CN1283048A

CN1283048A - Control scheme for constrained queue system

Info

Publication number: CN1283048A
Application number: CN 99111196
Authority: CN
Inventors: 顾钧; 农革
Original assignee: Individual
Current assignee: Individual
Priority date: 1999-07-30
Filing date: 1999-07-30
Publication date: 2001-02-07

Abstract

A control scheme of constrained queue system is disclosed. A constrained queue system NxM has N transmit sources, M destinations, a switch network, a throughput controller and a scheduler. The time of the system is disporsed into timesliccs. The data packets of transmit source are queued in the buffer to wait by queuing for scheduling and transmitting. The invented scheduler features use of priorities of destinations for dynamic scheduling. The high-performance exchanger and router are necessory for actual application.

Description

A kind of controlling schemes of constrained queue system

The invention belongs to a kind of scheme that the information of the network communication system of flowing through is controlled, comprise implementation method and device.

For adapting to current network increasing rapidly on the scale and on the demand capacity, it is with the bottom hardware of Fast Packet switch as router of new generation that a trend is arranged in the recent period.Therefore the utmost point need design a kind of packet exchange machine architecture that satisfies different network environments.The packet switch that adopts this architecture to produce can be deployed in the diverse location from the Access Network to the backbone network.For reaching this purpose, the exchanger system structure of employing must have favorable expansibility.The autgmentability of switch depends mainly on used queuing policy, promptly supplies the residing position of buffering area and the organization mechanism thereof of the usefulness of queuing.Usually, queuing scheme can be divided into input queuing, central authorities' queuing, output queuing or the combination between the former.In various schemes, the input queuing has best autgmentability.Yet it is that efficient controlling schemes of design is come the delay of control data bag through this switch that the switch of an input queuing reaches high performance key.

This paper provides a kind of high-performance data switch controlling schemes in conjunction with flow control and dynamic dispatching.This scheme can be used to control the delay of each packet process input queue switch to obtain expected switch performance.Yet the application of this scheme is not limited only to the input queue switch, and its application includes gauze and wireless network, electronic network and optical networking, or the like.According to these concrete application, we take out a constrained queue system model, and serve as with reference to launching argumentation of the present invention with this model.

The constrained queue system of a NxM has N transmission source and M point of destination.The time of this constrained queue system is unit with the discrete time period.Be scheduled at them and will line up at the buffering area in the source of transmission before being sent to the point of destination through arriving packet (or claiming client) that this system sends the source after the flow control.The ground that is without loss of generality, our tentation data bag are only in the transmission source that begins to arrive this system of time period.Packet behind the arrival transmission source will be dispatched the order that it passes past each corresponding point of destination by our DSDP method in each time period.According to this order, each packet will be sent to corresponding point of destination in the end of each time period.The delivery time of each packet is that unit measures with the time period.As not adding explanation, the delivery time of each packet is assumed to be a time period.The propagation delay of each packet is defined as its time interval that arrives corresponding transmission source and point of destination.And do not have each other can connect simultaneously under the situation at common transmission source and destination station path between a plurality of transmissions source and point of destination, the path that limited interference networks resource is only set up at the same time.Above-described constrained queue system is the abstract model of a series of networks, for example, and the switch of input queuing, wavelength-division optical networking, wireless network or the like.

The packet that sends the source through the difference that arrives this constrained queue system after the flow control at one time is sent to identical point of destination possibly, consequently causes the competition of transmission source to the point of destination.These competitions must be with solving someway.The packet of winning in competition will be sent to their point of destination along the path of setting up.The packet of failing in competition then continues to line up in corresponding buffering area, to wait scheduling next time.Be used for fabric queuing mechanism and to solve the transmission source be the key factor that influences this constrained queue system performance to the competing method of point of destination.Solving a plurality of packets is the main task of the used dispatching method of this constrained queue system in the transmission source to the competition of point of destination.Below, we discuss the queuing mechanism of packet in the buffering area of the source of transmission in this constrained queue system earlier, provide then to constitute the controlling schemes that this constrained queue system of control of the present invention reaches estimated performance in the 5th joint again.

Each packet that sends in the buffering area of source is formed a team by their point of destination, that is is that the packet that will deliver to same point of destination is formed an independently formation.Therefore for the system of a NxM, each sends the source buffering area can M formation at most, and whole system can have NxM formation at most.This bucket-brigade fashion can solve a so-called team blockage problem of puzzlement fifo queue effectively.(in first in first out queueing form, each transmission source is only kept the packet that a formation comes all to be arrived this transmission source buffering is provided in its buffering area.) packet in each formation ranks with their point of destination priority.Therefore file leader's packet has the tidemark station priority of all packets in this formation in each formation.About the specific definition and the purposes of the point of destination priority of each packet, we repeat after a while.

Fig. 1 has provided the constrained queue system of an aforesaid 3x3.

Present stage is existing a plurality of to a kind of dispatching method [

document

1,3,4,6,7,8,9,10,11,12,13,14,15,16,17,18,19] that is referred to as the input queue switch of dummy output terminal queue switch.These existing methods can could be controlled the delay of each packet by switch according to them and be divided into two big classes substantially.

The dummy output terminal queue switch of a NxM in fact also is a constrained queue system.In a dummy output terminal queue switch, each input (that is being the transmission source) has a buffer storage area.Should form by M (logic) formation by (physics) buffering area.Each formation corresponding an output (that is being the point of destination), i.e. private all the packets that hold of this formation coverlet from this transmission source to pairing output.Link to each other with a switching network between the input of this switch and output.The task of dispatching method be each time period Rigen control switching network to set up the path of transmitting corresponding selected packet according to the state of institute's data queued bag.Most of existing method is a target with the throughput that improves switch all, the only several delays that can control each packet by switch of only a few.Yet this SOME METHODS all exists the high and big problem of realization difficulty of complexity.

In the literature, in each time period the scheduling of data queued bag is described as usually according to the current load matrix of system and finds out a pairing matrix.Row and column in load matrix and the pairing matrix is the input and the output of the corresponding dummy output terminal queue switch that is scheduled respectively.Each element of load matrix is that (Wi, Wo), Wi in this dualistic formula and Wo represent the transmission source and destination station priority of a corresponding team packet respectively to a dualistic formula.Be without loss of generality, the element for " 0 " represents its pairing formation for empty.If Wi and Wo always equate that (Wi, Wo) an available scalar W represents dualistic formula.Form for the element of " 0 " or " 1 " by value according to the pairing matrix that a load matrix is found out.In each pairing matrix, every row or every row can have one " 1 " at most.Can be used to control switching network to set up corresponding path by pairing matrix intermediate value for the mapping one by one between the row and column that element constituted of " 1 ".

Looking for the pairing matrix according to load matrix is a typical matching problem [document 2] in the graph theory.Can be divided into maximum, optimum or stable according to its character with the pairing matrix of being found out.Wherein the complexity with the algorithm of finding out maximum pairing matrix is minimum again.Studied clearly quite thoroughly to various matching problems in the graph theory.For it being applied to the dummy output terminal queue switch, main task be how to design have special properties matching scheme to reach desired switch performance.In the matching process of existing transmission source and destination station priority based on packet, the transmission source priority of calculating each packet needs extra memory space and running time, thereby has limited its application under the hypervelocity environment.

The purpose of this invention is to provide a kind of efficient controlling schemes of constrained queue system controlling the performance that this system reaches expection, that is reach the controlled target of performance.

The delay of controlling each packet is to make a constrained queue system reach the fundamental way of this target.For reaching the purpose of the delay of controlling each packet, existing research has been done some effort and has been designed some schemes in the environment of input queue switch.Difference between different schemes is that mainly how the packet state from queuing achieves a solution a plurality of packets to sending the required schedule information of competition at source and destination station.Different with other scheme is that we combine dispatching method with flow control controlling schemes is unique and is creationary.Efficient, high-performance, the utmost point low complex degree that shows mainly that it reached and be easy to physics realization under hypervelocity.

Creatively carry out dynamically in the point of destination priority that each sends the client who lines up in the buffering area of source, maximum is at a high speed matched scheduling by basis for DSDP dispatching method in this scheme.With this dispatching method with combine to arriving the method that the packet respectively send the source carries out flow control, formed our each packet of may command efficient scheme by the delay of switch.Apply it in the actual constrained queue system, can realize at a high speed, performance is controlled and large-scale data switching exchane or router.And actual constrained queue system also comprises other a series of communication system, as wireless network and wavelength-division fiber optic network, or the like.

Particularly, the controlling schemes of our invention by to arrive this system send the source packet flow control and the data package dynamic scheduling two large divisions who waits in the buffering area of the source of transmission formed.The task of flow control is under given system parameters, and control arrives the data packet arrival process that this system sends the source, so that the number of data packets that waits in each sends the source buffering area maintains in the predetermined scope.The task of dynamic dispatching then is in each time period, to the packet of file leader in each formation satisfy following restriction, toward the transmission scheduling of its point of destination: in each time period, each transmission source can spread out of a packet at most can receive a packet at most with each point of destination.At length, in the beginning of each time period, dispatching method satisfies the scheduling of this restriction to team's packet of each formation.In the end of each time period, the packet that the method that is scheduled is chosen will be sent to their point of destination then.Fig. 2 has provided the scheduling in a constrained queue system and the sequential relationship of transfer operation.

To arriving the purpose that the packet respectively send the source carries out flow control, be to make in each sends the scope of client's numerical limitations in hope of waiting in line in the buffering area in source.The task of carrying out this flow control can adopt to comprise that emulation, mathematical analysis or any method that both combine, occur existing and future now realize according to system parameters.The present invention does not add any restriction to method and the form that how to realize flow control, and the task of dynamic dispatching is then realized by our the following method of invention.

When the competition that takes place the point of destination, the priority of the point of destination of each packet has been represented its degree of priority when solving competition, and promptly the packet of limit priority is won.Then can solve in any way the competition between packet with equal priority.The present invention does not add any restriction to how to calculate and compose with its point of destination priority of each packet.Under each packet has been composed prerequisite with its point of destination priority, in each scheduling that team's packet of individual queue is carried out, our dispatching method is found out a scheduling collection that meets the following conditions: suppose that Cij is team's packet that waits to be sent to point of destination j in the buffering area that sends source i, if packet Cij is not at this dispatching concentration, then this scheduling collection comprises such packet at least: this packet or sending source i, and maybe to deliver to point of destination j and have the priority to point of destination j higher than Cij.

Below the flow control to packet in the constrained queue system involved in the present invention, queuing mechanism, scheduling strategy and transport process have been done concise and to the point description.Flow control wherein and scheduling strategy are two parts of the present invention.And scheduling strategy is again the basis of carrying out flow control as one of system parameters.Therefore in following content, we will be further to core of the present invention, and promptly scheduling strategy is made the discussion on the implementation method and provided concrete example.

The dispatching method that design can be used for the dummy output terminal queue switch is the problem to be solved of having to go to the toilet in the recent period.Unfortunately, existing method is not only complicated but also be difficult to realization, thereby causes relatively poor switch performance.This section provides the DSDP dispatching method that can be used for high speed dummy output terminal queue switch.This method has general versatility, and its application is not limited only to the input queue switch, also can be used for including the network of gauze and wireless network, electronic network and optical networking or the like.

At first, we list some in order to simple and clear back with the symbol of discussing:

● Cij: team's packet that in the buffering area that sends source i, waits to be sent to point of destination j.

● DPL[j]: by queuing in all transmission sources and will deliver to the ordered list that team's packet of point of destination j is formed.Each element in this table is variable a: DP.The corresponding team's packet of element in each table.This table thereby N element can be arranged at most.When the packet that will deliver to point of destination j becomes a new team packet, its corresponding element will be added in this ordered list.After a packet is sent to point of destination j, its corresponding element will be deleted from this ordered list.

Our DSDP method is summarized as follows:

The first step: initialization, to set up the priority data structure and to make system be in dispatch state;

Second step: team's packet of corresponding each point of destination is carried out from high to low ordering by its point of destination priority;

The 3rd step:, find out a maximum pairing that sends between the source and destination station according to the state of every team's packet.(pairing is that maximum and if only if that do not have between any unpaired transmission source and destination station can be for the packet of scheduling.) pairing of this maximum meets the following conditions: if packet Cij is not matched, then have at least such packet to be matched: the packet that this is matched or sending source i, maybe to deliver to point of destination j and have the point of destination priority higher than Cij.

More than shown in the 3rd step of DSDP method need to carry out following scheduler task: according to the point of destination priority of the team's packet in the queuing, the process by the circulation pairing sends the pairing between the source and destination station.If a transmission source and a point of destination are matched in scheduler task, a path that connects this transmission source and destination station will be set up in order to transmit a corresponding team packet by the internet that sends between the source and destination station.

For narrating for simplicity, when we said that a packet Cij is matched, its meaning was meant that i transmission source and j point of destination are matched.And when we said the transmission source and destination station of a packet, its meaning was meant that the transmission source at this packet place and this packet will pass past point of destination.In addition, our the transmission source and destination station that defines a packet this packet that is that and if only if freely all is unpaired.The maximum pairing process that is used for DSDP method the 3rd step can realize by following endless form: matched pair technique: in each circulation, if free packet Cij all to be sent to j point of destination, freely

Have supreme good station priority in team's packet, then free packet Cij will be matched.

The DSDP method that provides can realize and be applied in any described constrained queue system with any program language existing now or that occur in the future.

The existing dispatching method based on maximum matching method has following defective:

1. all length of data package are fixed by hypothesis.The maximum matching method of existing point of destination priority based on each packet does not provide the processing to elongated packet.

2. existing scheme based on maximum matching method is in order to control the delay of each packet by switch, except needs utilize the point of destination priority of each packet, also needs to utilize the transmission source priority of each packet.

3. existing method is calculated it according to the point of destination priority of each packet mostly and is sent source priority, and this measure has increased the weight of the computation burden under the high velocity environment, thereby has increased the realization difficulty.

More than three big defectives of listed existing method when being high performance method that is applied under the hypervelocity environment of design must overcome.Our dispatching method only needs the point of destination priority of each packet to dispatch.In addition, with uniform way fixed length and elongated packet are carried out indiscriminate scheduling.

Our dispatching method is represented its point of destination priority to variables D P of each allocation of packets.Be without loss of generality, the point of destination priority of the DP variate-value representative that our hypothesis is more little is big more.The point of destination priority of each packet is represented the order of priority of corresponding packet when solving the point of destination competition.The present invention does not impose any restrictions the point of destination priority that how to define and calculate each packet.

We are to be solved to be how to utilize the point of destination priority of each packet to control its delay by switch.In the solution to this key issue, our method is distinguished.Be in particular on its performance that reaches, mainly contain: 1. hypervelocity: the complexity of a dispatching method is made up of the two large divisions, promptly is matching method and calculate the transmission source and the point of destination priority of each packet.Compare with existing method, because do not need to calculate the transmission source priority of each packet, so our dispatching method has extremely low complexity.In addition, owing to there is not the problem of the transmission source priority that will consider each packet in the pairing process, the complexity of matching method can drop to minimum.2. providing powerful support for service quality: the ability power that service quality is provided support is to estimate a leading indicator of following switch system performance.We combine flow control with dispatching method each packet of scheme may command is by the delay of switch, thereby provides the maximum degree of freedom and support for designing different method for ensuring service quality.3. be easy to realize: compare with other existing scheme, our scheme can further lower the realization difficulty of its dispatching patcher.4. good autgmentability: our scheme can be used for from small to large, the constraint queuing system of different scales, and its complexity only slowly increases with the increase of system scale.5. but distributivity: the form that our method can distribute realizes.It distributes and realizes further reducing realization, expansion and the hot maintenance difficulties of the system of being scheduled.

The present invention is described in further detail below in conjunction with accompanying drawing.

Shown in Figure 1 is one have 3 and send source S1[4], S2[5], S3[6] and 3 point of destination D1[8], D2[9], D3[10] the constrained queue system model.Send between the source and destination station and link to each other with interference networks [7].The square frame of each numbering is represented a packet of lining up.Numbering in the square frame is represented the point of destination of this packet.Formation [1], [2], [3] respectively in order to buffer storage from transmission source [4] to the point of destination packet of [8], [9], [10].Each team packet is the highest priority data bag of its place formation.

The scheduling that shown in Figure 2 is in a constrained queue system and the sequential relationship of transfer operation.As shown in the figure, in the beginning of each time period, dispatching method is dispatched each team packet; In the end of each time period, the team that the method that is scheduled an is chosen packet will be sent to their point of destination then.

The given controlling schemes of the present invention can realize with any way of serial, parallel, streamline, software, firmware, hardware or its combination or the like.Because of the present invention does not add any restriction to method and the form that how to realize flow control in this scheme, so locate only to discuss realization to the dispatching method in the scheme.

In 8.1 to 8.2 joints, we provide that the software to DSDP method described in the 5th joint is realized and to the processing of elongated packet.In 8.3 joints, we provide the architecture Design and the hardware of operation DSDP method and realize.

In (5.2) joint, we have provided a matching method.In the beginning of each time period, this matching method is found out the pairing that sends between the source and destination station.In the end of each time period, team's packet that should match will be sent to their point of destination mutually then.This matching method can have different realizations.Matching method with same performance can be realized with various forms, as long as also matched by the method for this new model by the transmission source and destination station that this matching method matched.Below provide a maximum matching method of the parallel circulation based on the point of destination priority of team's packet.The maximum matching method of parallel circulation

We provide a Parallel Implementation to above-mentioned matching method at this.In the beginning of method operation, all transmission source and destination stations all are unpaired.Two following steps will be repeated until carried out min (N, M) inferior circulation or do not have new pairing to set up:

Step 1: each unpaired point of destination has the transmission source at the packet place of tidemark station priority and sends request in its free packet;

Step 2: each requested transmission source optional one in the point of destination of sending request to it is answered.

In step 2, the point of destination how we do not answer in the request requested transmission source imposes any restrictions.In other words, point of destination in the request can be answered with any specific mode in requested transmission source.For example, at random, by requested packet first in first out or the order that last in, first out, or the like.

In above discussion, we suppose that always all length of data package fix.Therefore it also is constant each team that is scheduled away packet being sent to the required time of its point of destination.Yet in some cases, each length of data package is variable, thereby to cause it also be variable from the transmission source to transmitting the required time the point of destination.

In order to reduce the realization difficulty of the system of being scheduled, it is continuous that the scheduling of elongated packet is required the transport process of each packet usually.That is be to any packet Cij, in case the transmission of past its point of destination of beginning that is scheduled, then its transport process can not be interrupted by other packet from transmission source i to point of destination j.Yet the transport process of packet Cij can be interrupted by packet with the exception of this.

For reaching above-mentioned scheduling requirement to elongated packet, our DSDP method only need be done shown in the back, minimum change: be scheduled and will be sent to its point of destination when certain data is wrapped between the some time section, if the time has begun its transmission and has not finished as yet to have data to be wrapped in more early with this packet in addition with formation, then the latter continues its transport process, and the former will wait for next time and be scheduled.

Realize that the DSDP dispatching method can have multiple software and hardware architecture structure, can not enumerate one by one.It is following that what lift only is that a typical case realizes.Any realization and not requiring to the DSDP dispatching method be subjected to this typical case realize limit.

In the beginning of dispatching method operation, all transmission source and destination stations all are unpaired.The DPP processor is all M DPL tables of M point of destination structure according to the point of destination priority of each team packet.The device of a sorting unit by name will carry out from high to low ordering by its point of destination priority to the team's packet in each DPL table, that is, according to the DP variate-value in each ternary formula sort from low to high (the point of destination priority of the more little DP variate-value of our hypothesis representative is big more).This sorting unit can any software, the form of firmware, serial or parallel hardware realizes.In case after M PL table was established, in the time thereafter, we can come to finish with the one or more form of each time period adding joining a new team packet in the DPL table with the y-bend insertion.

Then, pair matcher will send the pairing between the source and destination station according to the DP variate-value of each team packet.Press

According to the pairing result, interference networks will be set up the path that sends between the source and destination station and transmit the team's packet that is matched.

The present invention has provided a kind of controlling schemes of constrained queue system.This kind constrained queue system is the abstract model of a series of networks, for example, and switch, wavelength-division optical networking, wireless network and wired network of input queuing or the like.Compare with other existing scheme, our scheme be the more important thing is to be easy to realize under the hypervelocity environment in that to guarantee that the Be Controlled system has under the prerequisite of superperformance not only more flexible.

Claims

A kind of scheme that client in the constrained queue system is controlled that above chapters and sections have been invented us.This constrained queue system is the abstract model that a series of network communication systems comprise switch, router, wavelength-division optical networking, wireless network and the wired network etc. of input queuing.Compare with other existing scheme, our scheme is in that to guarantee that the Be Controlled system has under the prerequisite of superperformance not only more flexible, the more important thing is to be easy under the hypervelocity environment, realize, thereby can be in order to realize the network communication apparatus and the system of high performance price ratio.

Under the prerequisite of knowledge that possesses this area and skill, can make multiple variation and change according to core concept of the present invention and scope.Therefore, related concrete form in our narration not can be used as the restriction to following claim only as helping to understand.

The word that we are used in the description of this invention not only comprises the implication that it is common, also comprises any special implication with regard to content of the present invention.Therefore, as any used word multiple special implication can be arranged in context of the present invention, then its appearance has in the claims also comprised implication that it is common and all the special implications in the present invention.

Used word or meaning group in following claim not only comprises its form herein, comprises that also all are in meaning and reach other the same form of essence to the effect.Therefore, any word in the following claim or meaning group can be different by form and other word or meaning group that essence is the same replace.

Under the prerequisite of the knowledge that possesses this area and skill (comprise known with occur in the future), the nonessential modification to content related in the claim has been done also is regarded as in the scope of work claim.Therefore, alternative also being regarded as in the scope of work claim that any content in the claim is done.

Therefore, the claim done of this place has also covered above-described explanation: comprise different concrete forms, identical notion, to word or meaning group substitute and in essence to the utilization of essential idea of the present invention.

We require:

(1) a kind of scheme that the client of the constrained queue system that arrives a NxM is controlled, it is characterized in that: the constrained queue system of a NxM has N transmission source, but the interference networks of a M point of destination and a dynamic-configuration, the path between a plurality of transmissions source and point of destination that do not have each other can connect simultaneously under the situation at common transmission source and destination station, the path that limited interference networks resource is only set up at the same time, the time of this constrained queue system is unit with the discrete time period, be scheduled at them and will line up at the buffering area in the source of transmission before being sent to the point of destination through arriving client that this system sends the source after the flow control, dispatching method will the beginning of each time period according to individual queue in the file leader client the point of destination priority scheduling its pass order toward each corresponding point of destination, according to this order, each client will be sent to corresponding point of destination in the end of each time period.
(2) queuing of client in the buffering area in the source of transmission in the described constrained queue system of claim 1, it is characterized in that: each client who sends in the buffering area of source forms a team by their point of destination, that is be that the client that will deliver to same point of destination forms an independently formation, client in each formation sorts according to their point of destination priority, and each team packet has the limit priority of client in this formation.
(3) the point of destination priority of packet in the described individual queue of claim 2, it is characterized in that: when the competition that takes place the point of destination, the point of destination priority of each packet has been represented its degree of priority when solving competition, the packet that is limit priority is won, then can solve in any way the competition between the packet with equal priority, method and the form calculating and compose with its point of destination priority of each packet are not subjected to any restriction.
(4) flow control in the described controlling schemes of claim 1, it is characterized in that: to arriving the purpose that the packet respectively send the source carries out flow control, be to make at each sending the client's who waits in line in the buffering area in source numerical limitations in the scope of wishing arbitrarily, the task of carrying out this flow control can be according to system parameters, employing comprises that emulation, mathematical analysis or any method that both combine, occur existing and future now realize, realize that the method for this flow control and form are not subjected to any restriction.
(5) in the described controlling schemes of claim 1 to the scheduling of every team's packet, it is characterized in that: in each scheduling that team's packet of individual queue is carried out, find out a scheduling collection that meets the following conditions: suppose that Cij is team's packet that waits to be sent to point of destination j in the buffering area that sends source i, if packet Cij is not at this dispatching concentration, then this scheduling collection comprises such packet at least: this packet or sending source i, and maybe to deliver to point of destination j and have the priority to point of destination j higher than Cij.
(6) constrained queue system in the described controlling schemes of claim 1, it is characterized in that: this constrained queue system is the abstract model that a series of network communication systems comprise switch, router, wavelength-division optical networking, wireless network and the wired network etc. of input queuing, and the client's correspondence in this constrained queue system is by the packet in the abstract network communication system.
(7) packet in the described network communication system of claim 6, it is characterized in that: each length of data package can be a fixed length.
(8) packet in the described network communication system of claim 6, it is characterized in that: each length of data package can be elongated.
(9) dispatching method in the described controlling schemes of claim 1 is characterized in that: reach fast dispatch to client with following three steps:

The first step: initialization, to set up the priority data structure and to make system be in dispatch state;

Second step: team's packet of corresponding each point of destination is carried out from high to low ordering by its point of destination priority;

The 3rd step:, find out a maximum pairing that sends between the source and destination station, this maximum pairing according to the state of every team's packet

Meet the following conditions:, then have at least such packet to be matched if packet Cij is not matched:

The packet that this is matched or sending source i maybe will be delivered to point of destination j and have a point of destination higher than Cij preferential

Level.
(10) the 3rd in the described dispatching method of claim 9 step, it is characterized in that:, find out one with any one maximum matching method and meet maximum described condition, that send between the source and destination station and match according to the state of every team's packet.
(11) the described maximum matching method of claim 10, it is characterized in that: a kind of maximum matching method of circulation is as follows: in the beginning of pairing process operation, all transmission source and destination stations all are unpaired, thereafter in each circulation, if free packet Cij all to be sent to j point of destination, in team's packet supreme good station priority is arranged freely, then free packet Cij will be matched, carrying out min (N, M) inferior circulation or finish this pairing process when not having new pairing to produce.
(12) the maximum pairing process of the described circulation of claim 11, it is characterized in that: the Parallel Implementation to this maximum pairing process is: in the beginning of method operation, all transmission source and destination stations all are unpaired, two following steps will be repeated until carried out min (N, M) inferior circulation or do not have new pairing to set up:

Step 1: each unpaired point of destination has the sending out of packet place of tidemark station priority in its free packet

Send the source to send request;

Step 2: each requested transmission source optional one in the point of destination of sending request to it is answered.
(13) step 2 in the described Parallel Implementation of claim 12, it is characterized in that: the point of destination how answer in the request in requested transmission source is not subjected to any restriction, that is point of destination in the request can be answered with any specific mode in requested transmission source.
(14) first step of the dispatching method of narrating according to claim 9 is characterized in that: (ⅰ) set up a metadata structure that stores corresponding to the priority information of each team packet: DP, (ⅱ) set up the data structure of creating each point of destination priority list.
(15) second of the dispatching method according to claim 9 step, it is characterized in that: all teams packet is divided into groups by point of destination under it, the team's packet in every group is carried out from high to low ordering to constitute a point of destination priority list by the priority of its point of destination.
(16) sorting operation according to claim 15 is characterized in that: sorting operation available software, firmware, hardware or its make up to be realized.
(17) foundation to a point of destination priority list according to claim 15 is characterized in that: a kind of method is to come a new team packet that will join in the point of destination priority list is finished with the one or more form of each time period adding with the insertion ranking method.
(18) according to claim 17 a new team packet is added to the method for its corresponding point of destination priority list, it is characterized in that: can use any insertion ranking method, comprise that y-bend inserts ranking method.
(19) operation according to claim 9 is characterized in that: each operation not only can be carried out the fixed-length data bag, also can be applied on the elongated packet.
(20) application to elongated packet according to claim 19, it is characterized in that: a kind of method is to be scheduled in certain time period will be sent to its point of destination the time when data wrap in, if there is one to begin its transmission with the packet of formation early than this time period but transmit as yet to finish in addition with this packet that is scheduled, then this another packet continues its transport process in this time period, and this packet that is scheduled is then waited for next time and is scheduled.
(21) operation of narrating according to claim 9 is characterized in that: each mode of operating available serial, parallel, streamline, software, hardware, firmware or their combination realizes.
(22) implementation of narrating according to claim 21 is characterized in that: this implementation can be concentrated, and also can distribute.
(23) controlling schemes of narrating according to claim 1, it is characterized in that: each client of this controlling schemes may command can be used for designing various transmission to packet and carries out the method that service quality guarantees by the delay of this constrained queue system in the network communication environment.
(24) controlling schemes of narrating according to claim 1, can design and realize the network communication apparatus that to control the delay of the packet on it of flowing through, it is characterized in that: this kind equipment of a NxM has N transmission source, M point of destination, but the interference networks of a dynamic-configuration, a flow controller and a scheduler, at length: the path between a plurality of transmissions source and point of destination that do not have each other can connect simultaneously under the situation at common transmission source and destination station, the path that limited interference networks resource is only set up at the same time, this equipment is that unit carries out synchronous operation with the discrete time period, be scheduled at them and will in buffering area, independently form a team the point of destination by them before being sent to separately point of destination through arriving packet that this equipment sends the source after the flow control, client in each formation sorts according to their point of destination priority, each team packet has the limit priority of client in this formation, scheduler will it passes the order toward each corresponding point of destination according to the point of destination priority scheduling of every team's packet in the beginning of each time period, according to this order, each packet will be sent to corresponding point of destination in the end of each time period.