CN103888379A - Improved queue scheduling algorithm based on trusted scheduling - Google Patents

Abstract

The invention discloses an improved queue scheduling algorithm based on trusted scheduling. The provided queue scheduling algorithm is mainly composed of two parts: a flow controller and a packet scheduler. The flow controller adopts a token bucket algorithm, which is most commonly used for distinguishing service networks, to mainly control flow of EF PHB business so as to control scheduling of the businesses with high priority and prevent starving of queues with low priority caused by long time service of queues with high priority. The packet scheduler is realized via combination of the improved algorithm DWRRI based on a weighted cycle scheduling algorithm and a priority algorithm primary rate interface.

Description

A kind of improved queue dispatching algorithm based on credible scheduling

Technical field

The present invention relates to communication technical field, a kind of improved queue dispatching algorithm based on credible scheduling of special design.

Background technology

Differentiated Services a kind of IP QoS model that IETF working group proposes for the poor expandability of the integrated service model of proposition before improving just, it has simplified signaling, sorting particles degree to Business Stream is thicker, and industry generally believes that differentiated service is feasible IP QoS model in a kind of future.The feature that differentiated service is implemented is to adopt stream flowcollector aggregation scheme FlowCollector that some Business Streams with identical characteristics are aggregating, for whole aggregated flow provides service, and no longer towards single stream.Therefore, be in the borderline router of differentiated services network and keep every flow state information, and core router is only responsible for forwarding data bag and non-hold mode information, thereby autgmentability is better, can be used in catenet.

In differentiated services network, a network has multiple different different service domains, and different different service domains can provide different Differentiated Services business.In typical differentiated services network, each edge router is connected with the router of other different service domains, is responsible for shaping and strategy to convergence service stream, and the preservation of flow state information is also provided.Between each DS territory, the connection between Ye Shi operator, can carry out the negotiation between service quality rating SLA conventionally.Distinguish each class communication according to the QoS mechanism of implementing, require it to classify and flow control according to business service at edge router place, and provide service to each business, comprise the reserved and queue scheduling of allocated bandwidth etc.Core router is generally positioned at Differentiated Services Autonomous Domain, the general only high speed forward of responsible packet, and this is a significant difference of Differentiated Services and Integration Services.Integration Services provides service for single stream, but Differentiated Services converges the stream with identical characteristics business, and it doesn't matter with service condition, so just greatly reduce the handled number of queues of dispatching algorithm, therefore the consumption of route resource is also reduced greatly, have than the better autgmentability of integrated service.In core router, define one or more groups every jumping route, mainly forwarded according to the different choice particular schedule strategy of DSCP value, guaranteed piecemeal the behavior of every jumping route at each span line.

User communicates via differentiated services network, needs through following process:

1. user is first to self connected boundary node transmission data, boundary node is responsible for carrying out business classification and flow adjustment at network edge, business classification can also can be based on other features based on DS territory, and the means such as flow adjustment is used measurement, mark, abandons, shaping realize;

2. boundary node identifies after business, mark in IP stem dscp field according to type of service, according to the every jumping route behavior for miscellaneous service defines of different DSCP values, then to the core node forwarding data in this DS territory, the service in DS region provides strategy to be determined by PHB, DS node forwards according to PHB attribute, and the aggregated flow that just a large number of services stream with identical DSCP value is pooled to minority is carried out simple forwarding according to the PHB corresponding with DSCP to message.

3. if desired via different Diffserv forwarded, internal core node is forwarded to after the boundary node of this DS territory, boundary node and other Diffserv network boundary node transmission of informations, because can there be different PHB in different DS territories, provide strategy to realize different services, between boundary node, need to provide through consultation trans-regional service, SLA is service level agreement, consult the treatment that should obtain while transmission in network about Business Stream, TCA is flow adjustment agreement, consults about business sorting criterion, business model and respective handling.

4. through after repeatedly forwarding, packet is joined Diffserv network boundary node by objective network and is forwarded to objective network.

Summary of the invention

Technical problem solved by the invention is:

The present invention mainly studies the scheduling strategy of QoS under DiffServ, proposes a kind of improved queue dispatching algorithm based on credible scheduling.Packet scheduling algorithm in credible scheduling model is take weighted round-robin algorithm as basis, there is the fixing shortcoming of weights for DWRR (Deficit weighted Round Robin) algorithm, a kind of improved algorithm DWRRI (DWRR Improvement) has been proposed, it changes the weights of giving it for the flow velocity of packet in network, compensate its defect thereby realize paying when a small amount of cost guarantees DWRR advantage.

In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is:

Improved queue dispatching algorithm based on credible scheduling, this queue scheduling algorithm is mainly made up of two parts: flow controller and packet scheduler.Flow controller adopts the most frequently used token bucket algorithm of differentiated services network, and the flow that is mainly used to control EF PHB business, to control the scheduling of high-priority service, prevents from serving for a long time high-priority queue and causes Low Priority Queuing hungry to death.Packet scheduler combines to realize by the improvement algorithm DWRRI on WRR algorithm basis and priority algorithm PRI.In the time that Business Stream grouping arrives, first enter edge router, if grouping is marked as EF business, it must be first by the filtration of token bucket, will be proceeded to high-priority queue and forwarded if meet traffic requirement, otherwise wait for and forwarding forwarded to scheduling that other queues receive DWRRI algorithm by mark again.If the grouping arriving is marked as AF business, first it will accept the scheduling of DWRRI queue scheduling algorithm, then sends to the queue of low priority to wait in line to be forwarded.Combination like this, can make full use of the advantage of repeating query and priority algorithm, overcomes shortcoming separately, meets as much as possible time delay and fairness, supports better Differentiated Services.

The invention has the beneficial effects as follows:

First this algorithm guarantees the operation of high-priority service, then allow other business contention bandwidth, in the time that certain class service rate exceedes the flank speed of its regulation, algorithm reduces gives its weights, thereby reduces it to the taking of bandwidth, in addition, can make full use of the advantage of repeating query and priority algorithm, overcome shortcoming separately, meet as much as possible time delay and fairness, support better Differentiated Services.

Accompanying drawing explanation

Fig. 1 is the processing procedure of the present invention's token bucket in flow controller design.

Fig. 2 is the present invention's DWRR algorithmic dispatching schematic diagram in packet scheduler design.

Fig. 3 is this method DWRR algorithm function block diagram in packet scheduler design.

Fig. 4 is the tactful schematic diagram of the improved queue dispatching algorithm based on credible scheduling in the present invention.

Embodiment

Below in conjunction with Figure of description, the present invention is further illustrated.

The design of flow controller and packet scheduler be achieved as follows:

1. flow controller

Traffic policing is the specification that supervision enters a certain flow of network, within it is limited in to a rational scope, or the flow that exceeds restriction is punished, with the interests of protecting network resource and operator.Common way is to use CAR (Committed Access Rate: agreement access rate) to limit the flow of certain class message.CAR utilizes token bucket to carry out flow control, just continues normal forward process if meet traffic requirement for the flow that enters token bucket, if there is no enough tokens in token bucket in order to send message, dropping packets.Realize the flow control to certain class message by the token of continuous generation like this.Herein, for arrive time token bucket do not had token in order to send flow be not to adopt the principle directly abandoning, but these flows are carried out to the mark again of DSCP value, it is designated as to AF class flow from EF flux scale, allow the flow competition network resource of these flows and normal priority business, so just limited the delay that the maximum flow of high-priority service has also avoided the bag because directly abandoning excess flow to cause simultaneously.

What Fig. 1 showed is the course of work of token bucket.The generation speed of token can be set according to user's demand, the capacity of token bucket also can arrange, and in the time that token produces the capacity that exceeds to a certain extent bucket, token amount no longer increases, after grouping is processed by token bucket, if there are enough tokens in order to send grouping, to divide into groups to be sent by continuation, correspondingly in bucket, in bucket, token number does corresponding minimizing by the block length of processing, when token in bucket can not send when grouping very little again, grouping will be by mark again, and in bucket, token number is constant.So just limit the packet count that high-priority service can only be allowed at most to send all token quantity in bucket, when high-priority service sudden when little, in bucket, token quantity is enough used for sending data, when high-priority service sudden excessive or while having malicious stream preempting resources, at most only allow to send the packet count that in bucket, all tokens represent, after token is finished, only have and wait until that new token generation just can continue to send, the transmission speed of controlling like this EF Business Stream with regard to having realized is less than or equal to token formation speed, reach the object of the EF service traffics of restriction high priority.

Here introduce two major parameters of token bucket:

CIR (Committed Information Rate): committed information rate, represents to the speed of throwing in token in token bucket, i.e. port Mean Speed for a long time.

CBS (Committed Burst Size): Committed Burst Size, this parameter determines the maximum burst size before flow exceeds CIR, that is the capacity of token bucket.This parameter must arrange the maximum length that is greater than message.

In fact also have an implicit parameter TI (Time Interval), this value is CBS/CIR, and it mainly defined to the time interval that adds token in token bucket.Join in token bucket with regard to having new token every the TI time, the quantity adding is the size of CBS.

For committed information rate CIR, on the interface of any type, minimum particle size is all 8000bit, and unit is bit/s, and minimum speed can only be restricted to 8000bit/s.For committed burst flow CBS, unit is bit, should not be less than the maximum length of message.Therefore CIR is set to 100Kbit/s, and CBS is set to 2000kbyte.It is contemplated that token bucket is a bucket, its capacity is 2000L, and bottom Water discharge flow speed is 200L/s, and top water inlet flow velocity is 100L/s, obviously unable to make ends meet.In 20s, inner bucket water has flowed, and at this moment Water discharge flow speed becomes 100L/s, and before 20s, flow velocity keeps 200L/s always, burst that Here it is.Just can control burst flow by the CBS parameter that token bucket is set.

2. packet scheduler

DWRR algorithm is improved by WRR algorithm, and WRR is mainly suitable for, in network that point group leader is fixing, deciding the sending order of grouping when scheduling according to the weights size of setting, and in a round robin cycle, must transmit packet in strict accordance with the weights that are assigned to.In the network of the such fixed packet length of ATM, can reach complete justice, but in differentiated services network, packet between individual queue is convergence flow, business is all to transmit based on IP, data package size normal range (NR) is in 64-1518 byte, the bandwidth that the grouping that length is longer takies is more, thereby affects fairness.DWRR has introduced the concept of difference counter, during by scheduling, becomes take byte as unit by forwarding grouping, no matter the block length packet size how at every turn scheduling can forward is fixed, has overcome the elongated unjustness bringing that divides into groups.

The explanation of giving one example: as shown in Figure 2, queue scheduling algorithm has three to treat scheduling queue, and weights are respectively 200,300,100, and first queue has three packets to be sent, and length is respectively 100,100, and unit is byte; Second queue has two packets to be sent, and length is respectively 200,150 bytes; The 3rd queue has three packets to be sent, and length is respectively 50,100,100 bytes.When queue is serviced, first weights are added to DCi upper, each queue of repeating query successively.Queue 0 weights are 200, now DC ₀be 200, queue header data length is that 100 bytes are less than DC ₀therefore scheduler forwarding data, simultaneously DC ₀become 100, be greater than 0, the second long data packet 100 and just equal DC ₀continue to forward, queue 0 is empty, repeating query queue 1.Queue 1 header data bag long 200 is less than DC ₁, forward, simultaneously DC ₁become 100, be greater than 0, but second data packet length is 150 to be greater than DC ₁, not sending this packet, repeating query queue 2, forwards header data bag DC equally ₂become 50, then no longer send packet below, start second and take turns circulation.When the second repeating query, because queue 0 has been empty, no longer, to its scheduling, all the other queue weights are added to DC _iupper, now DC _i=400, DC ₂=150, queue 1 can complete scheduling, and queue 2 remains the packet of 100 bytes must wait for that next repeating query sends.

Can be found out by upper example, DWRR roughly retrains the output size of packet with weights, guaranteed fairness, because the data unit forwarding is byte, therefore supports elongated packet simultaneously, makes scheduling more accurate.But same shortcoming is also clearly, in upper example, queue 1 one take turns scheduling after an only surplus packet to be sent, but it is still very large to distribute to its weights, adaptivity is very poor, for the distribution underaction of Internet resources.DWRRI in this paper improves algorithm has increased a right value update module in order to improve this shortcoming just on the basis of DWR, makes it avoid defect paying in guaranteeing advantage on certain complexity cost basis as far as possible.

The basic thought of DWRRI is to increase a measuring appliance and decision device to dispatching algorithm, dynamically changes the weights of respective queue according to the result of measuring appliance.Its integral module figure as shown in Figure 3.This module is made up of measuring appliance, decision device, queue weight table and common repeating query queue scheduling.

In DWRRI algorithm block diagram, (this grouping is through DSCP mark module mark classification above in every grouping, be only AF1-AF4 class business and BE business through the grouping of this module) herein, according to the data arrival rate of the outcome measurement corresponding with service classification of classification, this grouping is inserted in corresponding subqueue simultaneously, the result of measuring appliance is sent to decision device, determine how to change weights corresponding to each service type in queue weight table for decision device, thereby affect scheduling behavior.

QW _irepresent the initialization weights of subqueue i, dispatched each time QW _ithe actual weights W of subqueue i when next time dispatching _i.Definition SW _ifor distributing to the fixing weights of queue, Δ W _irepresent, with respect to the increment of fixing weights, to be determined QW according to measurement result by decision device _i=SW _i+ Δ W _i, R _minirepresent the lower limit of subqueue speed, can determine according to real network situation, correspondingly, R _maxifor queue rate-limit.

1, determine QW _i;

In the time that a grouping arrives, first will divide into groups to insert in corresponding queue according to the DSCP value of grouping, then measuring appliance is measured the speed R that such business arrives, and decision device moves as follows according to R value:

If a. R < R _mini, Δ W _i=0;

If b. R _mini< R < R _maxi, $\mathrm{\&Delta;Wi}=\frac{R-R_{\min i}}{R_{\max i}-R}\max \left(\mathrm{\&Delta;Wi}\right)$

If c. R > R _maxi, Δ W _i=max (Δ W _i);

d.QW _i＝SW _i+ΔW _i，

In order to guarantee each traffic differentiation, for max (the Δ W of each subqueue i _i) must limit to some extent, with guarantee weight increase after can not impact other business, therefore max (Δ Wi) is because there being a upper limit.For the R of each subqueue _minand R _maxoptional different value, for the larger subqueue of weight, can make its R _minand R _maxless and make max (Δ W _i) larger, Δ Wi changes comparatively fast like this, can meet the demand that its data speed is larger.

2, algorithm operating procedure;

The QW changing each time _ivalue all will be written in queue weight table and go, and scheduler often completes after once scheduling, read QW corresponding in table _ithe actual weights of individual queue i are then dispatched when dispatching next time, and concrete steps are as follows:

A. initialize queue weight table, carries out assignment: W according to corresponding weights in table to each subqueue weights _i=QW _i;

B. when grouping arrives, affecting decision device according to measuring appliance result changes Δ Wi and court verdict is write to queue weight table;

C. from table, read weights, W _i=QW _i+ Δ W _i

D. dispatch each subqueue by DWRR, byte W of every scheduling _isubtract 1, until W _i=0 or queue serve next queue for idle running;

E. one take turns finishing scheduling, go to step c.

More than show and described basic principle of the present invention and principal character and advantage of the present invention.The technical staff of the industry should understand; the present invention is not restricted to the described embodiments; that in above-described embodiment and specification, describes just illustrates principle of the present invention; without departing from the spirit and scope of the present invention; the present invention also has various changes and modifications, and these changes and improvements all fall in the claimed scope of the invention.The claimed scope of the present invention is defined by appending claims and equivalent thereof.

Claims (1)

1. the improved queue dispatching algorithm based on credible scheduling, it is characterized in that: the improved queue dispatching algorithm of this credible scheduling is mainly made up of two parts: flow controller and packet scheduler, flow controller adopts the most frequently used token bucket algorithm of differentiated services network, the flow that is used for controlling EF PHB business is to control the scheduling of high-priority service, preventing from serving for a long time high-priority queue causes Low Priority Queuing hungry to death, packet scheduler combines to realize by the improvement algorithm DWRRI on weighting loop scheduling algorithm basis and priority algorithm primary rate interface, in the time that Business Stream grouping arrives, first enter edge router, if grouping is marked as EF business, it must be first by the filtration of token bucket, if meeting traffic requirement will be proceeded to high-priority queue and forwarded, otherwise the scheduling that is forwarded to other queues by mark again and receive DWRRI algorithm is waited for and being forwarded.If the grouping arriving is marked as AF business, first it will accept the scheduling of DWRRI queue scheduling algorithm, then sends to the queue of low priority to wait in line to be forwarded.

Images