CN1317838C - Light burst exchange network edge node structure and realizing method - Google Patents

Abstract

The present invention relates to a light burst exchange network edge node structure and a realization method, which is used in the technical field of light communication. The edge node structure comprises a control packet and a light packet transmitting modules, a wavelength division multiplexer, a self-adaptive flow assembly module and a bias confirmation module, wherein the self-adaptive flow assembly module is connected with the bias confirmation module and then is connected to the control packet transmitting module. The light packet transmitting module and the control packet transmitting module are together connected to the wavelength division multiplexer to transfer. The output of a control packet trigger module in the self-adaptive flow assembly module is connected to an AAR(p) filter; the output of 1(n-I) and e(n) memories is also connected to the AAR(p) filter as input; the output of the AAR(p) filter is connected with a packet length confirmation module; the output of the packet length confirmation module is connected with a dynamic BAT decision module. In the method of the present invention, the self-adaptive flow assembly module comprising the AAR(p) filter, and the bias confirmation module tightly combined with the self-adaptive flow assembly module are added in the light exchange network edge node structure to simultaneously ensure the fairness of data packets and provide QoS guarantee at edge nodes.

Description

Light burst exchange network fringe node structure and implementation method

Technical field

The present invention relates to a kind of network configuration and implementation method that is used for the optical communication technique field, particularly a kind of light burst-switched fringe node structure and implementation method of real time business dynamic self-adapting assembling Network Based.

Background technology

Light burst-switched (OBS) network has reduced the requirement to optical switch and light buffer memory, and the sudden Packet Service of the fine support of energy is realized big packet switching and transparent transmission under current optical tech condition, become one of focus of optical communication field research.It is made of some fringe nodes and core node, fringe node is responsible for finishing the interface with legacy network, ingress edge node is dressed up burst to the conditional electronic package according to the requirement of destination address and service quality (QoS), and create a control for each burst and divide into groups to send in independent dense wave division multipurpose (DWDM) channel prior to burst a period of time, be called offset time during this period of time.Core node carries out light-electrical-optical conversion and is treated to corresponding burst packets setting up full light path to control grouping, and burst is through direct transparent transmission in the full optical channel that sets in advance after one section delay.The information that the outlet edge node comprises according to the control grouping again is reduced to pocket of electrons to burst and finally arrives the legacy network terminal.The OBS network edge node must be given full play to electricity and the enough intelligence of light, burst assembling (burst assembling) is the key technology of OBS network, and fringe node has directly determined the core optical networks performance to the integration of traffic carrying capacity, the generation and the output light bag characteristic of controlling packet.In order to make full use of existing electronic equipment storage flexibly and controlled function and in conjunction with the big capacity advantage of optical communication, requirement is at time delay fairness and the QoS of network edge to flow processing in real time and assurance different business class, and design has proposed very big challenge to the edge node structure for this.

Find by prior art documents, fringe node structure (Yi jun Xiong, Marc Vandenhoute, the control structure of and Hakki C.Cankaya.Control Architecture in Optical Burst-Switched WDMNetworks＜light burst-switched wavelength division multiplexed network 〉, IEEE Journal on SelectedAreas in Communications＜Institute of Electrical and Electric Engineers selected field periodical of communicating by letter 〉, 2000,18 (10): 1838-1850) continue to use the flow rate mode of legacy circuit-switched, still adopt fixed packet length and the assemble method of time, assembling and offset time have brought extra time delay, can't adapt to the sudden and self-similarity requirement of data network; Existing fringe node structure does not relate to the fairness of differentiated service, the inhomogeneity business is to the delay requirement difference in the network, high-priority service must guarantee lower-delay, and low priority traffice can be tolerated big relatively time delay, and guaranteed qos has just lost meaning on the basis that does not have the time delay fairness; Tradition fringe node structure is the extra offset time of unilateral increase, come the QoS of priority to guarantee of differentiated service class at core node by big offset time, this has increased the time delay of overall optical network greatly, to multi-hop number, long radius network even lost the advantage of the low time delay of OBS network uni-directional reservation; The assembling granularity is predefined in the existing fringe node design, and the real network flow is lacked flexibility and increased expense and time delay.These limiting factors of existing fringe node structure have increased the pressure of core optical networks, and network throughput mis-behave, the offset time behind the traffic shaping form mechanism and increased light net time delay, have also reduced the robust property of optical-fiber network simultaneously.

Summary of the invention

The objective of the invention is at the deficiencies in the prior art, a kind of light burst exchange network fringe node structure and implementation method are proposed, make it improve network utilization and network flexibility, more adapt to the relevant self similarity characteristics of the professional long-range of existing network, the estimation network discharge characteristic makes wide utilance of light guipure and throughput more meet real-time optical-fiber network discharge characteristic in real time.Strengthen optical-fiber network self to the jumbo adaptive capacity of multi-service, helped making up a self adaptation, light burst exchange network flexibly.Fringe node structure of the present invention can guarantee and fairness for the inhomogeneity business provides QoS simultaneously, fundamentally reduce long time delay and the overhead introduced owing to professional assembling, and laid a good foundation for optical core network network QoS, route and competition solve.

The present invention is achieved by the following technical solutions, light burst exchange network fringe node structure of the present invention comprises: light bag transmitter module and controlling packet transmitter module, wavelength division multiplexer, adaptive flow Knockdown block, biasing determination module, its annexation is: the adaptive flow Knockdown block connects the biasing determination module and is connected into the controlling packet transmitter module then, and light bag transmitter module and controlling packet transmitter module are connected into wavelength division multiplexer together and transmit.Adaptive flow Knockdown block inside comprises controlling packet trigger module, self adaptation autoregression AAR (p) (AdaptiveAuto-Regressive) filter, l (n-I) and e (n) memory, the long determination module of bag, dynamic BAT decision module, the output of controlling packet trigger module is connected into AAR (p) filter, l (n-I) also is connected to AAR (p) filter as input with the output of e (n) memory, the output of AAR (p) filter connects the long determination module of bag, wraps long determination module output and connects dynamic burst built-up time (BAT:Burst Assembly Time) decision module.

The present invention has also made significant improvement to adaptive flow Knockdown block and biasing determination module:

The AAR of adaptive flow Knockdown block (p) filter estimates that with long weighted sum p rank autoregression (AR (p)) the model combination of the bag in the predicted time bag of this secondary burst is long;

The dynamic BAT decision module of adaptive flow Knockdown block is dynamically determined this BAT according to estimating the long and last time load of bag, guarantees the time delay fairness of region class; The biasing determination module is a key character of the present invention, and it is according to the dynamically definite offset time differentiation based on inhomogeneity QoS of BAT value.

The present invention is based on above structure technology feature, fringe node is the estimation network discharge characteristic in real time, dynamically adjust the set of bursts ETL estimated time of loading, before finishing prerequisite, the burst assembling estimates that thereby the long and definite offset time of bag sends controlling packet, guarantees for Network provides adaptivity, flexibility, time delay fairness and QoS.

Controlling packet trigger module notice AAR (p) linear prediction filter of adaptive flow Knockdown block wraps long estimation, call l (n-I) and calculate adaptation coefficient with the past bag length and the estimated value of e (n) memory, check the predicted time value simultaneously, add the flow l (n+1) in this burst assembling predicted time and calculate this flow weight coefficient w (1), estimate the long predicted value L (n+1) of this bag at the long determination module of bag, estimate load by dynamic BAT decision module according to the relation of the last Bao Changyu built-up time again, dynamically adjust built-up time, for inhomogeneity is provided with the time delay fairness that different threshold value realizations is distinguished based on class, the dynamic self-adapting of service traffics characteristic assembling when finishing more near network implementation.The biasing determination module is determined to realize that based on the offset time value of inhomogeneity QoS the QoS that distinguishes based on class guarantees according to the dynamic BAT of Knockdown block, because regardless of different kinds of the equal outer offset time of occupying volume not of offset time, unique estimated time that reduces burst packets, and the dynamic self-adapting packing algorithm of the present invention's employing provides assurance for this problem just in this case.Being arranged on that max-thresholds and offset time are distinguished do not change the QoS that has fully guaranteed little fairness of high-priority service time delay and big offset time under the time delay condition.The estimation bag is long and offset time is definite enters the wavelength division multiplexer transmission after the controlling packet transmitter module carries out the electric light conversion, and light bag transmitter module is changed multiplexing afferent echo division multiplexer with data burst bag electric light and transmitted behind offset time.

Below light burst exchange network fringe node structure of the present invention is further made qualification:

Light burst exchange network fringe node structure of the present invention also can further comprise: ply-yarn drill routing iinformation processing module, electron exchanger, the formation of a plurality of fringe node data centralization, wavelength selector, route Wavelength dispatching device, controlling packet produce.Its annexation is: ply-yarn drill routing iinformation processing module is the interface with outside router, the other end links to each other with electron exchanger, the electron exchanger outlet is connected to the data centralization formation of a plurality of fringe nodes, data centralization formation simultaneously is connected with the adaptive flow Knockdown block is two-way: the data centralization formation of fringe node connects the controlling packet trigger module in the adaptive flow Knockdown block, also connect l (n-I) and the input of e (n) memory, the output feedback of its dynamic BAT decision module connects the data centralization formation of back edge edge node.The biasing determination module is connected into route Wavelength dispatching device, and route Wavelength dispatching device connects wavelength selector, output access control packet generation module.The data centralization formation outlet of fringe node connects wavelength selector.

Electric packet from access-layer router is finished packet header identification and Route Selection by ply-yarn drill routing iinformation processing module, enter electron exchanger and select corresponding output port of this packet and service class formation, buffer memory queuing in the data centralization formation of corresponding edge node class, simultaneously mutual with the adaptive flow Knockdown block, when the electronic data bag is first arrival bag of formation in the current data set, trigger the adaptive flow Knockdown block, and accept the dynamic group ETL estimated time of loading of its feedback.After dynamic assembling is finished in the formation of a plurality of fringe node data centralization, it is that the burst packets that corresponding data centralization set of queues installs is selected output wavelength that route Wavelength dispatching device drives wavelength selector, transmits thereby enter the multiplexing afferent echo division multiplexer in light bag transmitter module electric light conversion back.

The implementation method of light burst exchange network fringe node structure of the present invention, specific as follows:

In light network edge node structure, add one and comprise the adaptive flow Knockdown block of AAR linear prediction filter and the biasing determination module of combining closely with it.The relevant self-similarity nature of adaptation network edge access service long-range that the adaptive flow Knockdown block is real-time, adopt normalization minimum mean-square error prediction solution, the AAR of adaptive flow Knockdown block (p) filter is in predicted time, and the long and adaptation coefficient of p bag and the current flow that arrived shift to an earlier date the self adaptation autoregression and estimate to wrap length in the past according to corresponding port and service queue.The dynamic BAT decision module of adaptive flow Knockdown block according to last time offered load dynamically adjust different port and inhomogeneous set of bursts ETL estimated time of loading to adapt to the real-time traffic characteristic, and for high-priority service designs less threshold value, for low priority designs relatively large threshold value, provide the time delay fairness for the inhomogeneity business in advance at network edge node like this.Thisly estimate to consider simultaneously the BAT of inhomogeneity time delay fairness determines it is important innovations of the present invention based on the last load; The offset time determination module passes through weights simultaneously, in the set of bursts ETL estimated time of loading, design big offset time for high-priority service, for low priority traffice designs relatively little offset time, outside occupying volume not under the offset time condition, the offset time of inhomogeneity business is divided into burst packets provides QoS to guarantee in core optical networks transmission exchange like this.This dynamic BAT according to the self adaptation Knockdown block estimates to determine that the offset time differentiating method based on inhomogeneity QoS is an important innovations of the present invention.These 2 important innovations of the present invention are on dynamic self-adapting maximum burst built-up time decision basis, not only guaranteed the fairness that the high-priority service time delay is little but also guaranteed the QoS of big offset time, obtained the high usage of distinguishing based on service class, low time delay and low packet loss ratio at the optical-fiber network core node.

Below by step the present invention is done further concrete the qualification:

(1) packet of electric router carries out the routing iinformation processing at the ply-yarn drill place, finish packet header identification and Route Selection, select output port and service class formation by electronic data interchange again, finish the assembling of dynamic self-adapting flow in the queuing of the corresponding data centralization formation of fringe node buffer memory then.

(2) corresponding port and service class are concentrated corresponding self adaptation autoregression AAR linear prediction filter of first bag arrival this burst of triggering in the formation.Call last time the long predicted value L (n) of bag simultaneously so that error of calculation information e (n)=l (n)-L (n), and when predicted time finishes, concentrate formation that current arrival wrapped long letter and ceased l (n+1) and pass to the AAR filter by burst.

(3) call over p time the bag long value l (n-I), I=0,1, ... p, the control information e (n) that provides according to previous step calculates adaptation coefficient, estimates that in conjunction with the current bag long letter breath l (n+1) that has arrived bag is long then, and expression formula (1) (2) has provided the method for estimation of this long predicted value L of bag (n+1), wherein w (1) is the weighting that has arrived flow l (n+1) in this predicted time of introducing, and it is burst assembling in the light burst-switched JET agreement

$L(n+1)=\underset{i=0}{\overset{p-1}{\mathrm{\Σ}}}w\left(i\right)l(n-i)+\mathrm{\αl}(n+1)\frac{T_a}{T_a-T_o}=\underset{i=-1}{\overset{p-1}{\mathrm{\Σ}}}w\left(i\right)l(n-i)---\left(1\right)$

$\stackrel{\→}{w}(n+1)=\stackrel{\→}{w}\left(n\right)+\frac{\mathrm{\μ}[l\left(n\right)-L\left(n\right)]\stackrel{\→}{l}(n-1)}{{\left|\right|\stackrel{\→}{l}(n-1)\left|\right|}^2}---\left(2\right)$

The ratio weighting of time and current n+1 volume forecasting time.This discharge characteristic no doubt is closely connected with nearest p Bao Changyou of past, yet the arrival in this predicted time is surrounded by bigger correlation, and it is important innovations of the present invention that bag weighting in the predicted time is combined with AR (p) model.

(4) previous step obtains the long predicted value L (n+1) of this bag suddenly, pass to dynamic BAT decision module, it estimates load according to the relation of the last Bao Changyu built-up time, dynamically adjust the built-up time that this port and service class are concentrated formation, for high-priority service is set less threshold value, for low priority class is set bigger threshold value, the threshold zone of built-up time branch has guaranteed that the low time delay of real-time traffic has realized fairness in the process of dynamically adjusting like this.The dynamic self-adapting of service traffics characteristic assembling when finishing more near network implementation, this BAT based on the last load estimation and time delay fairness determines it is important innovations of the present invention.

(5) this long predicted value L of bag (n+1) that accepts step (3) passes to the biasing determination module simultaneously, according to the definite offset time value of the dynamic BAT of Knockdown block based on inhomogeneity QoS, the bigger offset time of design in the high priority class business, and in the low priority class business, designing less relatively offset time, this offset time differentiating method of estimating based on dynamic BAT is an important innovations of the present invention.The burst packets of corresponding data centralization formation is carried out route and wavelength selection then.This moment controlling packet produce through the conversion of light bag transmitter module electric light with advance optical fiber and in core net, transmit through the emission of the data burst light bag behind the offset time is multiplexing.

In the fringe node structure implementation method step of the present invention (4), controlling packet is carried immediately after the formation predicted time finishes and is estimated that information such as bag is long send, if the long estimated value of bag is greater than actual value then carry out invalid filling and send; If estimate that bag is long less than actual value then keep remaining data in the corresponding ports formation, wait for next burst assembling and merge with it sending.Consider the relation of built-up time, offset time and predicted time, implementation method has comprised following two kinds, optional one during use:

1〉dynamically change pure autoregression (AR) algorithm of built-up time, do not have the long weighting of bag in the current predicted time, the adaptively changing built-up time, offset time distributes (as 6:4:3:2) according to inhomogeneous weights.

2〉inhomogeneous built-up time and the offset time dynamic AAR algorithm that changes in certain threshold range, this bag length of introducing is the improvement to pure AR model in the predicted time, self adaptation estimating burst flow makes the nearest discharge characteristic of the long adaptation of bag.

Compared with prior art, the present invention has the following advantages:

(1) traditional light burst-switched controlling packet sends after built-up time and offset time are finished fully and has brought very big assembling time delay, and do not consider self similarity business characteristic and current network conditions to wrapping long estimation, data burst and long-range correlation are brought very big pressure to core node.Adaptive forecasting method of the present invention has successfully been used for reference the DSP technology of traditional maturation and has been improved the weighting of having introduced this volume forecasting, accurately estimated the bursty data discharge characteristic, relation according to the last Bao Changyu built-up time is estimated load, dynamically adjusts the inhomogeneous built-up time of different port to adapt to the real-time traffic characteristic by this estimation bag length and load again.Improved network utilization and network performance, fundamentally reduced burst packets assembling time delay and overhead, made wide utilance of light guipure and throughput more meet real-time optical-fiber network discharge characteristic.

(2) traditional light net does not guarantee the fairness of packet simultaneously and provides QoS to guarantee at fringe node.The present invention sets less threshold value for high-priority service on the basis that the dynamic self-adapting built-up time is determined, for low priority class is set bigger threshold value, the threshold zone of built-up time is divided the low time delay that has guaranteed real-time traffic in the process of dynamically adjusting like this, has realized the fairness of different business class time delay.

(3) according to dynamic self-adapting built-up time and offset time weighting algorithm inhomogeneous offset time is set, the bigger offset time of design in the high priority class business, and in the low priority class business design less relatively offset time, because regardless of different kinds of the equal outer offset time of occupying volume not of offset time, unique estimated time that reduces burst packets, and the dynamic self-adapting packing algorithm that adopts in this case provides assurance for this problem just.

(4) being arranged on that max-thresholds and offset time are distinguished do not change the QoS that has fully guaranteed little fairness of high-priority service time delay and big offset time under the time delay condition, for providing QoS simultaneously, JET agreement inhomogeneity business guarantees and fairness, and be that optical core network network QoS, route and competition solution are laid a good foundation, obtain high usage, low time delay and low packet loss ratio at the optical-fiber network core node.

The present invention has strengthened optical-fiber network self to multi-service, jumbo adaptive capacity, help making up a self adaptation, light burst exchange network flexibly, can both regulate automatically any business model and load, on the transmission opportunity of control burst group, reduce the generation of conflict.Generally acknowledged that now Packet Service has self similarity and long-range correlation properties, have only the intelligent of increase edge routing node for this business, make full use of the buffer memory flexible nature of electricity layer, could improve the adaptive ability of network, simplify the control and the management of core optical networks, could fundamentally improve network performance, strengthen the robustness of network self.

Description of drawings

Fig. 1 dynamic self-adapting flow assembling of the present invention light burst-switched fringe node structure chart

Wherein, four couple in router packet electron exchange are to N-1 edge nodes, λ wherein ₁, λ ₂, λ ₃, λ ₄Expression enters the transmission wavelength of core optical networks.

Fig. 2 dynamic self-adapting flow assembling of the present invention light burst-switched the whole network schematic diagram

Wherein, the edge is an edge nodes, and core is the core router node.

Fig. 3 differentiated service threshold value of the present invention fairness and biasing QoS schematic diagram

Among the figure, white threshold value comprises the offset time that the black oblique line is represented.

Embodiment

As shown in Figure 1, edge nodes structure of the present invention comprises: light bag transmitter module 1, wavelength division multiplexer 2, adaptive flow Knockdown block 3, biasing determination module 4, controlling packet transmitter module 5, its annexation is: adaptive flow Knockdown block 3 connects biasing determination module 4 and is connected into controlling packet transmitter module 5 then, light bag transmitter module 1 and controlling packet transmitter module 5 are connected into wavelength division multiplexer 2 together and transmit, adaptive flow Knockdown block 3 inside comprise controlling packet trigger module 6, AAR (p) filter 7, l (n-I) and e (n) memory 8, wrap long determination module 9, dynamic BAT decision module 10,6 outputs of controlling packet trigger module are connected into AAR (p) filter 7, l (n-I) also is connected to AAR (p) filter 7 as input with 8 outputs of e (n) memory, the output of AAR (p) filter 7 connects the long determination module 9 of bag, wraps long determination module 9 outputs and connects dynamic BAT decision module 10.

Fringe node structure of the present invention also comprises: ply-yarn drill routing iinformation processing module 11, electron exchanger 12, a plurality of fringe node data centralization formation 13, wavelength selector 14, route Wavelength dispatching device 15, controlling packet produces 16, its annexation is: ply-yarn drill routing iinformation processing module 11 is the interfaces with outside router, the other end links to each other with electron exchanger 12, electron exchanger 12 outlets are connected to the data centralization formation 13 of a plurality of fringe nodes, data centralization formation 13 and 3 two-way connections of adaptive flow Knockdown block simultaneously: the controlling packet trigger module 6 that the data centralization formation 13 of fringe node connects in the adaptive flow Knockdown block 3, also connect l (n-I) and 8 inputs of e (n) memory, the output feedback of its dynamic BAT decision module 10 connects the data centralization formation 13 of back edge edge node.Biasing determination module 4 is connected into route Wavelength dispatching device 15, and route Wavelength dispatching device 15 connects wavelength selector 14, output access control packet generation module 16, and data centralization formation 13 outlets of fringe node connect wavelength selector 14.

Controlling packet trigger module 6 notice AAR (p) linear prediction filters 7 of adaptive flow Knockdown block 3 wrap long estimation, call l (n-I) and calculate adaptation coefficient with the past bag length and the estimated value of e (n) memory 8, check the predicted time value simultaneously, add the flow l (n+1) in this burst assembling predicted time and calculate this flow weight coefficient w (1), estimate the long predicted value L (n+1) of this bag at the long determination module 9 of bag, estimate load by dynamic BAT decision module 10 according to the relation of the last Bao Changyu built-up time again, dynamically adjust built-up time.Biasing determination module 4 is according to the definite offset time value based on inhomogeneity QoS of the dynamic BAT of Knockdown block, the estimation bag is long and offset time is definite enters wavelength division multiplexer 2 transmissions after controlling packet transmitter module 5 carries out the electric light conversion, and light bag transmitter module 1 is changed multiplexing afferent echo division multiplexer 2 with data burst bag electric light and transmitted behind offset time.

Electric packet from access-layer router is finished packet header identification and Route Selection by ply-yarn drill routing iinformation processing module 11, enter electron exchanger 12 and select corresponding output port of this packet and service class formation, buffer memory queuing in the data centralization formation 13 of corresponding edge node class, simultaneously mutual with adaptive flow Knockdown block 3, represented the concentrated formation of two classes of first fringe node among Fig. 1.When the electronic data bag is first arrival bag of formation in the current data set, triggers adaptive flow Knockdown block 3, and accept the dynamic group ETL estimated time of loading of its feedback.After dynamic assembling is finished in a plurality of fringe node data centralization formation 13, route Wavelength dispatching device 15 drives the burst packets selection output wavelength that wavelength selector 14 installs for corresponding data centralization set of queues, transmits thereby enter the multiplexing afferent echo division multiplexer 2 in light bag transmitter module 1 electric light conversion back.

Fig. 2 has provided the light burst-switched example of a whole network at place of the present invention, the light burst exchange network fringe node router one 8 different port dynamic self-adaptings that wrap in that Access Layer electricity router 17 sends are assembled into inhomogeneous burst group, when finishing, predicted time generates the routing iinformation control grouping of carrying this burst group, and send along control channel prior to one section offset time of burst group, burst group is sent along data channel subsequently, control is grouped in core network light switch router 19 through processed after the opto-electronic conversion, be the burst group reservation output port that arrives soon after, then after necessary modifications has been done in some territory, continue to pass toward next hop node along control channel through the electrical-optical conversion, up to outlet edge node router, to be reduced to ordinary groups after the data burst packet fragmentation, be forwarded to the purpose electricity router that is attached thereto.The thin black line of band arrow is represented the control channel wavelength among the figure, and heavy line is a bursty data channels.

Fig. 3 guarantees to have done vivid explanation to differentiated service threshold value fairness and the biasing QoS that the present invention proposes, thereby after having changed the built-up time of different business class dynamically with dynamic self-adapting assembler prediction bag is long, for inhomogeneous offset time distributes different weights, the weights of four priority class among the figure are 6:4:3:2 successively, fairness is to distribute different max-thresholds to each class, in different built-up times, distinguish the dynamic inhomogeneous offset time that changes then according to weights, do not need to determine absolute value, thereby improve resource utilization.

Claims (4)

1, a kind of light burst exchange network fringe node structure, comprise: light bag transmitter module (1), wavelength division multiplexer (2), controlling packet transmitter module (5), it is characterized in that, comprise: adaptive flow Knockdown block (3), biasing determination module (4), its annexation is: adaptive flow Knockdown block (3) connects biasing determination module (4) and is connected into controlling packet transmitter module (5) then, light bag transmitter module (1) and controlling packet transmitter module (5) are connected into wavelength division multiplexer (2) together and transmit, adaptive flow Knockdown block (3) inside comprises controlling packet trigger module (6), AAR (p) filter (7), l (n-I) and e (n) memory (8), wrap long determination module (9), dynamic BAT decision module (10), controlling packet trigger module (6) output is connected into AAR (p) filter (7), l (n-I) also is connected to AAR (p) filter (7) as input with e (n) memory (8) output, the output of AAR (p) filter (7) connects bag long determination module (9), wrap long determination module (9) output and connect dynamic BAT decision module (10), the fringe node architectural feature also comprises: the ply-yarn drill routing iinformation processing module (11) in the light burst-switched fringe node, electron exchanger (12), a plurality of fringe node data centralization formation (13), wavelength selector (14), route Wavelength dispatching device (15), controlling packet generator (16), its annexation is: ply-yarn drill routing iinformation processing module (11) is the interface with outside router, the other end links to each other with electron exchanger (12), electron exchanger (12) outlet is connected to the data centralization formation (13) of a plurality of fringe nodes, data centralization formation simultaneously (13) and two-way connection of adaptive flow Knockdown block (3): the data centralization formation (13) of fringe node connects the controlling packet trigger module (6) in the adaptive flow Knockdown block (3), also connect l (n-I) and e (n) memory (8) input, the output feedback of its dynamic BAT decision module (10) connects the data centralization formation (13) of back edge edge node, biasing determination module (4) is connected into route Wavelength dispatching device (15), route Wavelength dispatching device (15) connects wavelength selector (14), output access control packet generation module (16), data centralization formation (13) outlet of fringe node connects wavelength selector (14).

2, a kind of implementation method of light burst exchange network fringe node structure, it is characterized in that, in light network edge node structure, add one and comprise the adaptive flow Knockdown block of AAR linear prediction filter and the biasing determination module of combining closely with it, the relevant self-similarity nature of adaptation network edge access service long-range that the adaptive flow Knockdown block is real-time, adopt normalization minimum mean-square error prediction solution, the AAR of adaptive flow Knockdown block (p) filter is in predicted time, long and the adaptation coefficient of p bag and the current flow that arrived shift to an earlier date the self adaptation autoregression and estimate to wrap length in the past according to corresponding port and service queue, the dynamic BAT decision module of adaptive flow Knockdown block according to last time offered load dynamically adjust port and set of bursts ETL estimated time of loading and adapt to the real-time traffic characteristic, and design less threshold value for high-priority service, for low priority designs relatively large threshold value, provide the time delay fairness for all kinds of business in advance at network edge node like this; The offset time determination module passes through weights simultaneously, in the set of bursts ETL estimated time of loading, design big offset time for high-priority service, for low priority traffice designs relatively little offset time, like this under extra offset time is zero condition, the offset time of all kinds of business is divided into burst packets provides QoS to guarantee in core optical networks transmission exchange, electric packet from access-layer router is finished packet header identification and Route Selection by ply-yarn drill routing iinformation processing module, enter electron exchanger and select corresponding output port of this packet and service class formation, buffer memory queuing in the data centralization formation of corresponding edge node class, simultaneously mutual with the adaptive flow Knockdown block, when the electronic data bag is first arrival bag of formation in the current data set, trigger the adaptive flow Knockdown block, and accept the dynamic group ETL estimated time of loading of its feedback, after dynamic assembling is finished in the formation of a plurality of fringe node data centralization, it is that the burst packets that corresponding data centralization set of queues installs is selected output wavelength that route Wavelength dispatching device drives wavelength selector, transmits thereby enter the multiplexing afferent echo division multiplexer in light bag transmitter module electric light conversion back.

3, the implementation method of light burst exchange network fringe node structure according to claim 2 is characterized in that, below does further concrete the qualification by step:

(1) packet of electric router carries out the routing iinformation processing at the ply-yarn drill place, finish packet header identification and Route Selection, select output port and service class formation by electronic data interchange again, finish the assembling of dynamic self-adapting flow in the queuing of the corresponding data centralization formation of fringe node buffer memory then;

(2) corresponding port and service class are concentrated corresponding self adaptation autoregression AAR linear prediction filter of first bag arrival this burst of triggering in the formation, call last time the long predicted value L (n) of bag simultaneously so that error of calculation information e (n)=l (n)-L (n), and when predicted time finishes, concentrate formation that current arrival wrapped long letter and ceased l (n+1) and pass to the AAR filter by burst;

(3) call over p time the bag long value l (n-I), I=0,1, ... p-1, the control information e (n) that provides according to previous step calculates adaptation coefficient, estimate that in conjunction with the current bag long letter breath l (n+1) that has arrived bag is long then, this long predicted value L of bag (n+1) be two-part and: the weighted sum of first for wrapping long value l (n-I) p time in the past, wherein current n+1 time weight vector be last time weight vector and control information e (n) multiply by the weighted sum of normalization bag long vector: second portion is introduced the weighting that has arrived flow l (n+1) in this predicted time, and coefficient is set of bursts ETL estimated time of loading and the ratio weighting of current n+1 volume forecasting time in the light burst-switched JET agreement;

(4) previous step obtains the long predicted value L (n+1) of this bag suddenly, pass to dynamic BAT decision module, it estimates load according to the relation of the last Bao Changyu built-up time, dynamically adjust the built-up time that this port and service class are concentrated formation, for high-priority service is set less threshold value, for low priority class is set bigger threshold value, the threshold zone of built-up time branch has guaranteed that the low time delay of real-time traffic has realized fairness in the process of dynamically adjusting like this;

(5) this long predicted value L of bag (n+1) that accepts step (3) passes to the biasing determination module simultaneously, according to the definite offset time value of the dynamic BAT of Knockdown block based on all kinds of QoS, the bigger offset time of design in the high priority class business, and in the low priority class business design less relatively offset time, the burst packets of corresponding then data centralization formation carries out route and wavelength is selected, this moment controlling packet produce through the conversion of light bag transmitter module electric light with advance optical fiber and in core net, transmit through the emission of the data burst light bag behind the offset time is multiplexing.

4, the implementation method of light burst exchange network fringe node structure according to claim 3, it is characterized in that, in the step (4), controlling packet is carried immediately after the formation predicted time finishes and is estimated that bag long letter breath sends, if the long estimated value of bag is greater than actual value then carry out invalid filling and send, if estimate that bag is long less than actual value then keep remaining data in the corresponding ports formation, wait for next burst assembling and merge with it and send, the specific implementation method comprises following two kinds, optional one during use:

1〉dynamically change the pure autoregression algorithm of built-up time, the long weight term of the bag in the current predicted time is zero, produce in burst and constantly finish prediction, and the adaptively changing built-up time, offset time distributes according to all kinds of weights;

2〉the AAR algorithm that in threshold range, dynamically changes of all kinds of built-up time and offset time, it is long to introduce this bag in the predicted time, is the improvement to pure AR model, and self adaptation estimating burst flow makes that bag is long to adapt to nearest discharge characteristic.

Images