4、Qos拥塞管理

Congestion is caused by a lack of network resources and falling rate、The introduction of additional delay a phenomenon.Congestion causes the transmission delay of a message、Low throughput, and resource cost of.而在IPPacket switching and more business coexist under the complex environment of,Congestion is very common.Such as wan bandwidth usually smaller than the bandwidth of the LAN,当一个局域网的用户向另一个局域网的用户发送数据时,由于广域网的带宽小于局域网的带宽,数据将不可能按局域网发送的速度在广域网上传输.

Congestion management by specifying packet scheduling order to ensure the high priority priority was dealing with.

4.1、generation of congestion

Congestion is the main scenes：

• 速率不匹配：报文从高速链路进入设备,再由低速链路转发出去;
• 汇聚问题：Message from multiple interfaces into the equipment at the same time,由一个没有足够带宽的接口转发出去.
  
  The negative impact of congestion happening:
  (1)、Increases the packet transmission delay and jitter;
  (2)、过高的延迟会引起报文重传;
  (3)、Make effective network throughput is reduced,造成网络资源的损害;
  (4)、Play a lot of network resources（特别是存储资源）,不合理的资源分配甚至可能导致系统陷入资源死锁而崩溃.

4.2、Overview of Congestion Management

Congestion management is refers to the network in the case of congestion use队列技术,将从一个接口发出的所有报文放入多个队列,使用不同的调度算法来发送队列中的报文流.

4.2.1、Network device queue constitute

什么是队列？
The queue is in the cache for message sorting logic.当流量的速率超过接口带宽或超过为该流量设置的带宽时,报文就以队列的形式暂存在缓存中.报文离开队列的时间、顺序,以及各个队列之间报文离开的相互关系则由队列调度算法决定.

当网络发生拥塞时,Network equipment without forward to complete the rest of the packet,This part of the packet is stored in the network equipment in memory,按照一定的规则排列.Each queue can be arranged10个1500字节的数据包.And therefore the number of units is usually a message queue rather than bytes message.

Network equipment in the software and hardware queue queue

In the network equipment of the outlet,By one of the most commonly basic queue,该队列以FIFO（First IN First Out）的方式调度,We put the queue in the hardware queue.In the hardware queue before there is aQoSQueue management and control software,拥塞发生时,Will be stored in a memory packet to be transmitted form the queue,Wait for scheduling algorithm to send queue the message flow.

4.2.1.1、Queue composition of routers

When the network congestion happens,The interface into the message routing equipment directly into its own hardware queue,send queue according toFIFO(先进先出)forwarded in order.

当拥塞发生时,The message into the router and their own equipment message queue cache into the software,By scheduling algorithm for forwarding.

The router equipment in the cache queue is divided into two types：
(1)、系统队列
router network,Maintain equipment operation by the hardware network,保障网络连通性,The accessibility of the various control protocol,组成的队列.Queue system is divided into two kinds of：

• 紧急队列：Link control layer of various protocols(PPP协议等)
• 协议队列：路由协议(OSPF、RIP、ISIS、BGP等)

(2)、用户队列
From router interface into the flow of various business users use.Routers commonly used queues areFIFO、PQ、CQ、RTPQ、WFQ、CBQ、CBRWaiting queue technique,默认使用FIFO队列.

4.2.1.2、Switch of queue

交换机工作在二层,The traffic flow data processing,带宽高,High transmission efficiency is required,Therefore switch using high-speed hardware directly with,Implemented by chipQoS队列.这种方式下,Switches use head identification data type on the second floor,Main flow is used802.1p的VLANmessage-based,and from the third floorDSCP的流.

Switches according to the priority rules of trust,查找QoS映射表,The mapping results to internal message priority tag,Message is mapped to a local queue.

A message in the port,According to the mapping table into the local queue;
The message is on the outgoing port,After sending queue scheduling;
常用的队列：SP、WRR、HWFQ.

4.2.2、Router Congestion Management

4.2.2.1、FIFO队列

FIFO(First In First Out Queuing,先入先出)All message to forward to the time sequence of the interface,Send a first-in, first-outFIFOThe queue messages.

优点：处理简单,开销小.
缺点：没有公平性,Between different flow is not isolated from each other,The bandwidth of the key business is not guaranteed.

FIFO的配置：
[sysname-Ethernet1/0/3] qos fifo queue-length queue-length
queue-length参数指定了FIFO队列的长度,Is the largest can accommodate the number of message.增加FIFOThe length of the queue can reduce packet loss,But at the same time also can increase the delay.

4.2.2.2、PQ队列

PQ(Priority Queuing,优先队列)调度,Is in strict accordance with the queue priorityhigh and low order进行调度.只有高优先级队列中的报文全部调度完毕后,低优先级队列才有调度机会.

采用PQ调度方式,Put delay sensitive key business in high priority queue,The non-critical business into the low priority queue,To ensure that critical business is preferred to send.

PQ队列的4个队列：
Top（High priority queue）、Middle（中优先队列）、Normal（normal priority queue）和boottom（low priority queue）.默认是Normal（normal priority queue）.

二层交换机配置PQ调度后,按照7、6、5、4、3、2、1、0In turn from high to low priority schedule,发送报文.

PQ队列原理：

• Queue by category,Discard after the queue is full
• The high priority queue scheduling first
• 加入top队列一直是满的,Leads to the low priority queue,never been scheduled（Normalthe default queue）

PQ队列特征：

• 可以使用ACL对报文进行分类,According to the need to message into the queue;
• Packet strategy using tail discarded(Tail Drop)机制,And only this one mechanism;
• 队列长度可以设置为0,Indicates that the queue is infinite,The access to the message queue will not beTail Drop机制丢弃,unless you run out of memory;
• 队列内部使用FIFO逻辑;
• When from the queue scheduling packet,From high priority queue to low priority scheduling message.

PQ队列缺点：

• queue starvation：PQScheduling mechanism can make the lower priority queues the message not scheduling chance.High priority queue there's always a message there,The low priority queue message delivery has been can not get the service.
• UDP流量：If high priority deliveryTCP流量,Low priorityUDP流量,则TCPincrease transfer rate,导致UDPFlow can't get enough bandwidth.

1、PQQueue Configuration Matters：

• 配置PQL（priority queue list）
  • configured by default,Queue that traffic enters
  • 配置PQQueue classification rules（acl）
  • 配置PQLThe length of the queue
  • Default queue configuration
• 将PQL应用到接口

2、Queue that traffic enters：
No corresponding configuration rules into the default queue by message：（The default queue isPQ的4one of the queues,是普通队列）
[RA] qos pql pql-index default-queue{bottom | middle | normal | top }

配置PQQueue classification rules：

• Configuration based on the classification rules of the interface：
  [RA] qos pql pql-index inbound-interface interface-type interface-number queue {bottom | middle | normal | top }

• Configuration based on the classification rules of the agreement：
  [RA] qos pql pql-index protocol ip [queue-key key-value] queue {bottom | middle | normal | top }
  

3、配置PQLThe length of the queue
[RA] qos pql pql-index queue {bottom | middle | normal | top } queue-length queue-length

4、在接口上应用PQL
[RA-Ethernet1/0/0] qos pqpql pql-index

5、PQ队列信息
[RA] dis qos pql 5
[RA] dis qos interface Ethernet1/0/0

4.2.2.3、CQ队列

CQ(Custom Queuing,自定义队列),According to certain rules are grouped into16类(对应于16个队列),Grouped according to their own categories according to the first in first out strategy to enter the correspondingCQ队列.

CQQueue scheduling using polling mode.According to the pre-configured lines in turn from1到16号用户队列中取出一定数量的报文发送.If one queue polling is empty,Then go into the next queue.This security key business to get more bandwidth,And non-critical business also can get the corresponding bandwidth.

接口拥塞时,Message was sent into corresponding queue according to the matching rules;如果报文不匹配任何规则,Has been sent to the default queue.The default queue defaults toCQ的队列1,可以修改.

• Queue by category,Discard after the queue is full,可配置为WRED.
• Each queue polling schedule
• 其实有17个队列,0Is used to queue system,用户不能修改.
• By default each queue1500个队列（阈值）

缺点：Because of using polling scheduling each queue,CQEnsure that any data flow delay.

CQ队列调度：

上图中,First determine the queueN(N从1开始)The current sending quota of(NQCS)The current is greater than the queue first enqueued message length(PL),If more than to send a message,Then send the queue to send current limit minus the length of a message（NQCS=NQCS-PL）.Then the next queue queue until next send amount is less than before at length when ready(NQCS<PL),put the current queueNSending limit and configuration values(NQCS=NQCS+NQDS),to the next queue(N+1)调度(N=16时,下一队列N+1=1).

这样CQQueue in accordance with the order queue scheduling,当某个队列为空时,Polling the queue scheduling.因此CQScheduling mechanism can automatically increase the existing categories, can account for the bandwidth of a message.It can also be seen from the above,One of the queue to send quota after use,Need to wait for other queue send line after use,Will dispatch to the message again,所以CQQueue time delay is larger,Not suitable for time sensitive business.

CQQueue Configuration Matters：

• 配置CQL.系统预定义了16个CQL,The user can choose one of them to configure their need to customize the queue.
  • CQMatching rules of each queue in the
  • CQThe length of queue and send in line
  • 默认队列
• CQQueue applied to interface.The application is configuredCQL,在接口应用CQ队列.

Configuration has no corresponding classification into the default queue by message：
[RA] qos cql cql-index default-queue queue-number

配置CQ队列长度：
[RA] qos cql cql-index queue queue-number queue-length queue-length

配置CQqueue share（The size of the modified queue cache）Number of bytes sent in a row（默认1500）
[RA] qos cql cql-index queue queue-number serving byte-count

Configuration based on the classification of the interface：
[RA] qos cql cql-index inbound-interface interface-type interface-number queue queue-number

Configuration based on the classification rules of the agreement：
[RA] qos cql cql-index protocol ip [queue-key key-value ] queue queue-number

接口上应用CQL
[RA-Ethernet1/0/0] qos cql cql-index

配置参数：

• 队列长度1~1024 默认值：20;
• Send quota,取值1~16777215,默认值1500 字节;
• Both correspond to rule packets,The default queue is the queue1;

CQQueue information display：
[RA] display qos cql
[RA] display qos cql interface Ethernet1/0/0

4.2.2.4、WFQ队列

公平队列FQ（Fair Queuing）The purpose is to share network resources as fair as possible,使所有流的延迟和抖动达到最优,And fair scheduling for different queue chance,从总体上均衡各个流的延迟.与FQ相比,WFQ（Weighted Fair Queue）在计算报文调度次序时增加了优先权方面的考虑.从统计上,WFQGive high priority message priority scheduling more opportunities than low priority message.

WFQ(Weighted Fair Queuing,Weight polling queue)对报文按流特征进行分类.根据IP报文的五元组(源IP地址、目的地址、源端口号、目的端口号、协议号、ToS域中的优先级)等特征,运用HashAlgorithm to different characteristics of the flow points into the category of different queue,This process is also called a hash.

Each queue can be viewed as a kind of flow,其报文进入WFQthe same queue.WFQAs fair as possible to share network resources,使所有流的延迟和抖动达到最优,Let fair scheduling for different queue opportunity.WFQ 以 bit 为单位进行调度,Bit-by-bit 调度模型可以完全按照权重分配带宽.

1、WFQ入队机制
Classified into the queue of traffic before,有两种分类方式：

• 按流的“会话”信息分类：根据报文的协议类型、源和目的TCP或UDP端口号、源和目的IP地址、ToSThe priority of the domain an automatic flow classification such as,And provide the queue as much as possible,To each flow evenly into different queue,从而在总体上均衡各个流的延迟.
  IP网络中IP报文进入WFQ队列,是根据IPThe quintuple of the message willHashto different groups,Within each group and then according to the different distribution of different priority queue number.
  

  上图中,对于IP报文设置IP Precedencewhen the weight,由于IPpriority only0~7,队列为8个.配置WFQ队列数为16时,According to the characters of five yuan groupHashmaximum share2个组,每个组包含8个队列;设置DSCPWhen priority is weight,则WFQQueue number at least configured to64,Then all message belongs to a group of.

• 按优先级分类：通过优先级映射把流量标记为本地优先级,每个本地优先级对应一个队列号.每个接口预分配8个队列,报文根据队列号进入队列.默认情况,队列的WFQ权重相同,流量平均分配接口带宽.The user can change the weights by configuring,高优先权和低优先权按权重比例分配带宽.

2、WFQ队列调度
WFQ队列调度时,与CQQueue scheduling is similar,also round-robin scheduling,The same queue to visitFIFO方式.

Different is sending limit initial value(NQIS)Equal to queue priority plus1后与100的乘积.如果优先级是0queue sending quota(NQCS)初始值为100,优先级为4Queue to limit the initial value for500.From the queue to scheduling1开始,判断NQCS-PL>=0,从1开始直到N队列,直到当NQCS＜PL时,the current queueNQCS＋NQIS,转到下一个N＋１队列调度.Can be seen from on high priority send limit value is big,In polling keep sending message,Therefore the message sending data;The low priority send whenNQCS＜PL时,等待下一次调度,wait many times,Send message data volume relatively less priority.

从上图也能看到,Tabloid article in queue becauseNQCS-PL>=0,can send faster,And the length of the big always smooth send message,When small value can only wait for the next round of transmission line of sending.

Bit-by-bit调度模型
Bit-by-bit调度模型可以完全按照权重分配带宽,防止长报文比短报文获得更多带宽,从而减少大小报文共存时的时延抖动.

3、Distribution of export bandwidth：
在出队的时候,WFQ按流的优先级（precedence）来分配每个流应占有出口的带宽.优先级的数值越小,所得的带宽越少.优先级的数值越大,所得的带宽越多.

Fairness is better make a variety of traffic queue fairer forward,prevent a certain kind of traffic（权重大的）Keep the data link bandwidth,最多支持4096个队列,WFQThe queue is in accordance with5tuple,cannot be manually configured.

WFQ的思想：
(1)、Create a dedicated queue for each flow,Avoid queue starvation,延迟,抖动等;　
(2)、Fair among all streams,Allocate bandwidth correctly;　　
(3)、WFQ使用 [IP优先级] Allocation of bandwidth as weight.
(4)、短报文和长报文获得公平的调度：如果不同队列间同时存在多个长报文和短报文等待发送,让短报文优先获得调度,从而在总体上减少各个流的报文间的抖动.

WFQ是在FQThe tabloid on the basis of priority,When calculating the packet scheduling order to increase the priority considerations.从统计上,WFQGive a high-priority message priority scheduling more opportunities than low priority message.

FQCall into a queue of message flow,Treat each flow system is equal,Each flow will equally share the bandwidth of the currently available.

FQAlso concerned about the length of the message flow the queue,If between the queue at the same time there are multiple long and short message waiting to send message,Short message will gain scheduling priority,The first dispatch teams lined up first tabloid article,这使FQCan slow down the flow of packet jitter between.

WFQ队列特点：

• 优点：
  • 配置简单(No manual sorting);
  • To guarantee the fair scheduling to all flow,At the same time to take care of high-priority message;
  • 有利于小包的转发,To reduce user interaction class operation to the corresponding time;
  • Most platforms support;
  • 与WRED组合应用,For congestion avoidance control.
• 缺点：
  • Can't the classification of convection manual intervention;
  • A number of different flow may be drawn in the same queue(The number of flow over the configuration of the queue number),无法保证每个流获得的实际资源量;
  • Resource consumption big and not adapt to high bandwidth link;
  • WFQCan balance the flow of delay and jitter,But the same is not suitable for delay sensitive business application.

WFQQueue configuration tasks：
配置WFQweight type、队列长度、队列总数：
[huawei] qos wfq [precedence | dscp ] [queue-length max-queue-length [ queue-numbe total-queue-number] ]

WFQQueue length configuration in the range of1~1024,默认为64,The maximum number of queues is16-4096,默认为256.

权重类型：对于使用IP precedence的IP报文,当配置WFQ队列数为16时,WFQThe message according to the characters of five yuan groupHashcan be divided into2个组,Each group contains8个队列.如果使用DSCPpriority as weight,则WFQQueue for minimum to be configured to64,All message belongs to a group at this time.

4.2.2.5、PQ+WFQ(Hybrid queue scheduling)

集合了PQ调度和WFQ调度各有优缺点.单纯采用PQ调度时,The lower priority queues the message flow long bandwidth,而单纯采用WFQLow latency requirements when scheduling the business（如语音）得不到优先调度,If combine two kinds of scheduling formPQ+WFQ调度,不仅能发挥两种调度的优势,而且能克服两种调度各自的缺点.

PQ+WFQ调度,调度PQ队列,多个PQQueue scheduling according to priority high and low order.WFQ或WRRQueue weighted round robin scheduling.

between queuesPQ+WFQ调度流程

The device interface8A queue is divided into two groups,用户可以指定其中的某几组队列进行PQ调度,other queuesWFQ调度.

1、The device interface8A queue is divided into two groups,用户可以指定其中的某几组队列进行PQ调度,other queuesWFQ调度.只有WANSide interface supportPQ+WFQ调度.

2、WRR、DRR及WFQUsers according to their own need to define the weight,默认权值为10, PQno weight required,因为PQAlways use the interface bandwidth priority.Queue4、Queue3、Queue2、Queue1和Queue0队列包含自己的权值,Decide to bandwidth division ratio between each other.

3、Users according to different business needs,Message flow in each queue scheduling is not average.When the scheduling method isDRR、WRR、WFQ时,Through setting different weights to each queue,Can according to the weight of queue scheduling,The greater the weight of queue is scheduling the more the number of times the relative.

华为Queue-profile队列实现
The system can set up to defined priority is0~7的8个队列.The queue index number is divided into0、1、2、3或0、…7.Equipment according to the internal priority selection queue,and serve according to the scheduling mechanism.

Huawei device can be usedQueue-profileQueue to global definitions can be applied to the interface of software,When the hardware queue is congested,Queue-profile The queue system began to work.

queue bandwidth allocation：
If the port plastic rate for100Mbps,The interface of each business type and configuration of bandwidthPIR如下.

PQ+WFQ配置事项：

• 调度方式：仅WFQ、仅WRR、仅DRR、仅PQ或PQ+WFQ/DRR/WRRmixed scheduling.
• 定义PQ队列：如果Queue-profile中定义了多个PQ队列,则多个PQAccording to the priority between high and low order schedule.
• Queue-profile中定义了多个WFQ/WRR/DRR的队列：PQAfter queue scheduling is complete,再对DRR、WFQ或WRR队列进行调度,Share the bandwidth of the remaining.因为PQqueue first,所以若PQQueue for packet,Other queue will face“饥饿”问题.Design proposal is to limit can enterPQThe number of packets in the queue,Don't be too much bandwidth.
• 应用接口：Can be applied to the logical interface or physical interface,Logical interface configuration is better than the physical interface work.如果Queue-profileApplications to the physical interface of the logical interface,the logical interface.

output from a devicewanThe port shaping rate is100Mbps,The device input port bandwidth and output portPIRThe values ​​are in the table below：

The bandwidth allocation process is as follows：
(1)、PQ队列调度：Packet scheduling according to the priority queue,最高优先级CS7、CS6The control protocol is sent first.首先保证PQstream queuePIR带宽.PQ调度后,remaining bandwidth=100Mbps-15Mbps-10Mbps=75Mbps.

(2)、WFQ调度：根据计算公式：WFQqueue bandwidth=remaining bandwidthxThe weight of this queue/所有WFQsum of queue weights,WFQThe queue of bandwidth=75Mbps*The weight of this queue/15.

第一轮EF带宽=75Mbps*5/15=25Mbps,>EF的PIR10M,EFactual bandwidth=PIR=10Mbps,分配给EFbandwidth remaining15Mbps.

第一轮AF4带宽=75Mbps*4/15=20Mbps,>AF4的PIR15M,AF4actual bandwidth=PIR=15Mbps,分配给AF4bandwidth remaining5Mbps.

第一轮AF3带宽=75Mbps*3/15=15Mbps ,>AF3的PIR,AF3actual bandwidth=PIR=10Mbps,分配给AF3bandwidth remaining5Mbps.

第一轮AF2带宽=75Mbps*2/15=10Mbps ,=AF2的PIR10M,AF2The actual input bandwidth is only5Mbps,因此分配给AF2bandwidth remaining5Mbps.

第一轮AF1带宽=AF1带宽=75Mbps*1/15=5Mbps,<AF1的PIR10M,AF1actual bandwidth=5Mbps,分配给AF1no bandwidth left.

Port residual bandwidth=15Mbps+5Mbps+5Mbps+5Mbps=30Mbps,AF1的PIR是10M,in front only5Mbps,Then the remaining bandwidth30Mbps分配5Mbps达到AF1的PIR所设定10M,last remaining bandwidth25M分配给BE.

4.2.2.6、RTPQ队列

即RTP（Real-time Transport Protocol,实时传输协议）优先对列,Is a kind of guarantee real-time business（包括语音与视频业务）The quality of service queue fast.

emergency queue priority＞RTPQ队列优先级＞Protocol queue priority＞other queue priorities(PQ、CQ、WFQ、FIFO).

1、RTPQ队列原理
carrying voice or videoRTP报文送入高优先级队列,使其得到优先发送,Ensure the delay and jitter decreased to the minimum.

RTPQSorting the queue by：将RTPMessage is defined as a port number within a certain range,and is evenUDP报文,As classified on the basis of.

As can be seen from the above structure：
(1)、Matching real-time business into theRTPQ队列,Other business into other data queue processing
(2)、Real-time business team speed according to the bandwidth of the configuration before processing
(3)、RTPQQueue to tail after the expiration of the discarded
(4)、RTPQ将RTPMessage is defined as a port within a certain range and is evenUDP报文,To classify the basis.
(5)、RTPqueue can passFIFO、PQ、CQ和WFQ结合使用,and it has the highest priority
(6)、为了防止RTPQQueue message full bandwidth,Lead to other queue“饿死”,RTPQIn a message flow regulation processing before team,超过RTPQThe flow of the reserved bandwidth directly discarded,Flow within the reserved bandwidth to team.

注意：可以与CQ、PQ、WFQ共用,但是不能与CBQ共用,因为CBQ里面有EF队列

2、RTPQ队列调度

RTPQAfter the emergency queue queue priority,Equivalent to protocol queue,Higher than that of other data queue.队列调度时,Check whether emergency queue is empty,如果不空,Scheduling emergency queue messages sent,Otherwise, round-robin schedulingRTPQand protocol queue.RTPQ队列内部采用FIFOway of leaving.RTPQEach message queue after the team forward,Whole will dispatch to emergency queue,If the emergency queue is not empty,Send an emergency queue messages,Otherwise send agreement queue scheduling.After the completion of the queue scheduling agreement,开始下一轮的调度.如果RTPQAnd protocol queue is empty,Began to other data queue scheduling,Other data queue is empty,After completion of or a schedule,Begin the next round of queue scheduling.

3、RTPQconfiguration and display
在接口下配置RTPQ队列：
[RA-GigabitEthernet1/0/0] qos rtpq start-portfirst-rtp-port -number end-portlast-rtp-port-numberbandwidthbandwidth [cbsburst]

主要参数：
(1)、start-port：RTPthe first of the messageUDP端口号,范围：2000—65535.
(2)、end-port：RTPlast messageUDP端口号,范围：2000—65535.
(3)、bandwidth：RTPBandwidth used by the queue,范围：8—1000000,单位为Kbps.
(4)、cbs burst ：Specify Commitment Burst Size,单位为字节,范围1500—2000000字节.

显示RTPQ队列
[RA] display qos queue rtpq interface GigabitEthernet1/0/0

4.2.2.7、华为CBQ队列

CBQ(Class-Based Queue,基于类的队列)Queue is huawei's recommended queue configuration way,目前,in many enterprise subnets,CBQThe queue has become a mainstream way of queue configuration.不同于基于qos-profileBased on the traditional command line configuration mode,CBQCan realize accurate classification of queue,Implement classified according to the customer demand.CBQ软件队列,区别于Queue-Profile的是,CBQ使用MQC来配置.

1、CBQ队列原理
CBQ首先根据IP优先级或者DSCP优先级、入接口、IP报文的五元组等规则来对报文进行分类,然后让不同类别的报文进入不同的队列.对于不匹配任何类别的报文,Will be sent to the system definition the default class.

CBQ提供4类队列：

(1)、EF队列
EF（Expedit Forwarding,加速转发）队列,适用于低延时、Low drop probability、Bandwidth is not very big business,For example, important business or audio message/视频业务.User defined priority class for delay sensitive business,Each class has a priority queue,The inside of the system can fast forward the queue data,But due to the strict priority queue“too much prioritization(or run out)接口带宽”,因此,EFThe bandwidth of the queue is set limit to limit the use of bandwidth.

(2)、LLQ队列
LLQ(Low Latency Queuing,低延时队列),its delay compared toEFlower queue.The application of time delay is very sensitive(如VoIP业务)Can provide a better service and quality assurance,它较EFThe queue has a better priority transmit ability,但与EF不同的是,LLQBandwidth can never be higher than the limit set by the,Even if the other queue have idle bandwidth,LLQDon't grab with idle bandwidth.当存在EF队列和LLQ队列时,出现拥塞时,首先丢弃LLQ队列.

(3)、AF队列
AF（Assured Forwarding,确保转发）队列,i.e. ensure that the forwarding queue.Meet the need of bandwidth to ensure key business data.每个AFQueues corresponding to a user a message,Users can set each type in the bandwidth of message.At the time of system scheduling packet out of the team,按用户为各类报文设定的带宽将报文出队发送,可以实现各个类的队列的公平调度.When the interface has the remaining bandwidth,AFThe queue according to the weight to share the rest of the bandwidth.同时,在接口拥塞的时候,仍然能保证各类报文得到用户设定的最小带宽.When other queue unused bandwidth,Interface bandwidth can be assigned to the current queue to use.Can ensure that there is no more than the minimum flow of business network to ensure that the bandwidth under the condition of,This queue in the packet loss probability is very low.Ensure the forwarding is suitable for large flow and need to be guaranteed business.对于AF队列,When the queue length reaches maximum length of queue,The default tail discarded strategy,But the user can also choose to useWRED丢弃策略.

(4)、BE队列
BE(Best Efford,尽力而为)队列,The best queue.Satisfaction does not need to be strictQoS保证的尽力发送业务.When the message of all categories set by the user does not match,The default message will be sent to the system definitionBE（Best Effort,尽力传送）类.BEThe interface queue remaining bandwidth andWFQScheduled to send.

2、华为CBR队列实现
华为CBRQueue to realize the configuration parameters：
(1)、The user to define the class definition of the corresponding popular as,其内容可以是AF、EF、LLQ、BE (定义给default-class).

(2)、AFNeed to display the definition of bandwidth,The bandwidth is the minimum guarantee bandwidth,Also when congestion occurs the smallest available bandwidth.

(3)、BE使用WFQ队列机制,Don't need to display definition bandwidth,可使用BW=Minimum{10%,100-EF-LLQ-AF},BWValues for the interface bandwidth at least1%.