QoS 配置基础

　　Configuration of QoS

　　In last month's newsletter I discussed the mechanisms QoS provides for

　　delivering traffic with different priorities and assurances. This month

　　we are going to look at the specific configurations for enabling the

　　QoS assurances. Configuration of QoS is somewhat confusing as there are

　　many types of QoS "knobs" that can be configured on a router or a

　　switch. Hopefully this newsletter will provide the necessary

　　instruction for getting through some of the complexities of QoS

　　configurations. This newsletter focuses on the QoS configurations

　　relevant to Cisco IOSnot CatOS. Layer2 QoS is another topic entirely

　　and will be reserved for future newsletters.

　　To refresh our memory from last month, let's review QoS briefly. QoS is

　　a set of tools, or "knobs" as they are sometimes called, that are

　　configured on network elements (usually routers and switches) to

　　enforce traffic delivery policies for certain types of traffic. QoS can

　　be broken down into three levels or models to provide end-to-end levels

　　of service to traffic delivered over the network infrastructure. These

　　models are called best-effort service, integrated service and

　　differentiated service. Within these models, QoS defines different

　　techniques to handle traffic. These techniques are congestion

　　avoidance, congestion management, policing and traffic shaping. We will

　　focus on the configuration of the congestion management techniques in

　　this newsletter. Congestion avoidance, policing and traffic shaping

　　techniques will be discussed next month.

　　Congestion Management

　　Congestion management techniques control the prioritization of traffic

　　as it leaves the queues, it does not minimize congestion. Cisco

　　provides several different techniques for providing congestion

　　management. The mechanisms include First In First Out Queuing (FIFO),

　　Weighted Fair Queuing (WFQ), Priority Queuing and Custom Queuing. In

　　addition, there is Class Based Weighted Fair Queuing (CBWFQ), IP RTP

　　Priority (or Priority Queue WFQ) and Low Latency Queuing (LLC).

　　First in First Out Queuing (FIFO)

　　FIFO has no concept of priority or classes of traffic and does not

　　provide any prioritization of packets

　　Weighted Fair Queueing (WFQ)

　　WFQ provides allocation of bandwidth to network conversations. Weights

　　are assigned based on source and destination patterns. WFQ

　　automatically gives precedence for low-volume traffic such as telnet

　　and HTTP over high volume traffic such as FTP. WFQ is enabled by

　　default on interfaces that run below E1 speeds (2.048 Mbps).

　　Priority Queuing (PQ)

　　PQ allows you to define four queues for traffic prioritization. Traffic

　　is matched using match criteria such as IP address, port numbers and

　　interfaces. The four priorities are defined as high, medium, normal and

　　low. The queues are served in a preferential manner. The router will

　　not service the lower-priority queues until all packets from the

　　higher-priority queues have been delivered. This can be detrimental if

　　not configured properly as all of the bandwidth can be consumed by the

　　higher-priority queues, not allowing the router to transmit data out of

　　the other queues. Let's say you want to do the following for traffic

　　being forwarded out interface serial 0.

　　Telnet (TCP port 23) traffic is assigned to the high-priority queue.

　　AppleTalk and IPX are given the medium-priority queue.

　　All other IP traffic is assigned to the normal-priority queue.

　　All other traffic is assigned to the low-priority queue (as default)

　　Below is the configuration:

　　priority-list 2 protocol ip high tcp 23

　　priority-list 2 protocol ip high list 1

　　priority-list 2 protocol interface ethernet 0 medium

　　priority-list 2 protocol ip normal

　　priority-list 2 queue-limit 15 20 20 30

　　!

　　access-list 1 permit 131.108.0.0 0.0.255.255

　　!

　　interface serial 0

　　priority-group 2

　　You can change the default number of packets in each queue by using the

　　following command:

　　Priority-list list-number queue-limit high-limit medium-limit normal-

　　limit low-limit

　　This is not recommended!!!!!

　　Custom Queuing (CQ)

　　CQ provides the same type of classification of packets as PQ with 16

　　queues instead of four. In addition, CQ allows the number of bytes to

　　be specified for forwarding each time the queue is serviced. The queues

　　are serviced in a round-robin fashion. By servicing each queue in a

　　round-robin fashion, CQ ensures that no application receives more than

　　a predetermined amount of overall bandwidth. Let's say you want to do

　　the following for traffic being forwarded out interface serial 0.

　　Traffic from E0 is assigned to queue 1.

　　IP traffic is sent to queue 2.

　　IPX traffic goes to queue 3.

　　AppleTalk traffic goes to queue 4

　　Queue 5 is the default queue. Traffic that is not from E0 or is not

　　IP-, IPX- or AppleTalk-based is sent to this queue.

　　The configuration would be as follows:

　　queue-list 1 interface E0 1

　　queue-list 1 protocol ip 2

　　queue-list 1 protocol ipx 3

　　queue-list 1 protocol appletalk 4

　　queue-list 1 default 5

　　!

　　Interface serial 0

　　custom-queue-list 1

　　CQ allows for granularity on the number of packets in any queue and the

　　number of bytes delivered from a queue. This is done by using the

　　following commands:

　　queue-list 1 queue 10 limit 40 (limit number of packets)

　　queue-list 1 queue 10 byte-count 1400 (set byte count)

　　In addition, access lists can be used to identify what applications are

　　serviced by each queue. Instead of just configuring IP in queue 2 you

　　could add a protocol number or access list to the queue list command.

　　For example:

　　queue-list 1 protocol ip 2 TCP 23 (TCP) OR

　　queue-list 1 protocol ip 2 list 10 (where 10 is an access list defining

　　certain traffic types)

　　More Congestion Management Techniques

　　The last three types of congestion management techniques are presented

　　separately as they utilize enhancements and combinations of the

　　previous queuing techniques. The following queuing techniques are used

　　primarily in IP telephony environments as they have the ability to give

　　more granular prioritization to voice traffic.

　　Class Based Weighted Fair Queuing (CBWFQ)

　　CBWFQ extends WFQ to provide user-defined classes using match criteria.

　　Queues are reserved for each class, and characteristics are assigned to

　　each queue. The characteristics include bandwidth, weight and queue

　　limit. The way CBWFQ works is as follows: Match criteria filters and

　　classify packets first and then each packet is assigned a weight. The

　　weight is derived from the bandwidth you assigned to the class.

　　Finally, each packet is placed in the appropriate queue and serviced

　　according to the weights. In order to configure this on a Cisco router,

　　you must perform the following three processes:

　　* Class Map Process: Define the traffic classes (i.e., what packets

　　are going to be matched and by what criteria).

　　* Class Policy Process: Specify the policies for each class of

　　traffic.

　　* Apply Policy Process: Apply the policies to the interfaces on the

　　router.

　　The commands necessary for each of these processes is shown below:

　　Class Map Process

　　Match a packet type

　　access-list 101 permit udp host 10.10.10.10 host 10.10.10.20 range

　　16384 20000

　　access-list 101 permit udp host 10.10.10.10 host 10.10.10.20 range

　　53000 56000

　　class-map class1

　　match access-group 101

　　class-map class2

　　match access-group 102

　　Instead of matching an access-list you can use other match criteria

　　such as:

　　match input-interface interface-name

　　or

　　match protocol protocol

　　or

　　match mpls-experimental number number

　　Class Policy Process

　　Assign packet class characteristics; if traffic does not meet any of

　　the defined match criteria, it is assigned to a default class. The four

　　characteristics that can be assigned are as follows:

　　* Class

　　* Bandwidth

　　* Fair-queue (for class-default class only)

　　* Queue-limit or random-detect

　　One or more of the characteristics can be assigned to the matched

　　packets for each class. You can assign up to 64 classes. Total

　　allocated bandwidth for the classes should not exceed 75 percent of the

　　available interface bandwidth.

　　policy-map policy1

　　class class1

　　bandwidth 3000

　　queue-limit 30

　　class class2

　　bandwidth 2000

　　If you use WRED on the policy map, WRED cannot be configured on the

　　interface to which you apply the policy. The examples above use tail

　　drop for congestion avoidance; if you want to use WRED, it has to be

　　explicitly configured as follows:

　　class-map class1

　　match access-group 101

　　policy-map policy1

　　class class1

　　bandwidth 1000

　　random-detect (enables WRED)

　　interface serial0/0

　　service-policy output policy1

　　The default class is used to classify traffic that does not meet the

　　match criteria. The default class uses WFQ by default, however if the

　　bandwidth policy-map class is used, WFQ is overwritten. The commands

　　used to configure the default class are the same as the other classes

　　with the exception of one and the rules regarding tail drop and WRED

　　still apply. The command needed for the default class is as follows:

　　class class-default default-class-name

　　The difference is the class-default parameter before the class name.

　　Assign Policy Process

　　interface e1/1

　　service output policy1

　　interface fa1/0/0

　　service output policy1

　　IP RTP Priority or Priority Queuing (WFQ) and Low Latency Queuing (LLC)

　　In IP telephony voice environments, a priority queue is required for

　　VoIPpackets. LLC provides priority to certain classes along with a

　　minimum guarantee for other classes. This is similar to CQ, however it

　　is easier to configure. With LLC there are four queues: one high-

　　priority queue, two guaranteed-bandwidth queues and a default queue.

　　The default class can be given reserved bandwidth or it can be given an

　　equal share of the bandwidth that is available and unreserved. Below is

　　an example of LLC queueing:

　　access-list 200 permit udp any any range 16384 32000 (voice traffic)

　　access-list 200 permit tcp any any eq 1720 (voice traffic)

　　access-list 201 permit tcp any any eq 80 (web traffic)

　　access-list 202 permit tcp any any eq 62 (SNMP traffic)

　　!

　　class-map voice

　　match access-group 200

　　class-map web

　　match access-group 201

　　class-map snmp

　　match access-group 202

　　!

　　policy-map policy1

　　class voip (voice traffic given highest priority and 64Kbps)

　　priority 64

　　class web (web traffic given lower priority and 64Kbps)

　　bandwidth 64

　　class snmp (snmp traffic given lower priority and 32Kbps)

　　bandwidth 32

　　class class-default

　　fair-queue (all other traffic shares available bandwidth using WFQ)

　　!

　　interface Serial1/1

　　bandwidth 256

　　service-policy output policy1 (applied to outgoing serial interface)

　　Conclusion

　　As can be seen from the topics discussed, there are various ways to

　　manage congestion in a Cisco environment. The techniques discussed can

　　be used in a variety of ways depending on the QoS policies that you

　　want to adhere to. It is critical that you have a complete

　　understanding of the traffic patterns within your environment before

　　you configure any of the queuing techniques discussed.