QoS
{Back to Index}

1. 概述
- 1.1. QoS 工作流程
- 1.2. QoS 模型
2. DiffServ

1 概述

1.1 QoS 工作流程

标记数据包
使用 ACL ，优先级，或基于源目地址，端口等手段对数据进行分类
不同的分类应用相应的策略

1.2 QoS 模型

Best-Effort

最简单的模型。应用程序可以在任何时候，发出任意数量的报文，而不需要通知网络。而网络则尽最大的可能来发送报文，但对时延、可靠性等性能不提供任何保证。
IntServ (操作复杂，灵活性较低)

应用程序在发送报文前，首先通过信令(RSVP:资源预留协议)向网络描述流量参数，申请带宽。在确定网络已经为其预留资源后，再发送报文。该模型下，可以保证报文的丢包率、延迟等要求。
DiffServ (应用广泛)

基于报文流的模型。该模型中，应用程序发出报文前，通过设置报文的 QoS 参数信息，来告知网络节点它的 QoS 需求。
网络根据每个报文流指定的 QoS 参数信息来提供差分服务，即对报文的服务等级划分，有差别地进行流量控制和转发，提供端到端的QoS保证。

2 DiffServ

Figure 1: 常见 QoS 处理流程

2.1 分类 (Classification)

如果不进行分类，所有流量将被同等对待。

2.1.1 ACL

通常使用 ACL 对流量进行分类，可以基于：

Incoming interface
IP precedence
DSCP
Source or destination address
Application

2.1.2 NBAR

基于网络的应用层识别，本质上也是对流量分类的技术，可以看作 ACL 的扩展。

NBAR 相比 ACL ，增加了如内容识别，有状态协议(FTP)的支持。

Figure 2: 实验拓扑

配置清单

----------------- R1 ----------------------
en
conf t
 hostname R1
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 192.168.1.1 255.255.255.0
  no sh
 exit
 int e0/1
  ip address 10.1.12.1 255.255.255.0
  no sh
 exit
 ip cef ! must have cef enabled
 ! define class maps
 class-map match-all URL_CM
  match protocol http url /abc ! nbar
 exit
 ! define policy map
 policy-map CBMARK
  class URL_CM
   set dscp af22
  class class-default
   set dscp default
 exit
 ! apply policy map to interface
 int e0/0
  service-policy input CBMARK
  ip nbar protocol-discovery ! enable nbar on this ifc
 exit
end

----------------- R2 ----------------------
en
conf t
 hostname R2
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 10.1.12.2 255.255.255.0
  no sh
 exit
 ip route 0.0.0.0 0.0.0.0 10.1.12.1
 ip http server
 username cisco privilege 15 password 0 cisco
 ip http authentication local
end

Figure 3: 匹配 url 的流量

2.2 标记 (Marking)

2.2.1 标记点

2.2.1.1 CoS (L2)

Class of Service ，在 802.1Q 帧中携带分类信息，在帧头的 Tag 字段中占 3bits ，范围为 0～7 。

Figure 4: CoS

2.2.1.2 ToS (L3)

Type of Service 服务类型，IP 包头携带一个字节的字段，标记 IP 包的服务类型， ToS 字段内可以是 IP Precedence ，也可以是 DSCP 值。

IP Precedence

共占 3bits ，范围为 0～7
DSCP

Differentiated Services Code Point ，共占 6bits ，范围为 0～63 ， 向下兼容 IP Precedence

Figure 5: ToS

2.2.2 标记方式

2.2.2.1 PBR ¹

即使用 route-map 工具来实施策略。(注: PBR 是优先于路由表的)

2.2.2.1.1 示例1

Figure 6: 根据入接口打标记

配置清单

----------------- R1 ----------------------
en
conf t
 hostname R1
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 10.1.13.1 255.255.255.0
  no sh
 exit
 int e0/1
  ip address 10.1.14.1 255.255.255.0
  no sh
 exit
 int e0/2
  ip address 10.1.12.1 255.255.255.0
  no sh
 exit
 
 route-map PRECEDENCE2 permit 10
  set ip precedence 2
 exit
 route-map PRECEDENCE0 permit 10
  set ip precedence 0
 exit
 
 int e0/0
  ip policy route-map PRECEDENCE2
 exit
 int e0/1
  ip policy route-map PRECEDENCE0
 exit
end

----------------- R2 ----------------------
en
conf t
 hostname R2
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 10.1.12.2 255.255.255.0
  no sh
 exit
 ip route 0.0.0.0 0.0.0.0 10.1.12.1
end

----------------- PC3 ----------------------
en
conf t
 hostname PC3
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 no ip routing
 ip default-gateway 10.1.13.1
 int e0/0
  ip address 10.1.13.3 255.255.255.0
  no sh
 exit
end

----------------- PC4 ----------------------
en
conf t
 hostname PC4
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 no ip routing
 ip default-gateway 10.1.14.1
 int e0/0
  ip address 10.1.14.4 255.255.255.0
  no sh
 exit
end

Figure 7: PC3 发起的流量在 R2 上抓包

Figure 8: PC4 发起的流量在 R2 上抓包

2.2.2.1.2 示例2

Figure 9: 实验环境

实验清单

----------------- R1 ----------------------
en
conf t
 hostname R1
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 192.168.1.1 255.255.255.0
  no sh
 exit
 int e0/1
  ip address 10.1.12.1 255.255.255.0
  no sh
 exit
 
 ip access-list extended WWW_TRAFFIC
  permit tcp any 10.1.12.2 0.0.0.255 eq www
 exit
 ip access-list extended TELNET_TRAFFIC
  permit tcp any 10.1.12.2 0.0.0.255 eq telnet
 exit
 
 
 route-map PBR_ROUTE_MAP permit 10
  match ip address WWW_TRAFFIC
  set ip precedence 1
 exit
 route-map PBR_ROUTE_MAP permit 20
  match ip address TELNET_TRAFFIC
  set ip precedence 2
 exit
 route-map PBR_ROUTE_MAP permit 30
  set ip precedence 0
 exit
 
 
 int e0/0
  ip policy route-map PBR_ROUTE_MAP
 exit
 
end

----------------- R2 ----------------------
en
conf t
 hostname R2
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 10.1.12.2 255.255.255.0
  no sh
 exit
 ip route 0.0.0.0 0.0.0.0 10.1.12.1
 username cisco password 0 cisco
 line vty 0 4
  login local
  transport input telnet
 exit
end

Figure 10: www 流量

Figure 11: telnet 流量

2.2.2.2 Class-Based Mark

CB Marking 可以使用 MQC² 方式来配置，可以支持 in/out 两个方向。

Figure 12: Class Map 即分类，Policy Map 即 Action (3 调用 2 , 2 调用 1)

2.2.2.2.1 示例

Figure 13: 实验环境

配置清单

----------------- R1 ----------------------
en
conf t
 hostname R1
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 192.168.1.1 255.255.255.0
  no sh
 exit
 int e0/1
  ip address 10.1.12.1 255.255.255.0
  no sh
 exit
 ip cef ! must have cef enabled
 ! define ACLs
 ip access-list extended WWW_TRAFFIC
  permit tcp any any eq www
 exit
 ip access-list extended TELNET_TRAFFIC
  permit tcp any any eq telnet
 exit
 ! define class maps
 class-map match-all TELNET_CM
  match access-group name TELNET_TRAFFIC
 exit
 class-map match-all WWW_CM
  match access-group name WWW_TRAFFIC
 exit
 ! define policy map
 policy-map CBMARK
  class TELNET_CM
   set dscp af11
  class WWW_CM
   set dscp af41
  class class-default
   set dscp default
 exit
 ! apply policy map to interface
 int e0/0
  service-policy input CBMARK
 exit
end

----------------- R2 ----------------------
en
conf t
 hostname R2
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 10.1.12.2 255.255.255.0
  no sh
 exit
 ip route 0.0.0.0 0.0.0.0 10.1.12.1
 username cisco password 0 cisco
 line vty 0 4
  login local
  transport input telnet
 exit
end

Figure 14: telnet 流量

Figure 15: www 流量

查看统计信息

R1#sh policy-map interface
 Ethernet0/0

  Service-policy input: CBMARK

    Class-map: TELNET_CM (match-all)
      67 packets, 3780 bytes
      5 minute offered rate 0000 bps, drop rate 0000 bps
      Match: access-group name TELNET_TRAFFIC
      QoS Set
        dscp af11
          Packets marked 67

    Class-map: WWW_CM (match-all)
      1 packets, 74 bytes
      5 minute offered rate 0000 bps, drop rate 0000 bps
      Match: access-group name WWW_TRAFFIC
      QoS Set
        dscp af41
          Packets marked 1

    Class-map: class-default (match-any)
      8 packets, 2415 bytes
      5 minute offered rate 0000 bps, drop rate 0000 bps
      Match: any
      QoS Set
        dscp default
          Packets marked 8

2.3 整形(Shaping)与监管(Policing) ³

Figure 16: Shaping vs Policing

Figure 17: Shaping vs Policing (Cont.)

整形机制
- Generic Traffic Shaping (GTS)
- Frame Relay Traffic Shaping (FRTS)
- Class-Based Shaping
监管机制
- Committed Access Rate (CAR)
- Class-Based Policing

2.3.1 令牌桶原理 ⁴

Figure 18: 双令牌桶， CIR 即限制后的速率

2.3.2 Class-Based Shaping

Figure 19: CB Shaping Building Blocks

Figure 20: 实验环境

配置清单

----------------- R1 ----------------------
en
conf t
 hostname R1
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 192.168.12.1 255.255.255.0
  no sh
 exit
 int e0/1
  ip address 192.168.13.1 255.255.255.0
  no sh
 exit
 int e0/2
  ip address 192.168.14.1 255.255.255.0
  no sh
 exit
 ip cef
 ip access-list extended ACL_SUBNET12
  permit ip 192.168.12.0 0.0.0.255 any
 exit
 ip access-list extended ACL_SUBNET14
  permit ip 192.168.14.0 0.0.0.255 any
 exit
 ! define class maps
 class-map match-all CMAP12
  match access-group name ACL_SUBNET12
 exit
 class-map match-all CMAP14
  match access-group name ACL_SUBNET14
 exit
 ! define policy map
 policy-map CBSHAPE
  class CMAP12
   shape average 64000  ! set CIR to 64k
  class CMAP14
   shape average 128000 ! set CIR to 128k
 exit
 ! apply policy map to interface
 int e0/1
  service-policy output CBSHAPE
 exit
end

查看接口 policy-map 状态

R1#sh policy-map interface e0/1
 Ethernet0/1

  Service-policy output: CBSHAPE

    Class-map: CMAP12 (match-all)
      13027 packets, 19299125 bytes
      5 minute offered rate 114000 bps, drop rate 112000 bps
      Match: access-group name ACL_SUBNET12
      Queueing
      queue limit 64 packets
      (queue depth/total drops/no-buffer drops) 0/12594/0
      (pkts output/bytes output) 433/538334
      shape (average) cir 64000, bc 256, be 256
      target shape rate 64000 ! 限速 64K

    Class-map: CMAP14 (match-all)
      8681 packets, 12865787 bytes
      5 minute offered rate 133000 bps, drop rate 127000 bps
      Match: access-group name ACL_SUBNET14
      Queueing
      queue limit 64 packets
      (queue depth/total drops/no-buffer drops) 0/8290/0
      (pkts output/bytes output) 391/516533
      shape (average) cir 128000, bc 512, be 512
      target shape rate 128000 ! 限速 128K

    Class-map: class-default (match-any)
      82 packets, 10661 bytes
      5 minute offered rate 0000 bps, drop rate 0000 bps
      Match: any

      queue limit 64 packets
      (queue depth/total drops/no-buffer drops) 0/0/0
      (pkts output/bytes output) 82/10661

Linux2 测速

root@Linux2:/home/cisco# iperf3 -u -c 192.168.13.3 -b 5M -t 10 -i 2
Connecting to host 192.168.13.3, port 5201
[  5] local 192.168.12.2 port 57057 connected to 192.168.13.3 port 5201
[ ID] Interval           Transfer     Bitrate         Total Datagrams
[  5]   0.00-2.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   2.00-4.00   sec  1.19 MBytes  5.00 Mbits/sec  864
[  5]   4.00-6.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   6.00-8.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   8.00-10.00  sec  1.19 MBytes  5.00 Mbits/sec  863
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Jitter    Lost/Total Datagrams
[  5]   0.00-10.00  sec  5.96 MBytes  5.00 Mbits/sec  0.000 ms  0/4316 (0%)  sender
[  5]   0.00-22.12  sec   168 KBytes  62.3 Kbits/sec  9.977 ms  4184/4303 (97%)  receiver
iperf Done.

Linux4 测速

root@Linux4:/home/cisco# iperf3 -u -c 192.168.13.3 -b 5M -t 10 -i 2
Connecting to host 192.168.13.3, port 5201
[  5] local 192.168.14.4 port 49732 connected to 192.168.13.3 port 5201
[ ID] Interval           Transfer     Bitrate         Total Datagrams
[  5]   0.00-2.00   sec  1.19 MBytes  4.99 Mbits/sec  863
[  5]   2.00-4.00   sec  1.19 MBytes  5.01 Mbits/sec  864
[  5]   4.00-6.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   6.00-8.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   8.00-10.00  sec  1.19 MBytes  5.00 Mbits/sec  863
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Jitter    Lost/Total Datagrams
[  5]   0.00-10.00  sec  5.96 MBytes  5.00 Mbits/sec  0.000 ms  0/4316 (0%)  sender
[  5]   0.00-16.00  sec   243 KBytes   125 Kbits/sec  2.443 ms  4110/4282 (96%)  receiver

iperf Done.

2.3.3 Class-Based Policing

Class-based policing 定义了 3 中状态来表示当前速度与限速之间的关系：

Conforms

第一个令牌桶还有令牌
Exceeds

第一个令牌桶没有令牌了，但 burst 桶中还有令牌
Violates

两个桶中都没有可用的令牌了

每种状态下可以采取的 Action 有：

Drop (Exceeds 和 Violates 的默认行为)
Transmit (Conforms 的默认动作)
Set IP Precedence and transmit
Set IP DSCP and transmit
Set MPLS experimental bits and transmit
Set Frame Relay DE bit and transmit

Figure 21: 配置命令

Figure 22: 实验环境

配置清单

----------------- R1 ----------------------
en
conf t
 hostname R1
 no ip do lo
 line con 0
  exec-timeout 0 0
  logging synchronous
 exit
 int e0/0
  ip address 192.168.12.1 255.255.255.0
  no sh
 exit
 int e0/1
  ip address 192.168.13.1 255.255.255.0
  no sh
 exit
 int e0/2
  ip address 192.168.14.1 255.255.255.0
  no sh
 exit
 ip cef
 ! define class maps
 ip access-list extended ACL_SUBNET12
  permit ip 192.168.12.0 0.0.0.255 any
 exit
 ip access-list extended ACL_SUBNET14
  permit ip 192.168.14.0 0.0.0.255 any
 exit
 class-map match-all CMAP12
  match access-group name ACL_SUBNET12
 exit
 class-map match-all CMAP14
  match access-group name ACL_SUBNET14
 exit
 ! define policy map
 policy-map CBPOLICE
  class CMAP12
   police 128000 conform-action transmit exceed-action set-dscp-transmit af32 violate-action drop ! rate unit is bit
  class CMAP14
   police 256000 conform-action transmit exceed-action set-dscp-transmit af33 violate-action drop ! 256K
  class class-default
   police 64000
 exit
 ! apply policy map to interface
 int e0/1
  service-policy output CBPOLICE
 exit
end

测速

root@Linux4:/home/cisco# iperf3 -u -c 192.168.13.3 -b 5M -t 10 -i 2
Connecting to host 192.168.13.3, port 5201
[  5] local 192.168.14.4 port 52015 connected to 192.168.13.3 port 5201
[ ID] Interval           Transfer     Bitrate         Total Datagrams
[  5]   0.00-2.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   2.00-4.00   sec  1.19 MBytes  5.00 Mbits/sec  864
[  5]   4.00-6.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   6.00-8.00   sec  1.19 MBytes  5.00 Mbits/sec  863
[  5]   8.00-10.00  sec  1.19 MBytes  5.00 Mbits/sec  863
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate         Jitter    Lost/Total Datagrams
[  5]   0.00-10.00  sec  5.96 MBytes  5.00 Mbits/sec  0.000 ms  0/4316 (0%)  sender
[  5]   0.00-10.00  sec   317 KBytes   259 Kbits/sec  0.075 ms  4079/4303 (95%)  receiver

iperf Done

2.4 拥塞避免 ⁵

2.4.1 Class-Based WRED

2.5 队列 ⁶

Figure 25: 队列架构

虽然有不同的 Queue 技术，但功能组成通常分为三部分：

Classification (将数据包分类)
Insertion Policy (判断数据包是否需要入队或丢弃)
Service Policy (如何调度数据包到硬件队列)

2.5.1 FIFO

Classification	Insertion Policy	Service Policy
无需分类	tail-drop	FIFO

2.5.2 PQ (Priority)

Classification	Insertion Policy	Service Policy
4 种分类	tail-drop	按照分类的优先级高低进行抢占式调度

Figure 27: PQ 概貌

Figure 28: PQ 调度逻辑

配置示例

Figure 29: 配置命令

Figure 30: show interface

Figure 31: Queue 信息

2.5.3 CQ (Round Robin)

Classification	Insertion Policy	Service Policy
16 + 1 种分类	tail-drop	轮询与优先相结合的调度

Figure 32: 每个 queue 的大小 (byte-count) 为 1500

配置示例

Figure 33: 配置

Figure 34: show interface

2.5.4 Weighted Fair Queuing (WFQ)

Classification	Insertion Policy	Service Policy
基于 Flow⁷	使用 CDT/HQO 两个阈值	基于 Finish Time⁸

Figure 35: 所有 queue 能缓存的数据包总量是固定的

2.5.4.1 Congestive Discard Threshold (CDT)

Figure 36: 假设 CDT=8 ，如 incoming packet 属于最长队列，则丢弃( Early dropping )

Figure 37: 假设 CDT=8 ，如 incoming packet 不属于最长队列，则保留

2.5.4.2 Hold-Queue Out limit (HQO)

Figure 38: 假设 HQO=10 ，如 incoming packet 属于最长队列，则丢弃

Figure 39: 假设 HQO=10 ，如 incoming packet 不属于最长队列，则丢弃最长队列最后一个数据包，并将当前数据入队

配置示例

Router(config-if)# fair-queue 64    256      0
                              CDT   Dyn Qs   RSVP Qs
Router(config-if)# hold-queue 1000 out
                              HQO

2.5.5 Class-Based WFQ

Footnotes:

路由选路通常是以目标地址为导向的， PBR 的思想是以源地址为导向。参考策略路由

所有 Class-Based 配置方式都属于 MQC ，MQC 本质还是 CLI ，提供模块化的配置

整形与监管本质是为了限速

⁴

https://youtu.be/r3p0eZWjWS4?list=PLwLfEC5vHvKex1Z90_qz3fYXHGVmrnVFA&t=505

⁵

主要是处理如何 Early Dropping

⁶

通常(默认)作用于接口出方向

⁷

Flow 是基于六元组 (SrcAddr,SrcPort,DestAddr,DestPort,Protocol,ToS)，相比 Class 是人为定义的，Flow 是动态生成的，WFQ 即为 FBWFQ (Flow Based)

⁸

https://youtu.be/S21jv5Xn7pY?list=PLwLfEC5vHvKex1Z90_qz3fYXHGVmrnVFA

QoS {Back to Index} Expand All Details on Page

Table of Contents

1 概述

1.1 QoS 工作流程

1.2 QoS 模型

2 DiffServ

2.1 分类 (Classification)

2.1.1 ACL

2.1.2 NBAR

2.2 标记 (Marking)

2.2.1 标记点

2.2.1.1 CoS (L2)

2.2.1.2 ToS (L3)

2.2.2 标记方式

2.2.2.1 PBR 1

2.2.2.1.1 示例1

2.2.2.1.2 示例2

2.2.2.2 Class-Based Mark

2.2.2.2.1 示例

2.3 整形(Shaping)与监管(Policing) 3

2.3.1 令牌桶原理 4

2.3.2 Class-Based Shaping

2.3.3 Class-Based Policing

2.4 拥塞避免 5

2.4.1 Class-Based WRED

2.5 队列 6

2.5.1 FIFO

2.5.2 PQ (Priority)

2.5.3 CQ (Round Robin)

2.5.4 Weighted Fair Queuing (WFQ)

2.5.4.1 Congestive Discard Threshold (CDT)

2.5.4.2 Hold-Queue Out limit (HQO)

2.5.5 Class-Based WFQ

Footnotes:

QoS
{Back to Index}

2.2.2.1 PBR ¹

2.3 整形(Shaping)与监管(Policing) ³

2.3.1 令牌桶原理 ⁴

2.4 拥塞避免 ⁵

2.5 队列 ⁶