By Martha Cooper,2014-10-14 19:44
13 views 0



    重点在各个队列利弊、整形和策略、三种QOS model。以及每种队列的典型配置命令都是常考



    Lack of Bandwidth :maximum available bandwidth equals bandwidth of the slowest link.

     solution: update the link,

     forward the important packet first

     compress the payload of layer 2 frames

     compress ip packet headers

     available bandwidth: LLQ and RTP queuing ==============voice

     CBWFQ , TCP header compress ==============interactive traffic

    end-to-end delay : processing delay , queue delay , serialization delay , propagation delay

     Processing delay: The time that it takes for a router (or Layer 3 switch) to take the packet

    from an input interface and put it into the output queue of the output interface. The

    processing delay depends on various factors.

     Queuing delay: The time that a packet resides in the output queue of a router. Queuing

    delay depends on the number of packets already in the queue and their sizes. Queuing delay also depends on the bandwidth of the interface and the queuing mechanism.

     Serialization delay: The time that it takes to place a frame on the physical medium for

    transport. This delay is typically inversely proportional to the link bandwidth.

     Propagation delay: The time that it takes for the packet to cross the link from one end to the other. This time usually depends on the type of media

     solution : update link

     forward the important packet first

     enable re-prioritization of important packet

     compress the payload of layer2 frame

     compress ip packet header

     use of way to reduce delay

     Customer routers perform:

     TCP/RTP header compression



     ISP routers perform:

     Reprioritization according to the QoS policy Jitter : best effect delivery

Loss packet:

     update link

     guarantee enough bandwidth to sensitive packets =====LLQ

     prevent congestion by random dropping less important packet before congestion


    2implementing QOS

     identify traffic and its requirements

     divide traffic into classes

     define QOS policies for each class

     Qos function

     traffic control

     traffic shaping avoid policing

     traffic classification and marking

    3qos model

     best effects:

     advantage : high scalable

     no special mechanisms

     disadvantage: no service guarantees

     no service differentiation

     IntServ: =========resource reserve and admission control

     advantage: explicit resource admission control

     pre-request policy admission control

     signaling of dynamic port number

     disadvantage: continuous signaling because of stateful architecture

     flow-base approach not scalable to large



     Admission control





     advantage high scalable , many levels of quality possible

     disadvantage : complex mechanism , no absolute service guarantee

     marking :tos de exp \ ip precedence dscp

     classification : best effect(0) middle (1) high (2) signaling(3) video (4)voice (5) reserve

    (6-7) TOS


     tcp\ udp port 3455

     protocol id 46

     request qos parameters from all device that are between source and destination

     rate-sensitive and delay sensitive traffic

     used routing line , do not use own line


     enough available resource

     application in question is allowed access to these resources

    5 Match all: All conditions have to be met to bind a packet to the class.

    ? Match any: At least one condition has to be met to bind the packet to the class.

     Up to 256 traffic classes can be associated with a single traffic policy. 6auto QOS

     ip cef , nbar , correct bandwidth statement is configure on the interface ,use to interface 7classification

     is the most fundamental qos building block , without ,all packet are treated the same

     based on : incoming interface , ip precedence , dscp , application , source or destination address


     is the qos feature component that " colors" packet so is can be identified and distinguish from other packets on qos treatment.

     based on :

     link layer: ethernet frame cos field mpls exp bit, frame-relay de bit

     network layer : dscp ( cos field left 6 bits), ip precedence (cos field left 3 bits )


     With DiffServ, you can define service classes on a router and then sort packets into

    these classes. The router can then queue and forward packets with different levels of priority,

    according to their classification. Each queuing and forwarding treatment is called a Per-Hop Behavior (PHB).

     include : classes-selector , AF ,EF(101 110 The EF PHB guarantees and polices bandwidth while ensuring a minimum departure rate.) , default PHB

     ef: it not permit exceed guarantee bandwidth

     af can exceed guarantee bandwidth


     protocol discovery ip nbar protocol-discovery

     performs identification of application and protocol

     per-protocol and per-interface statistics

     traffic classification

     when loading PDLM supported new application 11queuing


    is designed to accommodate temporary congestion on an interface of a network device by

    storing excess packets in buffer

     until bandwidth become available


    Speed mismatch and aggregate bring congestion

    The tx-ring state is an indication of hardware interface congestion. queue components:

    classification \ insert policy \ service policy Queuing algorithm (four)


    First in first out

    Simplest of all

    One queue


    Users multiple queues

    Allows prioritization

    Always empties first queues before going to the next queue Lower priority queue will be "starve"


    Router(config)#priority-list 1 protocol ip high tcp www Router(config)#priority-list 1 protocol ip medium tcp telnet Router(config)#priority-list 1 default low !

    Router(config)#interface serial 0/1

    Router(config-if)#priority-group 1

Round robin

    User multiple queues

    No prioritization

    Dispatches one packet from each queue in each round WRR

    Allows prioritization

    Assign a weight to each queue

Dispatches packets from each queue pre-prioritization to an assigned weight

    The router is allowed to send the entire packet even if the sum of all bytes is more than the

    threshold. ===latency


    Switch(config)#interface gig 0/5

    Switch(config-if)#wrr-queue bandwidth 1 2 3 4

    Switch(config-if)#wrr-queue cos-map 4 5


    Per-flow FIFO

    WFQ automatically adapts to changing network traffic conditions Classification

    ? Source IP address

    ? Destination IP address

    ? Protocol number (identifying TCP or UDP)

    ? Type of service field

    ? Source TCP or UDP port number

    ? Destination TCP or UDP port number

    two mode drop

    earlier dropping when the congestive discard threshold is reached aggressive dropping when the hold-queue out limit is reached exception

    a packet classified into an empty queue is never dropped the packet ip precedence has no effect on the dropping scheme benefits

    simple configuration

    guarantee throughput to all flows

    drop packet of most aggressive flows

    supported in most platform

    supported in most cisco ios


    possible of multiple flows ending up in one queue only used E1 or below E1 line

    lack of control over classification

    can not provide fixed bandwidth guarantees



    hold-queue max-limit out ===the maximum number of packets that can be in all output

    queues on the interface at

     any time. show queue interface-name interface-number


    guarantee bandwidth to classless and accord weight assigned to traffic classes


    custom-defined classification

    minimum bandwidth allocation



    but voice traffic can still suffer unacceptable delay 默认情况下,分配给所有类别的带宽总和不能超过接口可用带宽的75%


    bandwidth \ bandwidth percent \ bandwidth remaining percent \ max-reserved-bandwidth

    queue-limit (default max 64 ) \ fair-queue (16-4096) show policy-map interface [interface]


    a priority queue is added to CBWFQ for real-time traffic high-priority classes are guarantee

    when congest appear ,high flows can not exceed their guarantees bandwidth

    lower priority classes use CBWFQ


    priority bandwidth \ priority percent N

    Router(config-pmap)#class-map VOICE

    Router(config-pmap-c)#priority 256

    show policy-map interface interface

12congest avoidance tx-ring indicate hardware interface congestion

    drop mechanism : car , class-based shaping , WRED , WFQ

    tail drop


    tcp synchronization

    tcp starve

    no differentiated drop


    is a mechanism that randomly drops packets before a queue is full increase drop rate as the average queue size increase


     (no drop)

    minimum threshold

     (random drop)

    maximum threshold

     (full drop)

    mark probability denominator


    the profile same to RED, it base on ip precedence and dscp WRED drop less important packets more aggressive than more important packets WRED prevent tail drop allow the implementation of DiffServ 's AF PHB


    CBWRED is available when configure in combination with CBWFQ.

    CBWRED allows implementation of DiffServ assured forwarding PHB stand-alone WRED

     base-on traffic class \ip precedence \dscp value



    random-detect [dscp-based | prec-based] // If neither dscp-based nor prec-based is specified, WRED defaults to prec-based.

    random-detect precedence

    random-detect dscp

flow-based WRED

    drop packets of aggressive flows a head of other packets no distributed ,no predictable , works only adaptive flows, depends on hop behavior

    ensure that flows respond to WRED packet drop prohibit a single flow from the buffer resource at an interface

    13policing and shaping


    incoming and out going direction

    out-of-profile packet are drop

    dropping causes tcp retransmit

    support packet marking or remarking

    discard the packet when token bucker no enough cache not be increase


    outgoing direction

    out-of-profile packet are queue until buffer gets full buffering minimum tcp retransmit

    support interaction whit FR congest notification indirection marking and re-marking not support

    wait for enough token to accumulate in the bucket why use policing

    to limit access to resource when high-speed access is need but not desired to limit the traffic rate of certain application or traffic classes to marking exceed traffic at layer 2 and layer 3 why use shaping

    to prevent and manage congestion

    to regulate (controlthe sending traffic rate to match the subscribed rate to implement shaping at the network edge

    14bc is normal burst size

     tc is the time interval ====default interval 125 ms

     cir is the committed information rate


15link efficiency mechanisms

    layer 2 payload compress ==========sticker\ predictor\ mppc header compress =================tcp \ real-time-transport protocol \ class-based tcp \

    class-based rtp

    compress ===obtain high bandwidth and lower latency link fragment interleave (LFI) ========reduces the delay and jitter of small packets


    0 best effort 1 middle priority date 2 high priority date 3 call signaling 4 video 5 voice 6and7 reserve 17pre-classification be used to VPN


    Permit implement QOS rate limit in control plan , protect control plan 's dos attack


     is used to mark packets to select a per-hop behavior

    how does dscp usage through the network ? PHB

    20dscp can be used mark packet

    car can be used mark and classification


    implement a simple based on destination forward instead of a fixed router search based on qos destination classification and marking packet router-map under packets received through the specified interface command

    route-map xxx

     match ip add 100

     set ip next-hope xxx

     set ip precedence xxx

     and so on


    flow-based DWFQ

    number of queue is 512


    do not support configuration of classification

    do not use ip precedence as weight


    automatic classification

    high performance

Tos-based DWFQ


    automatic classification

    guarantees throughput to all classes high performance


    only four classes

    do not configuration of classification fifo queuing within a single class

Qos-group-based DWFQ

    output queuing

    support 100 classes

    based Qos group parameter

    guarantees throughout to all classes a large number of classes

    high performance

    need set qos group


    CBWFQ : guarantees bandwidth

    CBLLQ : guarantees bandwidth and low-latency CB shaping: delay

    CB policing: limit traffic rate by dropping excess traffic

    CB marking: mark packets


    compression packet and segment header tcp header-compress : ip and tcp header , ppp link

    rtp header-compress : ip udp rtp header ppp link, it be used to slow link and delay sensitive


    compress depend on router platform compression algorithm hardware compression


    algorithm : stacker and predictor 25subnet bandwidth management

Report this document

For any questions or suggestions please email