cost will be zero. The edge routers of the differentiated services network know the congestion cost of the current network by sending probe packets to each other, and dynamically adjust the sending rate of each AF aggregation flow in real time according to the congestion cost information, so as to maximize the net utility of each AF aggregation flow change.

Therefore, the traffic regulation method of the present invention is mainly based on the principle that the congestion charging applied in the Internet can guarantee the fairness of the user ratio and by changing the PIR of the token bucket in the three-color marker (trTCM) based on the token bucket Parameters to adjust the sending rate of each AF aggregation flow: first establish a traffic regulation mathematical model of AF aggregation flow based on congestion charging in the DiffServ network, and periodically send detection packets between the edge routers of the DiffServ network to obtain the current The network congestion fee, and then calculate the optimal sending rate of each AF aggregation flow according to the congestion fee information, the target rate of each AF aggregation flow and the principle of net utility maximization, and adjust the token bucket of each AF aggregation flow according to the optimized sending rate Parameters to ensure that each AF aggregation flow obtains a bandwidth sharing share approximately proportional to its target rate, and achieve proportional fairness in resource sharing among AF aggregation flows. At the same time, the overall transmission efficiency of the DiffServ network is improved, and the overall effective throughput of the network is improved.

Referring to Fig. 2, specifically illustrate four main operation steps of the flow regulation method of the present invention:

(1) According to the service level agreement SLA between the user and the network, establish a mathematical model based on congestion charging to realize the flow adjustment of the AF service aggregation flow. This step mainly includes two operations:

(11) Define the set of AF converged flow users S={s ₁ , s ₂ ...}, where s is each AF converged flow user; and define the following mathematical model for each AF converged flow user s:

The utility function of user s is U _s =w _s logx _s ; where x _s is the current sending rate of user s; w _s is the coefficient of the utility function of user s, and the value of w _s is set as the target rate of user s. The target rate is determined according to the service level agreement SLA;

The net utility function of user s is J _s (x _s )=U _s (x _s )-x _s p ^s ; where p ^s is the fee that user s needs to pay for consuming unit bandwidth, that is, x _s p ^s is user s is the total fee to be paid for its AF convergence flow;

(12) Set the token generation rate CIR of the token bucket C to the target rate of the AF convergence flow, and set the token generation rate PIR of the token bucket P to twice the CIR.

(2) Measure the congestion cost information of each AF aggregation flow: send the detection packet of each AF aggregation flow at the ingress edge router, and stamp it with a time stamp; when the detection packet reaches the corresponding exit router, the exit router returns the detection packet immediately To the ingress router, the ingress router calculates the congestion charge based on the delay.

(3) The edge router calculates the optimal transmission rate of each AF aggregation flow under the current congestion fee according to the congestion fee information and the principle of maximizing the user's net utility, that is, the transmission rate that increases the user's net utility;

(4) According to the optimized transmission rate, adjust the token generation rate PIR of the token bucket P of each AF aggregation flow to its current optimal transmission rate x _opt in each rate adjustment period, that is, $\mathrm{PIR}=x_{\mathrm{the\; s}}^{\mathrm{opt}}$ , and compare the values of PIR and CIR to ensure that PIR is not less than the value of CIR, that is, PIR=Max(CIR, PIR); and then return to step (2) to enter the cycle operation of the next flow regulation cycle.

Referring to Fig. 3, the flow chart of each refinement step in step (2) measuring network congestion situation in detail:

(21) According to the priority characteristic of each AF aggregation flow, that is, the DSCP value, a detection packet with only the header is generated. The DSCP value of the detection packet is the same as the DSCP value of the AF aggregation flow, and is sent to each pair of ingress routers and egress routers. The detection packet is sent multiple times according to the time period T;

(22) Measure the round-trip delay r s and the one-way queuing delay q _{s of the AF aggregation flow user s passing through the DiffServ domain by sending a probe packet between the ingress router and the egress router, q s =} [rs _-min (r _s )]/2, where min( _rs ) is the minimum value of _rs , which is the approximate value of the sum of propagation delay and processing delay when not queuing;

(23) According to the one-way queuing delay q _s value, calculate the unit bandwidth congestion charge p ^s =γ·q _s for the AF converged flow user s passing through the DiffServ domain, which is a linear function of the one-way queuing delay q _s , and the coefficient γ The experimental experience value of γ is 0.01, and the larger the γ, the greater the impact of network congestion charges on users, but it will not affect the fairness among AF converged flows; then calculate the congestion charge that user s of AF converged flows needs to pay: x _s ·p ^s =γ·x _s ·q _s , where x _s is the current sending rate of user s.

Referring to Fig. 4, describe in detail the flow chart of each refinement step of step (3) calculation optimization rate:

(31) Utilize the metering function of the flow regulator to measure and calculate the current sending rate x _s of the AF convergence flow user s;

(32) Calculate the net utility function value J _s (x _s )=U _s (x _s )-x _s p ^s of the AF converged flow user s under the current sending rate x _s and congestion cost, and its first derivative: ${\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)=\frac{\&PartialD;J_{\mathrm{the\; s}}}{\&PartialD;x_{\mathrm{the\; s}}}=\frac{w_{\mathrm{the\; s}}}{x_{\mathrm{the\; s}}}-p^{\mathrm{the\; s}};$

(33) According to the principle of net utility maximization, calculate the optimal sending rate of the AF aggregation flow under the current congestion cost , where x _s is the current sending rate of user s, α and β are adjustment coefficients, and 0<α≤β; that is, if ${\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)>0$ , indicating that the net utility function value J _s (x _s ) will increase with the increase of x _s , so the user increases the sending rate, and the increase range is ;if ${\stackrel{\&Center\; Dot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)\&le;0,$ It shows that the net utility function value J _s (x _s ) will increase with the decrease of x _s , then the user reduces the sending rate, and the reduction range is ; When adjusting the traffic according to the above two formulas, the increase range is small when the transmission rate is increased, and the decrease range is large when the transmission rate is reduced, so that the network can quickly relieve congestion when the network is congested, so as not to cause the network state to fluctuate. Helps maintain network stability.

The traffic regulation method of the present invention has been carried out simulation test on the computer, to five kinds of situations (between response flow and non-response flow and between the response flow of four different attributes) that affect the fairness of "AF business" convergent flow Simulations have been carried out, and the method of the present invention has been compared with trTCM and the DCCS algorithm belonging to the TCP friendly control mechanism. The simulation results show that the traffic regulation method of the present invention is more effective in improving fairness and improving total effective throughput. Good performance, the test is successful, and the purpose of the invention has been realized.

The test process of improving the fairness of AF service aggregation flows with different target rates (CIR) will be briefly described below.

There are 10 TCP aggregation flows in the simulation, and the target rate (CIR) of each aggregation flow is 0.1Mbps, 0.2Mbps, 0.3Mbps, 0.4Mbps, 0.5Mbps, 0.6Mbps, 0.7Mbps, 0.8Mbps, 0.9Mbps and 1.0Mbps respectively , so the total target rate is 5.5Mbps. The bottleneck link bandwidth of the simulated network topology is set to 11Mbps, so the load is 50%.

In order to evaluate the fairness between AF aggregation flows, a fairness index (FI) is defined here:

$\mathrm{FI}={\left(\underset{i=1}{\overset{\mathrm{no}}{\mathrm{\&Sigma;}}}{\mathrm{\&eta;}}_i\right)}^2/\mathrm{no}\&times;\underset{i=1}{\overset{\mathrm{no}}{\mathrm{\&Sigma;}}}{\mathrm{\&eta;}}_i^2$ , where n represents the number of AF converged flow users, η _i = _xi / _ci represents the ratio of the actual throughput of the ith converged flow to the target rate, x _i represents the average effective throughput of the ith converged flow, c _i represents the target rate (CIR) of the i-th converged flow. The fairness index is a non-negative number between 0 and 1. The closer the value is to 1, the better the fairness. Ideally, each AF aggregation flow will obtain a bandwidth share proportional to its target rate, that is, 0.2Mbps, 0.4Mbps, 0.6Mbps, 0.8Mbps, 1.0Mbps, 1.2Mbps, 1.4Mbps, 1.6Mbps, 1.8Mbps and 2.0Mbps.

The simulation experiment compares the traffic regulation method of the present invention with trTCM and the DCCS algorithm belonging to the TCP friendly control mechanism respectively, the fairness index of result trTCM is 0.74, the fairness index of DCCS is 0.82, and the fairness index of the method of the present invention It is 0.96, which is close to the ideal state. This is because the traffic adjustment method of the present invention is different for the traffic adjustment of AF converged flows with different CIRs, and the CIR is used as the coefficient of the utility function of each converged flow, so the fairness target proportional to the CIR can be guaranteed.

Claims (7)

1. A traffic regulation method based on congestion charging to ensure fairness of "AF service" converged flow in a differentiated service network, characterized in that: the method is based on congestion charging used in the Internet to ensure fairness of user ratio The flow control method of DiffServ network is applied to DiffServ network: first establish the mathematical model of traffic regulation of AF aggregation flow based on congestion charging in DiffServ network, and send detection packets between edge routers of DiffServ network to obtain the current network congestion According to the congestion cost information and the principle of net utility maximization, the optimized sending rate of each AF aggregated flow is calculated, and the token bucket parameters of each AF aggregated flow are adjusted according to the optimized sending rate to ensure that each AF aggregated flow achieves its goal The rate is approximately proportional to the bandwidth sharing share, which realizes the proportional fairness of resource sharing between AF aggregation flows. 2. The flow regulating method according to claim 1, characterized in that it comprises the following steps: (1) According to the service level agreement between the user and the network, establish a mathematical model based on congestion charging to realize AF aggregation flow flow regulation; (2) The detection packet of each AF aggregation flow is sent by the ingress edge router, and stamped with a time stamp; when the detection packet reaches the corresponding egress router, the egress router immediately returns the detection packet to the ingress router, and the ingress router returns the detection packet to the ingress router according to the delay Calculate congestion charges; (3) The edge router calculates the optimal transmission rate of each AF aggregation flow under the current congestion fee according to the congestion fee information and the principle of maximizing the user's net utility, that is, the transmission rate that increases the user's net utility; (4) Adjust the token generation rate PIR of the token bucket P of each AF aggregated flow according to the optimized sending rate; then return to step (2) to perform the loop operation of flow adjustment. 3. The flow regulating method according to claim 2, characterized in that: said step (1) further comprises the following steps: (11) Define the set of AF converged flow users S={s ₁ , s ₂ ...}, where s is each AF converged flow user; and define the following mathematical model for each AF converged flow user s: The utility function of user s is U _s =w _s logx _s ; where x _s is the current sending rate of user s; w _s is the coefficient of the utility function of user s, and the value of w _s is set as the target rate of user s. The target rate is determined according to the service level agreement SLA; The net utility function of user s is J _s (x _s )=U _s (x _s )-x _s p ^s ; where p ^s is the fee that user s needs to pay for consuming unit bandwidth, that is, x _s p ^s is user s is the total fee to be paid for its AF convergence flow; (12) Set the token generation rate CIR of the token bucket C to the target rate of the AF convergence flow, and set the token generation rate PIR of the token bucket P to twice the CIR. 4. The flow regulating method according to claim 2, characterized in that: said step (2) further comprises the following steps: (21) According to the priority characteristic of each AF aggregation flow, that is, the DSCP value, a detection packet with only the header is generated. The DSCP value of the detection packet is the same as the DSCP value of the AF aggregation flow, and is sent to each pair of ingress routers and egress routers. The detection packet is sent multiple times according to the time period T; (22) Measure the one-way queuing delay q _s of each AF aggregation flow user s passing through the DiffServ domain: q _s =[ _rs -min(rs ₎ ]/2, where rs _is the detection packet between the ingress router and The round-trip delay between egress routers, min( _rs ) is the minimum value of _rs , which is the approximate value of the sum of propagation delay and processing delay when there is no queuing; (23) According to the one-way queuing delay q _s value, calculate the unit bandwidth congestion charge p ^s =γ·q _s for the AF converged flow user s passing through the DiffServ domain, which is a linear function of the one-way queuing delay q _s , and the coefficient γ The experimental experience value of γ is 0.01, and the larger the γ, the greater the impact of network congestion charges on users, but it will not affect the fairness among AF converged flows; then calculate the congestion charge that user s of AF converged flows needs to pay: x _s ·p ^s =γ·x _s ·q _s , where x _s is the current sending rate of user s. 5. The flow regulating method according to claim 2, characterized in that: said step (3) further comprises the following steps: (31) Utilize the metering function of the flow regulator to measure and calculate the current sending rate x _s of the AF convergence flow user s; (32) Calculate the net utility function value J _s (x _s )=U _s (x _s )-x _s p ^s of the AF converged flow user s under the current sending rate x _s and congestion cost, and its first derivative: ${\stackrel{\&Center\; Dot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)=\frac{{\&PartialD;J}_{\mathrm{the\; s}}}{{\&PartialD;x}_{\mathrm{the\; s}}}=\frac{w_{\mathrm{the\; s}}}{x_{\mathrm{the\; s}}}-p^{\mathrm{the\; s}};$ (33) According to the principle of net utility maximization, calculate the optimal transmission rate x _s ^opt of the AF convergence flow under the current congestion cost, if ${\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)>0,$ It shows that the net utility function value J _s (x _s ) will increase with the increase of x _s , so the user increases the sending rate: $x_{\mathrm{the\; s}}^{\mathrm{opt}}=x_{\mathrm{the\; s}}+\mathrm{\&alpha;}\&CenterDot;{\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right),$ where x _s is the current sending rate of user s, and the increase range is $\mathrm{\&alpha;}\&CenterDot;{\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right);$ if ${\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)\&le;0,$ It shows that the net utility function value J _s (x _s ) will increase as x _s decreases, and the user reduces the sending rate: $x_{\mathrm{the\; s}}^{\mathrm{opt}}=x_{\mathrm{the\; s}}-\mathrm{\&beta;}\&Center\; Dot;\left|{\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)\right|,$ where x _s is the current sending rate of user s, and the reduction range is $\mathrm{\&beta;}\&CenterDot;\left|{\stackrel{\&CenterDot;}{J}}_{\mathrm{the\; s}}\left(x_{\mathrm{the\; s}}\right)\right|;$ α and β in the above two formulas are adjustment coefficients, and 0<α≤β, that is, the increase range is small when the transmission rate is increased, and the decrease range is large when the transmission rate is reduced, so that the network can quickly recover when the network is congested. Alleviating congestion will not cause fluctuations in network status, which is conducive to maintaining network stability. 6. The flow regulation method according to claim 2, characterized in that: said step (4) includes the following operation steps: in each rate adjustment cycle, the token generation rate of the token bucket P of each AF converged flow PIR adjusts to its current optimal sending rate x _opt , ie $\mathrm{PIR}=x_{\mathrm{the\; s}}^{\mathrm{opt}},$ And compare the values of PIR and CIR to ensure that PIR is not less than the value of CIR, that is, PIR=Max(CIR, PIR); then return to step (2), and enter the cycle operation of the next flow adjustment cycle. 7. The traffic regulation method according to claim 1 or 5, characterized in that: the net utility maximization principle is the so-called net utility value obtained by subtracting the user's utility from the fee that the user needs to pay for this utility. Seek the maximum possible.

Images