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

Abstract

The invention applies flow control method based on congestion charging utilized in Internet for guaranteeing proportion fairness for users to distinguished service networks. The method includes steps: building flow control mathematical model of AF concourse flow based on congestion charging in distinguished service networks; detecting packets are sent between edge routers in distinguished service networks in order to obtain congestion fee of current network; based on information of congestion fee and principle of maximizing net utility, calculating optimal transmission rate of AF concourse flow; based on optimal transmission rate, adjusting proportion fairness of shared resources between concourse flows. Comparing with prior art, the invention realizes proportion fairness of target speed in AF concourse flow.

Description

The flow rate adjusting method of convergence flow fairness that a kind of guarantee " guarantees to transmit business "

Technical field

The present invention relates to the flow rate adjusting method of a kind of guarantee " guarantee transmit (AF) business " convergence flow fairness.Exactly, relate to a kind of flow rate adjusting method that in differentiated services network, ensures " AF business " convergence flow fairness, belong to the technical field of flow control and congested control in the Internet based on congestion pricing.

Background technology

According to the difference of charging way, the charging of Internet is divided into two big classes: the static charging and dynamically charging.The most frequently used charging way of current Internet is that static flat (flat) charges.Flat charges and by timing mode the user is collected cost of use, and its rate is fixed, and is decided by user's access rate.And the rate of dynamically chargeing dynamically changes according to network condition.Congestion pricing (congestion pricing) is exactly a kind of typical dynamically charging way, and its charging rate is dynamically to adjust according to the current network congestion condition, and feeds back to the user in real time.Congestion pricing is informed the congestion condition of user network indirectly by the feedback expense, and the congested more then price of network is high more.So that the user adjusts transmission rate according to continuous variable prices, when expense surpassed the user and expects, the user reduced the total cost of paying by reducing transmission rate, thus the congested control and the traffic management of network enabled effectively.At this moment, chargeing is no longer merely a kind of way of recycling of network investment cost, also is a kind of technological means of controlling flow simultaneously.

Flow control technique based on congestion pricing; usually can set up a market model earlier; Internet resources (mainly considering bandwidth resources) are abstracted into the commodity in the market; and service source (user) is abstracted into commodity purchaser, commodity purchaser (network user) need be paid certain expense when buying commodity (being the consume network bandwidth resource).At first set up effectiveness (Utility) function model and represent the satisfaction of user when consumption of natural resource, the bandwidth demand of promptly representing the user with the utility function form, network based current congestion condition is constantly to the congested expense of user feedback, the user obtains under the maximized interests driving of net utility in hope, to constantly adjust the transmission rate of user traffic flow according to the expense of feedback, reach the equilibrium state of system optimization at last.Studies show that, can ensure the proportional fairness of resource-sharing between each user based on the flow optimization control technology of congestion pricing.

The implication that ensures the proportional fairness of AF service convergence stream in Differentiated Services (DiffServ) network is: when network is underload, all AF service convergence streams all should be shared and the proportional remaining bandwidth share of targeted rate, be that targeted rate is identical, the remaining bandwidth share that obtains is also identical; When network over loading, all AF service convergence stream all should experience with the proportional throughput of targeted rate and descend, and promptly targeted rate is identical, and its throughput decline degree is identical.Therefore, proportional fairness is exactly in fact the bandwidth of regulating each AF service convergence stream with targeted rate pro rata.

Differentiated Services (DiffServ) is a kind of System Framework of supporting service quality (QoS) in the Internet that internet engineering task group IETF proposes, with respect to the resource reservation of integrated service (IntServ), Differentiated Services adopts the strategy based on priority to guarantee quality of services for users.Introduced the notion in territory in the Differentiated Services system, the set of routers of the Differentiated Services function of supporting same rule is called a DiffServ territory, the DiffServ territory is made up of border router and intermediate router.In differentiated services network, guarantee that with QoS relevant a series of complex mechanism are pulled to the edge in DiffServ territory.Edge router need be finished the function of Flow-rate adjustment (Traffic Conditioning), comprising metering, and mechanism such as shaping and mark, core router then only need be finished corresponding packet scheduling and drop policy.

In the DiffServ territory, all edges, territory that are grouped in that enter the territory are divided into several different grades, and by differentiated Services code point (DSCP) field identification in the grouping, the dscp field of network based each grouping provides required service to this grouping.In the DiffServ system, defined three kinds of grade of service types at present: quicken to transmit professional (EF): requiring has low time delay and delay variation, generally corresponding to real time business; Guarantee to transmit professional (AF): requiring has lower packet loss, and certain demand is arranged on bandwidth; Default (BE) business of transmitting as possible: traditional IP packet forward of doing one's best.In the DiffServ system, the priority of EF business is the highest, and the AF business is taken second place, and the BE business is taken second place again.

In the Differentiated Services architecture, flow control function acts on edge, DS territory, and the Business Stream that flows into or flow out is adjusted, it can be regarded as a flow regulator in the edge router, comprising metering, mechanism such as mark and shaping, the circuit structure of flow regulator is as shown in Figure 1.In edge router, packet is judged the type of service earlier through grader, to determine that flow regulator adopts the traffic specification of which kind of type to regulate.Meter is according to the discharge characteristic of this traffic specification measurement Business Stream, and whether the judgment data bag satisfies traffic specification.According to judged result, marker, reshaper and packet loss device will adopt corresponding strategy to handle this packet respectively again.According to the judged result of meter, marker can be the DSCP value of packet setting corresponding to its COS.Reshaper is adjusted Business Stream by one or more packets of delaying time, and makes it consistent with traffic specification.Because the buffering area in the reshaper is generally limited capacity, if the flow of time-delay has surpassed buffer pool size, the packet loss device will abandon some packets according to certain strategy.What in general, abandon mostly is and the inconsistent packet of traffic specification.

Because the metering and the mark function of BE and EF business are fairly simple, most tagging mechanisms are mainly at guaranteeing that transmitting (AF) class business carries out.Present tagging mechanism can be divided into three kinds: based on the algorithm of token bucket, based on the algorithm of speed with based on the algorithm of strategy.Based on the representative algorithm of token bucket is fair marker (Fair Marker) of IN/OUT that was proposed by Hyogon Kim in 1999 and tricolor marker device (TCM) and the expansion algorithm thereof that is proposed by J.Heinanen: single-rate three-color marker device (srTCM) and dual rate tricolor marker device (trTCM).Representative algorithm based on speed is the tricolor marker device (TSWTCM) by the time sliding window of W.Fang proposition in 2000.Representative algorithm based on strategy is the flow marker (TFTM) by the TCP close friend of F.Azeem proposition in 1999.Wherein dual rate tricolor marker device (trTCM) is current Flow-rate adjustment algorithm the most commonly used, and it is the transmission rate of regulating each AF Business Stream by the peak rate PIR parameter of regulating token bucket.Simply introduce the trTCM algorithm below:

TrTCM has four parameters: peak rate PIR (Peak Information Rate), peak burst size PBS (Peak Burst Size), targeted 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.Grouping surpasses PIR and just is labeled as redness, has surpassed CIR and surpasses PIR and just be labeled as yellow, otherwise be labeled as green.In the DiffServ network, redly, yellow, green be mapped as different DSCP values.

TrTCM has two token bucket P and C, and its token produces speed and is respectively PIR and CIR.The capacity of P bucket is PBS, and the capacity of C bucket is CBS.Token bucket P and C are full during beginning, i.e. token number Tp (0)=PBS, Tc (0)=CBS.The Tp per second increases by the byte of 1 PIR, until Tp=PBS; The Tc per second increases by the byte of 1 CIR, until Tc=CBS.

When the grouping that is the B bit of a size arrived, trTCM operated by the following method under the Color-Blind pattern:

If Tp (t)-B＜0, group markups is designated as redness, otherwise:

If Tc (t)-B＜0, group markups is designated as yellow, and Tp reduces the B bit;

If Tc (t)-B 〉=0, group markups is designated as green, and Tp and Tc reduce the B bit.

Studies show that in a large number the Flow-rate adjustment mechanism in the present DiffServ network system is difficult to the fairness between the assurance AF service convergence stream.Have response flow and non-response flow in AF service convergence stream, response flow is the application stream with end-to-end congested control of similar transmission control protocol TCP (Transmission Control Protocol); Non-response flow is not adopt the application stream of end-to-end congested control, typically as User Datagram Protoco (UDP) UDP (User Data Protocol) stream, particularly also can not reduce the application stream of transmission rate behind the packet loss.Obvious non-response flow can take more resources, and this is inequitable for response flow; The response flow of same different attribute also can cause inequitable consequence in the competition network resource.Therefore, the fairness between the AF service convergence stream is mainly reflected in two aspects: the one, and the fairness problem between response flow and the non-response flow, this is because the two is to due to the different disposal mode of congestion signal; The 2nd, there is inequitable resource allocation in the fairness problem between the response flow between the response flow of different attribute, and these attributes can reduce: (1) loopback delay, the miniflow number in (2) convergence flow, the size of (3) targeted rate, (4) block length.

Present three kinds of labeling methods using: based on the algorithm of token bucket, all can not ensure the proportional fairness of resource-sharing between the AF business based on the algorithm of speed with based on the algorithm of strategy, wherein the algorithm based on token bucket is comparatively commonly used.Recently a lot of researchers have carried out big quantity research, and have drawn some improved algorithms at the defective that the professional fairness of AF in the Differentiated Services can not be protected.For example the DCCS algorithm of article " DirectCongestion Control Scheme (DCCS) for Differentiated Services IP Networks " (referring to IEEE GLOBECOM ' 2001) introduction is exactly a kind of representative Flow-rate adjustment algorithm based on the friendly control of TCP.Be characterized in: at first calculate the maximum throughput that the TCP convergence flow can reach in the AF convergence flow, and according to other non-response flow (as UDP stream) of this throughput constraints, thereby reach the target that certain fairness ensures.This DiffServ Flow-rate adjustment algorithm based on the friendly control of TCP is very simple; and can improve the fairness that the AF service convergence flows to a certain extent; but this method has characteristics: protect competitiveness more weak stream relatively by the stream that suppresses competitiveness strong (as the more relatively TCP stream of non-response flow, linking number etc.), thereby guarantee fairness.This method is feasible in traditional Internet network, but in the DiffServ network of introducing targeted rate, but has problems: can't ensure and the proportional fairness of targeted rate.

Internet based on congestion pricing optimizes flow control technique, and proportional fairness can be provided between different user.If therefore the flow control technique based on congestion pricing is applied in the Flow-rate adjustment mechanism of DiffServ, for AF service convergence stream provides and the proportional resource allocation of targeted rate, should be able to reach the target of improving AF convergence flow fairness in the DiffServ network.

According to applicant's retrieval, in document in the past, do not find to have the mechanism of in differentiated services network, utilizing congestion pricing to carry out flow control and congested control as yet; And existing DiffServ Flow-rate adjustment mechanism can not ensure the proportional fairness of AF convergence flow, ensures that promptly the AF convergence flow obtains and the proportional bandwidth share of its targeted rate.

Summary of the invention

The purpose of this invention is to provide a kind of in differentiated services network the flow rate adjusting method based on congestion pricing guarantee " AF business " convergence flow fairness, this method has solved Flow-rate adjustment algorithm in the existing differentiated services network can not ensure the defective that realizes between the AF service convergence stream with the proportional fairness of targeted rate, realize the proportional fairness of AF service convergence stream shared bandwidth, also improved the entire system throughput simultaneously.

The object of the present invention is achieved like this: a kind of in differentiated services network based on the flow rate adjusting method of congestion pricing guarantee " AF business " convergence flow fairness, it is characterized in that: this method is to ensure that with being used in the Internet flow control methods based on congestion pricing of user's proportional fairness is applied to differentiated services network: set up in the differentiated services network Flow-rate adjustment Mathematical Modeling based on the AF convergence flow of congestion pricing earlier, and in the edge router of differentiated services network, send detection packet and obtain the congested expense of current network, this congested cost information of basis and net utility maximization principle calculate the optimization transmission rate of each AF convergence flow again, and the token bucket parameters of adjusting each AF convergence flow according to the optimization transmission rate, ensure that each AF convergence flow obtains and the approximately proportional bandwidth sharing share of its targeted rate, realizes the proportional fairness of resource-sharing between the AF convergence flow.

Comprise following operating procedure:

(1) Mathematical Modeling that realizes AF convergence flow Flow-rate adjustment based on congestion pricing is set up in agreement according to the grade of service between user and the network;

(2) send the detection packet of each AF convergence flow at the ingress edge router, and stamp timestamp; After this detection packet arrived corresponding egress router, egress router turned back to ingress router with this detection packet immediately, by ingress router according to the congested expense of time-delay calculation;

(3) edge router calculates the optimization transmission rate of each AF convergence flow under the current congestion expense according to congested cost information and user's net utility maximization principle, even get the transmission rate that user's net utility increases;

(4) adjust the token generation speed PIR of the token bucket P of each AF convergence flow according to optimizing transmission rate; Return step (2) then, carry out the cycling of Flow-rate adjustment.

Described step (1) further comprises following operating procedure:

(11) definition AF convergence flow user S set={ s ₁, s ₂, s is each AF convergence flow user in the formula; And each AF convergence flow user s defined following Mathematical Modeling:

The utility function of user s is U _s=w _sLogx _sX in the formula _sCurrent transmission rate for user s; w _sBe the coefficient of the utility function of user s, w _sNumerical value be set to the targeted rate of user s, this targeted rate is determined according to grade of service agreement SLA;

The net utility function of user s is J _s(x _s)=U _s(x _s)-x _sP ^sP in the formula ^sFor user s consumes the expense that unit bandwidth need be paid, i.e. x _sP ^sThe total cost that need pay for its AF convergence flow for user s;

(12) token of token bucket C produces the targeted rate that speed CIR is set to the AF convergence flow, and the token of token bucket P produces speed PIR and is set to 2 times CIR.

Described step (2) further comprises following operating procedure:

(21) according to the priority feature of each AF convergence flow, it is the DSCP value, generation has only the detection packet in packet header, and the DSCP value of this detection packet is identical with the DSCP value of this AF convergence flow, and repeatedly sends this detection packet according to period of time T between every pair of ingress router and egress router;

(22) measure the unidirectional queuing delay q of each AF convergence flow user s through the DiffServ territory _s: q _s=[r _s-min (r _s)]/2, r in the formula _sBe the round-trip delay of this detection packet between ingress router and egress router, min (r _s) be r _sMinimum value, the propagation delay when not lining up and the approximation of processing delay sum;

(23) according to unidirectional queuing delay q _sNumerical value calculates the unit bandwidth congested expense p of AF convergence flow user s through the DiffServ territory ^s=γ q _s, it is unidirectional queuing delay q _sLinear function, the experiment experience value of coefficient gamma is 0.01, the big more expression network congestion of γ expense is big more to user's influence, but can not influence the fairness between each AF convergence flow; Calculate the congested expense that AF convergence flow user s need pay: x again _sP ^s=γ x _sQ _s, x in the formula _sCurrent transmission rate for user s.

Described step (3) further comprises following operating procedure:

(31) utilize the current transmission rate x of the function of measuring measurements and calculations AF convergence flow user s of flow regulator _s

(32) calculate at current transmission rate x _sNet utility functional value J with AF convergence flow user s under the congested expense _s(x _s)=U _s(x _s)-x _sP ^s, and first derivative: ${\stackrel{\·}{J}}_s\left(x_s\right)=\frac{\∂J_s}{\∂x_s}=\frac{w_s}{x_s}-p^s;$

(33), calculate the optimization transmission rate x of AF convergence flow under the current congestion expense according to the net utility maximization principle _s ^OptIf, ${\stackrel{\·}{J}}_s\left(x_s\right)>0$ , net utility functional value J is described _s(x _s) can be along with x _sIncrease and increase, so the user increases transmission rate: $x_s^{\mathrm{opt}}=x_s+\mathrm{\α}\·{\stackrel{\·}{J}}_s\left(x_s\right)$ , x in the formula _sBe the current transmission rate of user s, the increase amplitude is

If ${\stackrel{\·}{J}}_s\left(x_s\right)\≤0$ , net utility functional value J is described _s(x _s) can be along with x _sReduce and increase, then the user reduces transmission rate: $x_s^{\mathrm{opt}}=x_s-\mathrm{\β}\·\left|{\stackrel{\·}{J}}_s\left(x_s\right)\right|$ , x in the formula _sBe the current transmission rate of user s, the amplitude of reducing is

: α and β in above-mentioned two formulas are the adjustment coefficient, and 0＜α≤β, the amplitude that increases when promptly increasing transmission rate is little, and the amplitude that reduces when reducing transmission rate is big, so that network alleviate congestion rapidly congested the time, be unlikely to cause the concussion of network state, help keeping network stabilization.

Described step (4) comprises following operating procedure: at each speed adjustment cycle the token of the token bucket P of each AF convergence flow is produced speed PIR and be adjusted into its current optimization transmission rate x _Opt, promptly $\mathrm{PIR}=x_s^{\mathrm{opt}}$ , and the value of PIR and CIR relatively guarantees that PIR is not less than the value of CIR, promptly PIR=Max (CIR, PIR); Return step (2) then, the cycling that enters next flow adjustment cycle.

Described net utility maximization principle be with the effectiveness by the user deduct the user for this reason the required payment of effectiveness expense and the so-called net utility numerical value that obtains is sought maximum as much as possible.

The present invention is a kind of flow rate adjusting method that ensures " AF business " convergence flow fairness in differentiated services network based on congestion pricing, and it is that the flow control methods based on congestion pricing in the Internet is applied in the differentiated services network.Method of the present invention is compared with traditional Flow-rate adjustment algorithm (is representative with trTCM) and improvement algorithm (is representative with the DCCS algorithm), solve the defective of Flow-rate adjustment algorithm in the existing various differentiated services network well, can ensure the proportional fairness of AF convergence flow; And improve the AF convergence flow according to targeted rate shared bandwidth fairness in, can also improve the entire system throughput, make the network user obtain high cost performance.

Flow rate adjusting method of the present invention belongs to the algorithm based on token bucket, and it is the dual rate tricolor marker device (trTCM) that utilizes standard, the transmission rate of regulating each AF convergence flow by the PIR parameter of regulating token bucket.Because the dynamic adjustments effect of congestion pricing, network is congested more, and then congested expense is high more, and each AF convergence flow will reduce flow in order to realize the maximized target of net utility, promptly reduces its transmission rate; And if network bandwidth degree more than needed is big more, then congested expense is low more, and each AF convergence flow will increase flow in order to realize the maximized target of net utility, promptly improves its transmission rate; So just can realize the congested control of differentiated services network, and ensure that each AF convergence flow can obtain and the approximately proportional bandwidth sharing share of its targeted rate, thereby reach the target that ensures proportional fairness between each AF convergence flow.

Description of drawings

Fig. 1 is the building-block of logic of the flow regulator in the differentiated services network.

Fig. 2 is the operating procedure overview flow chart of flow rate adjusting method of the present invention.

Fig. 3 is a congested expense operating procedure flow chart of measuring the AF convergence flow in the flow rate adjusting method of the present invention.

Fig. 4 is the optimization transmission rate operating procedure flow chart that calculates the AF convergence flow in the flow rate adjusting method of the present invention.

Embodiment

Describe the working mechanism and the concrete operations step of the inventive method in detail below in conjunction with accompanying drawing.

The present invention be a kind of in differentiated services network the flow rate adjusting method based on congestion pricing guarantee " AF business " convergence flow fairness, its working mechanism is that the differentiated services network resource is abstracted into commodity in the market, and the AF convergence flow of competitive resource is each other regarded as user in the market, and represent the satisfaction that user's bandwidth consumed resource is obtained with the form of logarithmic utility function.The resource that the user consumes is more will to obtain higher satisfaction, and each user is necessary for its Internet resources that consume payment corresponding cost simultaneously.The network using congestion pricing strategy here, promptly the expense of unit resource is relevant with network congestion degree, and the congested more then expense of network is high more, and expense is zero when network is not congested.The edge router of differentiated services network is known the congested expense of current network by sending detection packet each other, and come the transmission rate of each AF convergence flow of real-time dynamic adjustments according to this congested cost information, so that make the net utility maximization of each AF convergence flow.

Therefore, flow rate adjusting method of the present invention mainly be use for reference in the Internet, use can ensure the principle of user's proportional fairness and the transmission rate of regulating each AF convergence flow based on the PIR parameter of the token bucket in the tricolor marker device (trTCM) of token bucket based on congestion pricing by change: set up in the differentiated services network Flow-rate adjustment Mathematical Modeling earlier based on the AF convergence flow of congestion pricing, and between the edge router of differentiated services network, periodically send detection packet and obtain the congested expense of current network, again according to this congested cost information, the targeted rate of each AF convergence flow and net utility maximization principle calculate the optimization transmission rate of each AF convergence flow, and the token bucket parameters of adjusting each AF convergence flow according to the optimization transmission rate, ensure that each AF convergence flow obtains and the approximately proportional bandwidth sharing share of its targeted rate, realizes the proportional fairness of resource-sharing between the AF convergence flow.Simultaneously, improve the overall transfer efficient of differentiated services network, improved the effective throughput of network integral body.

Referring to Fig. 2, specify four main operational steps of flow rate adjusting method of the present invention:

(1) sets up the Mathematical Modeling that realizes AF service convergence stream Flow-rate adjustment based on congestion pricing according to the agreement of the grade of service between user and network SLA.This step mainly comprises two operations:

(11) definition AF convergence flow user S set={ s ₁, s ₂, s is each AF convergence flow user in the formula; And each AF convergence flow user s defined following Mathematical Modeling:

The utility function of user s is U _s=w _sLogx _sX in the formula _sCurrent transmission rate for user s; w _sBe the coefficient of the utility function of user s, w _sNumerical value be set to the targeted rate of user s, this targeted rate is determined according to grade of service agreement SLA;

The net utility function of user s is J _s(x _s)=U _s(x _s)-x _sP ^sP in the formula ^sFor user s consumes the expense that unit bandwidth need be paid, i.e. x _sP ^sThe total cost that need pay for its AF convergence flow for user s;

(12) token of token bucket C produces the targeted rate that speed CIR is set to the AF convergence flow, and the token of token bucket P produces speed PIR and is set to 2 times CIR.

(2) measure the congested cost information of each AF convergence flow: send the detection packet of each AF convergence flow at the ingress edge router, and stamp timestamp; After this detection packet arrived corresponding egress router, egress router turned back to ingress router with this detection packet immediately, by ingress router according to the congested expense of time-delay calculation.

(3) edge router calculates the optimization transmission rate of each AF convergence flow under the current congestion expense according to congested cost information and user's net utility maximization principle, even get the transmission rate that user's net utility increases;

(4), at each speed adjustment cycle the token of the token bucket P of each AF convergence flow is produced speed PIR and be adjusted into its current optimization transmission rate x according to optimizing transmission rate _Opt, promptly $\mathrm{PIR}=x_s^{\mathrm{opt}}$ , and the value of PIR and CIR relatively guarantees that PIR is not less than the value of CIR, promptly PIR=Max (CIR, PIR); Return step (2) then, the cycling that enters the next Flow-rate adjustment cycle.

Referring to Fig. 3, describe the flow chart of each refinement step in step (2) the Measurement Network congestion condition in detail:

(21) according to the priority feature of each AF convergence flow, it is the DSCP value, generation has only the detection packet in packet header, and the DSCP value of this detection packet is identical with the DSCP value of this AF convergence flow, and repeatedly sends this detection packet according to period of time T between every pair of ingress router and egress router;

(22) measure the round-trip delay r of AF convergence flow user s by the mode that between ingress router and egress router, sends detection packet through the DiffServ territory _sWith unidirectional queuing delay q _s, q _s=[r _s-min (r _s)]/2, min (r in the formula _s) be r _sMinimum value, the propagation delay when not lining up and the approximation of processing delay sum;

(23) according to unidirectional queuing delay q _sNumerical value calculates the unit bandwidth congested expense p of AF convergence flow user s through the DiffServ territory ^s=γ q _s, it is unidirectional queuing delay q _sLinear function, the experiment experience value of coefficient gamma is 0.01, the big more expression network congestion of γ expense is big more to user's influence, but can not influence the fairness between each AF convergence flow; Calculate the congested expense that AF convergence flow user s need pay: x again _sP ^s=γ x _sQ _s, x in the formula _sCurrent transmission rate for user s.

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

(31) utilize the current transmission rate x of the function of measuring measurements and calculations AF convergence flow user s of flow regulator _s

(32) calculate at current transmission rate x _sNet utility functional value J with AF convergence flow user s under the congested expense _s(x _s)=U _s(x _s)-x _sP ^s, and first derivative: ${\stackrel{\·}{J}}_s\left(x_s\right)=\frac{\∂J_s}{\∂x_s}=\frac{w_s}{x_s}-p^s;$

(33), calculate the optimization transmission rate of AF convergence flow under the current congestion expense according to the net utility maximization principle , x in the formula _sBe the current transmission rate of user s, α and β are the adjustment coefficient, and 0＜α≤β; If promptly ${\stackrel{\·}{J}}_s\left(x_s\right)>0$ , net utility functional value J is described _s(x _s) can be along with x _sIncrease and increase, so the user increases transmission rate, the increase amplitude is If ${\stackrel{\·}{J}}_s\left(x_s\right)\≤0,$ Net utility functional value J is described _s(x _s) can be along with x _sReduce and increase, then the user reduces transmission rate, the amplitude of reducing is When regulating flow according to above-mentioned two formulas, the amplitude that increases when increasing transmission rate is little, and the amplitude that reduces when reducing transmission rate is big, so that network alleviate congestion rapidly congested the time, be unlikely to cause the concussion of network state, help keeping network stabilization.

Flow rate adjusting method of the present invention has carried out the analogue simulation test on computers, five kinds of situations (between response flow and the non-response flow and between the response flow of four kinds of different attributes) to influence " AF business " convergence flow fairness have all been carried out analogue simulation, and the inventive method and trTCM and the DCCS algorithm that belongs to the friendly controlling mechanism of TCP compared, simulation result shows that flow rate adjusting method of the present invention all shows more performance aspect total effective throughput improving fairness and improve, test is successful, has realized goal of the invention.

Following simple declaration the inventive method is in the test process of the fairness of the AF service convergence stream that improves different target rate (CIR).

10 TCP convergence flows are arranged in the emulation, and the targeted rate of each convergence flow (CIR) is followed successively by 0.1Mbps, 0.2Mbps, 0.3Mbps respectively, 0.4Mbps, 0.5Mbps, 0.6Mbps, 0.7Mbps, 0.8Mbps, 0.9Mbps and 1.0Mbps, therefore total targeted rate is 5.5Mbps.The bottleneck link bandwidth of artificial network topology is made as 11Mbps, so load is 50%.

In order to assess the fairness between the AF convergence flow, define a fairness index (FI) here:

$\mathrm{FI}={\left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\η}}_i\right)}^2/n\×\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\η}}_i^2$ , wherein n represents AF convergence flow user's number, η _i=x _i/ c _iRepresent the goodput of i convergence flow and the ratio of targeted rate, x _iThe average effective throughput of representing i convergence flow, c _iThe targeted rate (CIR) of representing i convergence flow.The fairness index is a nonnegative number between 0 and 1.This value is more near 1, and the expression fairness is good more, and each AF convergence flow all will obtain and the proportional bandwidth share of its targeted rate under the ideal state, promptly be respectively 0.2Mbps, 0.4Mbps, 0.6Mbps, 0.8Mbps, 1.0Mbps, 1.2Mbps, 1.4Mbps, 1.6Mbps, 1.8Mbps and 2.0Mbps.

L-G simulation test compares flow rate adjusting method of the present invention and trTCM and the DCCS algorithm that belongs to the friendly controlling mechanism of TCP respectively, the fairness index of trTCM is 0.74 as a result, the fairness index of DCCS is 0.82, and the fairness index of the inventive method is 0.96, relatively near perfect condition.This is because flow rate adjusting method of the present invention is different for the Flow-rate adjustment of the AF convergence flow of different CIR, and CIR is used as the coefficient of the utility function of each convergence flow, therefore can guarantee proportional fairness target with CIR.

Claims (7)

1, a kind of in differentiated services network based on the flow rate adjusting method of congestion pricing guarantee " AF business " convergence flow fairness, it is characterized in that: this method is to ensure that with being used in the Internet flow control methods based on congestion pricing of user's proportional fairness is applied to differentiated services network: set up in the differentiated services network Flow-rate adjustment Mathematical Modeling based on the AF convergence flow of congestion pricing earlier, and between the edge router of differentiated services network, send detection packet and obtain the congested expense of current network, this congested cost information of basis and net utility maximization principle calculate the optimization transmission rate of each AF convergence flow again, and the token bucket parameters of adjusting each AF convergence flow according to the optimization transmission rate, ensure that each AF convergence flow obtains and the approximately proportional bandwidth sharing share of its targeted rate, realizes the proportional fairness of resource-sharing between the AF convergence flow.

2, flow rate adjusting method according to claim 1 is characterized in that: comprise following operating procedure:

(1) Mathematical Modeling that realizes AF convergence flow Flow-rate adjustment based on congestion pricing is set up in agreement according to the grade of service between user and the network;

(2) send the detection packet of each AF convergence flow at the ingress edge router, and stamp timestamp; After this detection packet arrived corresponding egress router, egress router turned back to ingress router with this detection packet immediately, by ingress router according to the congested expense of time-delay calculation;

(3) edge router calculates the optimization transmission rate of each AF convergence flow under the current congestion expense according to congested cost information and user's net utility maximization principle, even get the transmission rate that user's net utility increases;

(4) adjust the token generation speed PIR of the token bucket P of each AF convergence flow according to optimizing transmission rate; Return step (2) then, carry out the cycling of Flow-rate adjustment.

3, flow rate adjusting method according to claim 2 is characterized in that: described step (1) further comprises following operating procedure:

(11) definition AF convergence flow user S set={ s ₁, s ₂, s is each AF convergence flow user in the formula; And each AF convergence flow user s defined following Mathematical Modeling:

The utility function of user s is U _s=w _sLogx _sX in the formula _sCurrent transmission rate for user s; w _sBe the coefficient of the utility function of user s, w _sNumerical value be set to the targeted rate of user s, this targeted rate is determined according to grade of service agreement SLA;

The net utility function of user s is J _s(x _s)=U _s(x _s)-x _sP ^sP in the formula ^sFor user s consumes the expense that unit bandwidth need be paid, i.e. x _sP ^sThe total cost that need pay for its AF convergence flow for user s;

(12) token of token bucket C produces the targeted rate that speed CIR is set to the AF convergence flow, and the token of token bucket P produces speed PIR and is set to 2 times CIR.

4, flow rate adjusting method according to claim 2 is characterized in that: described step (2) further comprises following operating procedure:

(21) according to the priority feature of each AF convergence flow, it is the DSCP value, generation has only the detection packet in packet header, and the DSCP value of this detection packet is identical with the DSCP value of this AF convergence flow, and repeatedly sends this detection packet according to period of time T between every pair of ingress router and egress router;

(22) measure the unidirectional queuing delay q of each AF convergence flow user s through the DiffServ territory _s: q _s=[r _s-min (r _s)]/2, r in the formula _sBe the round-trip delay of this detection packet between ingress router and egress router, min (r _s) be r _sMinimum value, the propagation delay when not lining up and the approximation of processing delay sum;

(23) according to unidirectional queuing delay q _sNumerical value calculates the unit bandwidth congested expense p of AF convergence flow user s through the DiffServ territory ^s=γ q _s, it is unidirectional queuing delay q _sLinear function, the experiment experience value of coefficient gamma is 0.01, the big more expression network congestion of γ expense is big more to user's influence, but can not influence the fairness between each AF convergence flow; Calculate the congested expense that AF convergence flow user s need pay: x again _sP ^s=γ x _sQ _s, x in the formula _sCurrent transmission rate for user s.

5, flow rate adjusting method according to claim 2 is characterized in that: described step (3) further comprises following operating procedure:

(31) utilize the current transmission rate x of the function of measuring measurements and calculations AF convergence flow user s of flow regulator _s

(32) calculate at current transmission rate x _sNet utility functional value J with AF convergence flow user s under the congested expense _s(x _s)=U _s(x _s)-x _sP ^s, and first derivative: ${\stackrel{\·}{J}}_s\left(x_s\right)=\frac{{\∂J}_s}{{\∂x}_s}=\frac{w_s}{x_s}-p^s;$

(33), calculate the optimization transmission rate x of AF convergence flow under the current congestion expense according to the net utility maximization principle _s ^OptIf, ${\stackrel{\·}{J}}_s\left(x_s\right)>0,$ Net utility functional value J is described _s(x _s) can be along with x _sIncrease and increase, so the user increases transmission rate: $x_s^{\mathrm{opt}}=x_s+\mathrm{\α}\·{\stackrel{\·}{J}}_s\left(x_s\right),$ X in the formula _sBe the current transmission rate of user s, the increase amplitude is $\mathrm{\α}\·{\stackrel{\·}{J}}_s\left(x_s\right);$ If ${\stackrel{\·}{J}}_s\left(x_s\right)\≤0,$ Net utility functional value J is described _s(x _s) can be along with x _sReduce and increase, then the user reduces transmission rate: $x_s^{\mathrm{opt}}=x_s-\mathrm{\β}\·\left|{\stackrel{\·}{J}}_s\left(x_s\right)\right|,$ X in the formula _sBe the current transmission rate of user s, the amplitude of reducing is $\mathrm{\β}\·\left|{\stackrel{\·}{J}}_s\left(x_s\right)\right|;$ α and β in above-mentioned two formulas are the adjustment coefficient, and 0＜α≤β, the amplitude that increases when promptly increasing transmission rate is little, and the amplitude that reduces when reducing transmission rate is big, so that network alleviate congestion rapidly congested the time, be unlikely to cause the concussion of network state, help keeping network stabilization.

6, flow rate adjusting method according to claim 2 is characterized in that: described step (4) comprises following operating procedure: at each speed adjustment cycle the token of the token bucket P of each AF convergence flow is produced speed PIR and be adjusted into its current optimization transmission rate x _Opt, promptly $\mathrm{PIR}=x_s^{\mathrm{opt}},$ And the value of PIR and CIR relatively, guarantee that PIR is not less than the value of CIR, promptly PIR=Max (CIR, PIR); Return step (2) then, the cycling that enters next flow adjustment cycle.

7, flow rate adjusting method according to claim 1 or 5 is characterized in that: described net utility maximization principle be with the effectiveness by the user deduct the user for this reason the required payment of effectiveness expense and the so-called net utility numerical value that obtains is sought maximum as much as possible.

Images