HUAWEI NetEngine5000E Core Router V800R002C01. Feature Description - QoS. Issue 01 Date HUAWEI TECHNOLOGIES CO., LTD.

Transcription

1 V800R002C01 Issue 01 Date HUAWEI TECHNOLOGIES CO., LTD.

2 2011. All rights reserved. No part of this document may be reproduced or transmitted in any form or by any means without prior written consent of Huawei Technologies Co., Ltd. Trademarks and Permissions and other Huawei trademarks are trademarks of Huawei Technologies Co., Ltd. All other trademarks and trade names mentioned in this document are the property of their respective holders. Notice The purchased products, services and features are stipulated by the contract made between Huawei and the customer. All or part of the products, services and features described in this document may not be within the purchase scope or the usage scope. Unless otherwise specified in the contract, all statements, information, and recommendations in this document are provided "AS IS" without warranties, guarantees or representations of any kind, either express or implied. The information in this document is subject to change without notice. Every effort has been made in the preparation of this document to ensure accuracy of the contents, but all statements, information, and recommendations in this document do not constitute the warranty of any kind, express or implied. Huawei Technologies Co., Ltd. Address: Website: Huawei Industrial Base Bantian, Longgang Shenzhen People's Republic of China i

3 About This Document About This Document Intended Audience This document describes the QoS features in terms of its overview, principle, and applications. This document together with other types of document helps intended readers get a deep understanding of the QoS features. This document is intended for: Network planning engineers Commissioning engineers Data configuration engineers System maintenance engineers Related Versions (Optional) The following table lists the product versions related to this document. Product Name HUAWEI NetEngine5000E Core Router Version V800R002C01 Symbol Conventions The symbols that may be found in this document are defined as follows. Symbol Description Indicates a hazard with a high level of risk, which if not avoided, will result in death or serious injury. Indicates a hazard with a medium or low level of risk, which if not avoided, could result in minor or moderate injury. ii

4 About This Document Symbol Description Indicates a potentially hazardous situation, which if not avoided, could result in equipment damage, data loss, performance degradation, or unexpected results. Indicates a tip that may help you solve a problem or save time. Provides additional information to emphasize or supplement important points of the main text. Change History Updates between document issues are cumulative. Therefore, the latest document issue contains all updates made in previous issues. Changes in Initial field trial release. iii

5 Contents Contents About This Document...ii 1 QoS Overview Introduction to QoS Traditional Packets Transmission Application New Applications Requirements End-to-End QoS Model Best-Effort Service Model Integrated Service Model Differentiated Service Model Techniques Used for the QoS Application Traffic Classification Traffic Policing and Shaping Congestion Avoidance Configuration RSVP Congestion Avoidance and Management Introduction References Principles Basic Principles of Congestion Avoidance Principle of Congestion Avoidance and Management Application Port Queue Introduction to Port Queue References Enhancement Principles QoS Port Queue Queue Scheduling Technology Queue Scheduling Basic Principles of Congestion Avoidance Applications Traffic Classification...31 iv

6 Contents 4.1 Introduction to Traffic Classification References Enhancement Principles Simple Traffic Classification Complex Traffic Classification Applications Traffic Policing and Traffic Shaping Introduction to Traffic Policing References Enhancement Implementation Principle Basic Principles of Traffic Policing Basic Principles of Traffic Shaping Applications Rate Limit on Physical Interfaces Introduction to Rate Limit on Physical Interfaces References Enhancement Principles Principle of Rate Limit on Physical Interfaces Process of Rate Limit on the Physical Interface Applications Impact On the System Performance On Other Features QPPB Introduction to QPPB References Enhancement Principles Principle of QPPB Implementation Mechanism of QPPB Applications...67 v

7 1 QoS Overview 1 QoS Overview About This Chapter This chapter describes the performance measurement of services provided by the service provider. It also introduces some QoS solutions, such as RSVP and Diff-Serv Model. 1.1 Introduction to QoS This section describes the basic concepts of the Quality of Service (QoS), traditional packet delivery services, new demands resulting from new services, and QoS features supported by the product. 1.2 End-to-End QoS Model This section describes the end-to-end service of QoS. 1.3 Techniques Used for the QoS Application Techniques Used for the QoS Application 1

8 1 QoS Overview 1.1 Introduction to QoS This section describes the basic concepts of the Quality of Service (QoS), traditional packet delivery services, new demands resulting from new services, and QoS features supported by the product. QoS is a term that describes the relations between demands and supplies. It is a measurement reflecting the ability of a supplier to meet the demands of customers. Usually, the QoS assessment does not provide the accurate score. Instead, it focuses on the quality of service under certain conditions so that the quality of the service that is found defective can be improved accordingly. On the Internet, QoS is used to assess the ability of the network to transmit packets. The network provides a wide variety of services. Therefore, QoS assesses the service from different aspects. Generally, QoS is used to assess the ability to meet the core requirements for packet transmission, including delay, jitter, and packet loss ratio Traditional Packets Transmission Application It is difficult to ensure QoS in the traditional IP network. Because routers in the network handle all the packets equally and adopt First In First Out (FIFO) method to transfer packets. Resources used for forwarding packets are allocated based on the arrival sequence of the packets. All packets share the bandwidth of networks and routers. Resources are allocated according to the arrival time of the packets. This policy is called best effort (BE). The device in this mode tries its best to transmit packets to the destination. The BE mode, however, does not ensure any improvement in delay time, jitter, packet loss ratio, and high reliability. The traditional BE mode applies only to services such as World Wide Web (WWW), file transfer, and , which have no specific request for bandwidth and jitter New Applications Requirements With the rapid development of the network, increasing number of networks are connected to the Internet. The Internet expands greatly in size, scope, and users. The use of the Internet as a platform for data transmission and implementation of various applications is on the rise. Further, the service providers also want to develop new services for more profits. Apart from traditional applications such as WWW, , and File Transfer Protocol (FTP), the Internet has expanded to accommodate other services such as E-learning, telemedicine, videophone, videoconference, and video on demand. Enterprise users want to connect their branches in different areas through VPN technologies to implement applications such as accessing corporate databases or managing remote devices through Telnet. These new applications put forward special requirements for bandwidth, delay, and jitter. For example, videoconference and video on demand require high bandwidth, low delay, and low jitter. Telnet stresses on low delay and priority handling in the event of congestion. As new services spring up, the number of requests for the service capability of IP networks has been on the rise. Users expect improved service transmission to the destination and also better quality of services. For example, IP networks are expected to provide dedicated bandwidth, reduce packet loss ratio, avoid network congestion, control network flow, and set the preference of packets to provide different QoS for various services. 2

9 1 QoS Overview All these demand better service capability from the network, and QoS is just an answer to the requirements. 1.2 End-to-End QoS Model This section describes the end-to-end service of QoS. Different service models are provided for user services to ensure QoS according to users' requirements and the quality of the network. The common service models are as follows: Best-Effort service model Integrated service model Differentiated service model Best-Effort Service Model Best-Effort is an indiscriminate and the simplest service model. Application programs can, without notifying the network or obtaining any approval from the network, send any number of packets at any time. For the Best-Effort service, the network tries its best to send packets, but cannot ensure the performance such as delay and reliability. The Best-Effort model is the default service model of the Internet and can be applied to most networks, such as FTP and , through the First-in-First-out (FIFO) queue Integrated Service Model The integrated service model is called InterServ for short. IntServ is an integrated service model and can meet various QoS requirements. In this service model, before sending packets, an application program needs to apply for specific services through signaling. The application program first notifies the network of its traffic parameters and the request for special service qualities such as bandwidth and delay. After receiving the confirmation of the network that resources have been reserved for packets, the application program begins sending packets. The sent packets are controlled within the range specified by the flow parameters. After receiving the request for resources from the application program, the network checks the resource allocation. That is, based on the request and current available resources, the network determines whether to allocate resources for the application program or not. Once the network confirms that resources are allocated for the packets, and as long as the packets are controlled within the range specified by the flow parameters, the network is certain to meet the QoS requirements of the application program. The network maintains a state for each flow that is specified by the source and destination IP addresses, interface number, and protocol number. Based on the state, the network classifies packets and performs traffic policing, queuing, and scheduling to fulfil its commitment to the application program. IntServ can provide the following services: Guaranteed service: provides the preset bandwidth and delay to meet the requirements of the application program. For example, a 10 Mbit/s bandwidth and a delay less than one second can be provided for Voice over IP (VoIP) services. Controlled-load service: If network overload occurs, packets can still be provided with the service similar to that provided in the absence of network overload. That is, when traffic congestion occurs on the network, less delay and high pass rate are ensured for the packets of certain application programs. 3

10 1 QoS Overview Differentiated Service Model The differentiated service model is called Diff-Serv for short. In the model, the application program does not need to send its request for network resource before sending the packets. The application program informs network nodes of its demand for QoS by using QoS parameters in the IP packet header. Then routers along the path obtain the demand by analyzing the header of the packet. To implement Diff-Serv, the access router classifies packets and marks the class of service (CoS) in the IP packet header. The downstream routers then identify the CoS and forward the packets on the basis of CoS. Diff-Serv is therefore a flow-based QoS solution. Diff-Serv Model in IP Network Diff-Serv Networking The network node that implements Diff-Serv is called a DS node. A group of DS nodes that adopt the same service policy and the same per-hop behavior (PHB) is called a DS domain. See Figure 1-1. DS nodes are classified into the following two modes: DS border node: Connects DS domain with non-ds domain. This node controls traffic and sets Differentiated Services CodePoint (DSCP) value in packets according to the Traffic Conditioning Agreement (TCA). DS interior node: Connects a DS border node with other interior nodes or connects interior nodes in a DS domain. This node carries out only the simple traffic classification and traffic control based on the DSCP value. Figure 1-1 Diff-Serv networking diagram DS domain DS node DS node DS node Non-DS domain Non-DS domain DS Field and DSCP The Type of Service (ToS) octet in IPv4 packet header is defined in RFC791, RFC134, and RFC1349. As shown in Figure 1-2, the ToS octet contains the following fields: Precedence: 4

11 1 QoS Overview It is of three bits (bits 0 through 2). It indicates the precedence of the IP packet. D bit: It is of one bit and indicates delay. T bit: It is of one bit and indicates throughput. R bit: It is of one bit and indicates reliability. C bit: It is of one bit and indicates cost. The highest bit of ToS field has to be 0. The router first checks the IP precedence of packets to implement QoS. The other bits are not fully used. The ToS octet of IPv4 packet header is redefined in RFC2474, called DS field. As shown in Figure 1-2: Bits 0 through 5 in DS field are used as DSCP. Bit 6 and bit 7 are the reserved bits. Bits 0 through 2 are Class Selector CodePoint (CSCP), which indicate a type of DSCP. DS node selects PHB according to the DSCP value. Figure 1-2 ToS field and DS field IPv4 ToS DS Field Precedence DTRC 0 CSCP unused DSCP The DSCP field within the DS field is capable of conveying 64 distinct codepoints. The codepoint space is divided into three pools as shown in Table 1-1. Table 1-1 Classification of DSCP Code Pool Code Space Usage 1 xxxxx0 Standard action 2 xxxx11 EXP/LU (experiment or local use) 3 xxxx01 EXP/LU (can be used as the extended space for future standard action) Code pool 1 (xxxxx0) is used for standard action, code pool 2 (xxxx11) and code pool 3 (xxxx01) are used for experiment or future extension. Standard PHB The DS node implements the PHB behavior on the data flow. The network administrator can configure the mapping from DSCP to PHB. When a packet is received, the DS node detects its DSCP to find the mapping from DSCP to PHB. If no matching mapping is found, the DS node selects the default PHB (Best-Effort, DSCP=000000) to forward the packet. All the DS nodes support the default PHB. The following are the four standard PHBs defined by the IETF: Class selector (CS), Expedited forwarding (EF), Assured forwarding (AF) and Best-Effort (BE). The default PHB is BE. 5

12 1 QoS Overview CS PHB Service levels defined by the CS are the same as the IP precedence used on the network. The value of the DSCP is XXX000 where the value of "X" is either 1 or 0. When the value of DSCP is , the default PHB is selected. EF PHB EF means that the flow rate should never be less than the specified rate from any DS node. EF PHB cannot be re-marked in DS domain except on border node. New DSCP is required to meet EF PHB features. EF PHB is defined to simulate the forwarding of a virtual leased line in the DS domain to provide the forwarding service with low drop ratio, low delay, and high bandwidth. AF PHB AF PHB allows traffic of a user to exceed the order specification agreed by the user and the ISP. It ensures that traffic within the order specification is forwarded. The traffic exceeding the specification is not simply dropped, but is forwarded at lower service priorities. Four classes of AF: AF1, AF2, AF3, and AF4 are defined. Each class of AF can be classified into three different dropping priorities. AF codepoint AFij indicates AF class is i (1<=i<=4) and the dropping priority is j (1<=j<=3). When providing AF service, the carrier allocates different bandwidth resource for each class of AF. A special requirement for AF PHB is that the traffic control cannot change the packet sequence in a data flow. For instance, in traffic policing, different packets in a service flow are marked with different dropping priorities even if the packets belong to the same AF class. Although the packets in different service flows have different dropping ratio, their sequence remains unchanged. This mechanism is especially applicable to the transmission of multimedia service. BE PHB BE PHB is the traditional IP packet transmission that focuses only on reachability. All routers support BE PHB. Recommended DSCP Different DS domains can have self-defined mapping from DSCP to PHB. RFC2474 recommends code values for BE, EF, AFij, and Class Selector Codepoints (CSCP). CSCP is designed to be compatible with IPv4 precedence model. BE: DSCP= EF: DSCP= AFij codepoint AFij codepoint is shown in Table 1-2. Table 1-2 AF codepoint Service Class Low Dropping Priority, j=1 Medium Dropping Priority, j=2 High Dropping Priority, j=3 AF(i=4) AF(i=3) AF(i=2)

13 1 QoS Overview Service Class Low Dropping Priority, j=1 Medium Dropping Priority, j=2 High Dropping Priority, j=3 AF(i=1) In traffic policing: If j=1, the packet color is marked as green. If j=2, the packet color is marked as yellow. If j=3, the packet color is marked as red. The first three bits of the same AF class are identical. For example, the first three bits of AF1j are 001; that of AF3j are 011, that of AF4j are 100. Bit 3 and bit 4 indicate the dropping priority which has three valid values including 01, 10, and 11. The greater the Bit value, the higher the dropping priority. Class selector codepoint In the Diff-Serv standard, the CSCP is defined to make the DSCP compatible with the precedence field of the IPv4 packet header. The routers identify the priority of the packets through IP precedence. The IP precedence and the CSCP parameters map with each other. The user should configure the values for these parameters. In CSCP, the higher the value of DSCP=xxx000 is, the lower the forwarding delay of PHB is. The default mapping between CSCP and IPv4 precedence is shown in Table 1-3. Table 1-3 The default mapping between IPv4 precedence and CSCP IPv4 Precedence CSCP (in binary) CSCP (in dotted decimal) Service Class BE AF AF AF AF EF EF EF Other codepoints Besides the preceding DSCPs, other DSCPs correspond with BE services. Diff-Serv Model in the MPLS Network EXP field 7

14 1 QoS Overview Defined in RFC3032, MPLS packet header is shown in Figure 1-3. EXP field is of three bits. Its value ranges from 0 to 7 and indicates the traffic type. By default, EXP corresponds to IPv4 priority. Figure 1-3 Position of EXP field LABEL MPLS Header EXP S TTL Processing QoS Traffic in MPLS Domain Processing QoS Traffic on the Ingress Device On the Ingress device of MPLS domain, you can limit the data flow by setting the Committed Access Rate (CAR) to ensure that the data flow complies with MPLS bandwidth regulations. Besides, you can assign different priorities to the IP packets according to certain policies. One-to-one mapping can be achieved since the IP precedence field and the EXP field are both 3 bits. In Diff-Serv domain, however, the DSCP field of IP packet is 6 bits, which is different from the length of EXP and thus leads to many-to-one mapping. It is defined that the first 3 bits of DSCP (that is, CSCP) are mapped with EXP. Processing QoS Traffic on the Device in the MPLS Domain When forwarding the MPLS label, the LSR in MPLS carries out queue scheduling according to the EXP field in the labels of packets that are received. This ensures that packets with higher priority enjoy better service. Processing QoS Traffic on the Egress Device On the Egress device of MPLS domain, you need to map EXP field to DSCP field of IP packet. By standard, the first 3 bits of DSCP (that is, CSCP) take the value of EXP, and the last 3 bits take 0. It should be noted that QoS is an end-to-end solution, while MPLS only ensures that data can enjoy the services regulated in SLA. After the data enters the IP network, IP network ensures QoS. 1.3 Techniques Used for the QoS Application Techniques Used for the QoS Application The primary technologies for implementing Diff-Serv include: Traffic classification Traffic policing Traffic shaping Congestion management Congestion avoidance 8

15 1 QoS Overview Traffic classification is the basis of the QoS application. With this technique, packets are identified based on certain mapping rules. This is a precondition for providing differentiated services. Traffic policing, traffic shaping, congestion management, and congestion avoidance control the network traffic and resource allocation from different aspects. They feature the Diff- Serv concept. The following describes these techniques in detail: Traffic classification: Identifies objects according to specific rules. It is the prerequisite of Diff-Serv and is used to identify packets according to defined rules. Traffic policing: Controls the traffic rate. The rate of the traffic that enters the network is monitored and the traffic exceeding its rate limit is restricted. Only a reasonable traffic range is allowed to pass through the network. This optimizes the use of network resources and protects the interests of the service providers. Traffic shaping: Actively adjusts the rate of outputting traffic. It adjusts the volume of output traffic according to the network resources that can be afforded by the downstream router to prevent unnecessary dropping of packets and congestion. Congestion management: Handles resource allocation during network congestion. It stores packets in the queue first, and then takes a dispatching algorithm to decide the forwarding sequence of packets. Congestion avoidance: Monitors the usage of network resources, and actively drops packets in case of heavy congestion. This addresses the problem of network overload. For the common QoS features in the DiffServ model, see Figure 1-4. Figure 1-4 Common QoS features in the DiffServ model Configure complex traffic classification and traffic policing on the ingress of the network DS domain DS node DS node Configure simple traffic classification, queue scheduling, congestion management, and congestion avoidance. Traffic shaping DS node Non-DS domain Non-DS domain In the IntServ model, the Resource Reservation Protocol (RSVP) is used as signaling for the transmission of QoS requests. When a user needs QoS guarantee, the user sends a QoS request 9

16 1 QoS Overview to the network devices through the RSVP signaling. The request may be a requirement for delay, bandwidth, or packet loss ratio. After receiving the RSVP request, the nodes along the transfer path perform admission control to check the validity of the user and the availability of resources. Then the nodes decide whether to reserve resources for the application program. The nodes along the transfer path meet the request of the user by allocating resources to the user. This ensures the QoS of the user services. In addition, the link efficiency mechanism carries out packet header compression on low-rate links, which greatly improves the efficiency of links. The headers such as IP headers, and User Datagram Protocol (UDP) headers of packets transmitted on the link layer are compressed through the mechanism. This mechanism applies mainly to PPP link layers Traffic Classification When implementing QoS in Diff-Serv model, the router needs to identify each class of traffic. The following are the two methods for the router to classify traffic: Complex traffic classification refers to classifying packets according to more complex rules, for example, the rules combining the link layer, the network layer, and the transport layer information such as the source MAC address, destination MAC address, source IP address, destination IP address, user group number, protocol type, and TCP/UDP port number. Complex traffic classification is generally implemented on the edge routers in the Diff-Serv domain. Simple traffic classification refers to classifying packets according to the IP precedence or DSCP value of the IP packet, the EXP value of the MPLS packet, or the 802.1p priority of the VLAN packet. It is used to simply identify the traffic that has the specific precedence or class of service. A collection of packets of the same traffic classifier is called a Behavior Aggregate (BA). Simple traffic classification is generally implemented on the core device in the Diff-Serv domain Traffic Policing and Shaping In a Diff-Serv domain, traffic policing, and traffic shaping is completed by the traffic conditioner. A traffic conditioner consists of four parts: Meter, Marker, Shaper, and Dropper as shown in Figure 1-5. Meter: Measures the traffic and judges whether the traffic complies with the specifications defined in TCS. Based on the result, the router performs other actions through Marker, Shaper, and Dropper. Marker: Re-marks the DSCP of the packet, and puts the re-marked packet into the specified BA. The available measures include lowering the service level of the packet flow which does not match the traffic specifications (Out-of-Profile) and maintaining the service level. Shaper: Indicates the traffic shaper. Shaper has buffer which is used to buffer the traffic received and ensures that packets are sent at a rate not higher than the committed rate. Dropper: Performs the traffic policing action, which controls the traffic by dropping packets so that the traffic rate conforms with the committed rate. Dropper can be implemented by setting the Shaper buffer to 0 or a small value. 10

17 1 QoS Overview Figure 1-5 Traffic policing and shaping Meter Packets Classifier Marker Shaper/ Dropper In Diff-Serv, routers must support traffic control on the inbound and outbound interfaces simultaneously. The functions of routers vary with their locations. The functions of a router are as follows: The border router processes the access of a limited number of low-speed users. In this way, traffic control on the border router can be completed efficiently. A large amount of traffic classification and traffic control are completed by the border router. The core router only performs PHB forwarding of BA to which packets flow belong. In this way, PHB forwarding can be completed with high efficiency, which also meets the requirements of high-speed forwarding by Internet core network Congestion Avoidance Configuration Causes of Congestion Low QoS in the traditional networks is mainly caused by network congestion. When the available resources temporarily fail to meet the requirements of the service transmission, the bandwidth cannot be ensured. As a result, service rate decreases, resulting in long delay and high jitter. This phenomenon is called congestion. Congestion often occurs in complex packet switching environment of the Internet. It is caused by the bandwidth bottleneck of two types of links, as shown in Figure 1-6. Figure 1-6 Schematic diagram of traffic congestion 100M v1 v2 v2 100M 10M v1 100M 100M Traffic congestion on interfaces operating at different speeds v3 v4 100M Traffic congestion on interfaces operating at the same speed 11

18 1 QoS Overview Congestion Results Congestion Solutions RSVP Packets enter the router at high rate through v1, and are forwarded at low rate through v2. Congestion occurs in the router because the rate of v1 is greater than that of v2. Packets from multiple links enter the router at the rate of v1, v2, and v3. They are forwarded at the same rate of v4 through a single link. Congestion occurs in the router because the total rate of v1, v2, and v3 is greater than that of v4. Congestion also occurs due to the causes as follows: Packets enter the router at line speed. Resources such as available CPU time, buffer, or memory used for sending packets are insufficient. Packets that arrive at the router within a certain period of time are not well controlled. As a result, the network resources required to handle the traffic exceed the available resources. The impact of congestion is as follows: Increases the delay and the jitter in sending packets. Long delay can cause retransmission of packets. Reduces the efficiency of throughput of the network and result in waste of the network resources. Consumes more network resources, particularly storage resources when congestion is aggravated. If not properly allocated, the network resources may be exhausted, and the system may crash. Congestion is the main cause of low QoS. It is very common in complex networks and must be solved to increase the efficiency of the network. Congestion management classifies packets and sends the classified packets to corresponding queues. With the technique of queuing, packets are queued based on a certain queuing policy on the router and then forwarded from the interface based on a certain scheduling policy. Congestion avoidance is generally achieved by discarding packets. When congestion occurs or intensifies, a certain grouping drop policy is used to allocate bandwidths to services of different CoSs, such as EF and AF services. The grouping drop policies include Tail Drop, Random Early Detection (RED), and Weighted Random Early Detection (WRED). Tail Drop When the queue is full, subsequent packets that arrive are discarded. Random Early Detection (RED) When the queue reaches a certain length, packets are discarded randomly. This can avoid global synchronization due to slow TCP start. Weighted Random Early Detection (WRED) When discarding packets, the router considers the queue length and packet precedence. The packets with low precedence are discarded first and are more likely to be discarded. The NE5000E adopts WRED to avoid congestion problems. 12

19 1 QoS Overview RSVP is an end-to-end protocol. Requests for resources are transmitted between nodes through RSVP. The nodes allocate resources at the requests. This is the process of resource reservation. Nodes check the requests against current network resources before determining whether to accept the requests. If the current network resources are quite limited, certain requests can be rejected. Different priorities can be set for different requests for resources. Therefore, a request with a higher priority can preempt reserved resources when network resources are limited. RSVP determines whether to accept requests for resources and promises to meet the accepted requests. RSVP itself, however, does not implement the promised service. Instead, it uses the techniques such as queuing to guarantee the requested service. Network nodes need to maintain some soft state information for the reserved resource. Therefore, the maintenance cost is very high when RSVP is implemented on large networks. RSVP is therefore not recommended for the backbone network. 13

20 2 Congestion Avoidance and Management 2 Congestion Avoidance and Management About This Chapter 2.1 Introduction 2.2 References 2.3 Principles 2.4 Application 14

21 2 Congestion Avoidance and Management 2.1 Introduction Definition Congestion Avoidance Congestion avoidance is a traffic control mechanism that monitors the network resources such as queues and buffer memory. When network congestion is found of tending to intensify, the router actively discards packets to regulate network traffic so that the network is free from overload. Congestion Management Congestion management provides means to manage and control traffic when traffic congestion occurs. The queue scheduling technology is used to handle traffic congestion. Packets sent from one interface are placed into many queues which are identified with different priorities. Packets are then sent according to the priorities. A proper queue scheduling mechanism can provide packets of different types with reasonable QoS features such as the bandwidth, latency, and jitter. The queue here refers to the outgoing packet queue. Packets are buffered into queues before the interface is able to send them. Therefore, the queue scheduling mechanism works only when an outbound interface is congested. The queue scheduling mechanism can re-arrange the order of packets except those in First In First Out (FIFO) queues. Purpose Congestion avoidance and management is a traffic control mechanism to regulate network traffic so that the network is free from overload. Benefits Benefits Brought to Users Packets with high priorities will be assured firstly when traffic congestion occurs. 2.2 References None. 2.3 Principles Basic Principles of Congestion Avoidance Congestion avoidance is a traffic control mechanism used to discard packets according to the queue status when the network is congested. Through congestion avoidance, the QoS of traffic is improved when the network is congested. The traditional solution adopted by congestion avoidance is tail drop. That is, all arriving packets are discarded when the network is congested. If a large number of packets from a TCP connection are discarded, the TCP connection times out and enters the slow start state. Then, the TCP connection sends fewer packets. When packets from multiple TCP connections are discarded in a queue, these TCP connections enter the congestion avoidance and slow start state at the same 15

22 2 Congestion Avoidance and Management time, which is referred to as global TCP synchronization. Thus, these TCP connections simultaneously send fewer packets to the queue so that the rate of incoming packets is smaller than the rate of outgoing packets, which reduces the bandwidth usage. To avoid the preceding problems, packet discarding must be done before the queue is to be congested. WRED is a congestion avoidance mechanism used to discard packets to prevent queues from being congested. WRED discards at probabilities increasing packets that may cause congestion. Thus, the bandwidth consumed by outgoing interfaces of TCP connections is reduced slowly, which does not cause the slow synchronization of a large number of TCP connections. This also reduces the average queue length and shortens the delay for sending traffic. The NE5000E uses the both the tail-drop and the WRED algorithms for congestion avoidance. In the DiffServ model, the router preserves eight service queues for each port. The queues map the following service types respectively: BE, AF1 to AF4, EF, CS6, and CS7. By default, AF1 to AF4 and BE queues are applied with the WFQ scheduling; they are allocated with bandwidth according to configured weight parameters. EF, CS6, and CS7 queues are configured to the PQ scheduling by default. WRED Algorithm The RED algorithm can better solve the problem of the global TCP synchronization. This algorithm, however, cannot sense any QoS signaling: all types of packets are considered equally. Therefore, this algorithm is less flexible. To adopt differentiated discarding policies to different types of packets, the weighted random early detection (WRED) algorithm is introduced. The WRED algorithm is similar to the RED algorithm. In the WRED algorithm, each queue is also set with a minimum threshold and a maximum threshold. Apart from this, When a queue is shorter than the minimum threshold, the router does not discard packets. When a queue is longer than the maximum threshold, the router discards all incoming packets. When the length of a queue is between the minimum threshold and the maximum threshold, the router discards packets in a random order. By the random way, each arriving packet is applied with a random number. This random number is compared with the current discarding probability of the current. If the number is greater the discarding probability, this packet is discarded. The longer a queue, the higher the discarding probability. The probability, however, cannot exceed the high limit. In addition, the average queue length is used to compare with the minimum threshold and the maximum threshold so that burst traffic is processed unfairly. Different from the RED algorithm, the random numbers produced by the WRED algorithm is based on the precedence. In the WRED mechanism, the DSCP value that indicates the IP precedence is introduced to identify discarding policies. You can set different DSCP values for the queue length, queue threshold, and drop probability so that packets of different precedence are applied with different discarding probability. This is the important feature of the WRED algorithm. When the weighted fair queue (WFQ) is used in the queuing mechanism, packets of different precedence can be set with different minimum threshold, maximum threshold, and drop probability. In this way, packets of different precedence are provided with different discarding features. 16

23 2 Congestion Avoidance and Management When the FIFO, PQ, and CQ are used in the queuing mechanism, you can set different minimum threshold, maximum threshold, and drop probability for each queue so that packets of different types are provided with different discarding features. Figure 2-1 shows the relations between the WRED and queues. Figure 2-1 Relations between the WRED and queues WRED drop Queue1 weight1 Packets to be sent from this interface Queue2 weight2 Forwading queue... Classifying QueueN-1 weightn-1 Scheduling Forwarded packets QueueN weightn Dropped packets Congestion Avoidance Algorithms PQ The discarding policy for the PQ can be the tail-drop or the WRED. The services that demand high real-time performance are usually applied with the tail-drop policy. These packets must be provided with the QoS guarantee to a large extent. The tail-drop policy means that the router discards packets only when a queue reaches the length threshold. The PQ scheduling preempts the bandwidth of other services; therefore, when traffic congestion occurs, real-time services are guaranteed with the bandwidth to the maximum. WFQ The default discarding policy for the WFQ is tail-drop but in reality the WRED is mostly adopted. The WFQ scheduling is often applied to the packets of low precedence and those insensitive to latency. You can use the WFQ and the WRED together to configure different discarding parameters to different types of traffic so that different purposes are reached. You can configure a template on an device to realize the WRED. First define WRED templates: set the maximum and minimum thresholds for packets in different colors and set the drop probability. Then apply the WRED templates for different levels of quality on the interface. You can configure a maximum of eight WRED templates for queues on an interface. Each template supports the process of packets of no more than three colors. These packets are defined as red, yellow, and green packets. Generally green packets are set to low drop probability and high threshold while red packets are set to high drop probability and low threshold. You can configure packets of different colors with a different thresholds and drop probabilities flexibly. When traffic congestion occurs, a queue begins to buffer packets. According to the classification of packets, red packets are set to low threshold and high drop probability; 17

24 2 Congestion Avoidance and Management therefore, the red packets begin to be dropped first. When the queue is long enough, green packets begin to be dropped. When the queue length reaches the maximum threshold of a color, packets of this color begin to be applied with the tail-drop policy. Because the WFQ queues share the bandwidth in proportion, traffic congestion occurs easily. The use of the WRED policy can effectively prevent the global TCP synchronization. Currently, the device supports the application of the WRED policy only on outbound interfaces. The WRED parameter granularity is packet-based. Congestion Avoidance of Flow Queues The upstream and downstream FQs support the WRED and tail drop mechanisms. Congestion Avoidance of Port Queues The downstream CQ supports the WRED and tail drop mechanisms Principle of Congestion Avoidance and Management Common Queue Scheduling Algorithms FIFO FIFO is the simplest queuing algorithm. One interface has only one FIFO queue. Therefore, FIFO does not need traffic classification; one queue is unnecessarily to be scheduled. FIFO handles only the queue length, which has effect on the latency and packet drop ratio. FIFO queuing works under the tail drop mechanism. In the tail drop mechanism, when a queue is full, all further packets to join the queue are dropped. No means is provided to let later packets take up the positions of the packets already in the queue. Figure 2-2 FIFO queuing Queue Packets out Packets to be sent from this interface Scheduling PQ As shown in Figure 2-2, FIFO does not classify packets. When packets enter the interface at a rate higher than the ability the interface can support, FIFO lets the packets that come earlier to enter the queue first. At the outbound interface, FIFO lets the packets leave the interface in the same order as when the packets enter the interface. This is called first in, first out for short. In the FIFO mechanism, if a queue is defined to be too long, the queue is not easy to be full and fewer packets are discarded. But long queue results in long latency. If a queue is defined to be too short, latency is short but more packets are discarded. In configuration, you must balance between the two factors to achieve a favorable result. Such a problem also exists in other queue scheduling mechanisms. In the Priority Queuing (PQ) mechanism, queues are generally classified into four levels, namely, top, middle, normal, and bottom, from high to low in priority. 18

25 2 Congestion Avoidance and Management NOTE On the device, queues are classified into eight priority levels, from 0 to 7. As shown in Figure 2-3, when packets arrive, PQ organizes the packets into four classes. Each class of packets is sent to one of the four PQ queues. Figure 2-3 Priority queuing Queues Top Packets to be sent from this interface Classifying Middle Normal Bottom Scheduling Packets out When packets leave a queue, PQ lets the packets from the queue of the top priority go first. Packets from this queue keep being sent until the queue is empty. When the packets from the queue of the top priority are all sent, packets from the queue of middle priority are sent. When the packets from the queue of the middle priority are all sent, the packets from the queue of the normal priority are sent; finally, the packets from the queue of the bottom priority are sent. In this way, packets from the queue of high priority are sent earlier according to the classification. When congestion occurs, packets from the queue of high priority are still authorized to leave earlier. This makes the packets of important services such as the enterprise resource planning (ERP) service are handled earlier. The packets of not so important services such as the service are handled late until the packets of important services are all sent up and the network is idle. As a result, key services are handled first and network resources are also fully used. PQ has the following features: ACLs can be used for packets classification and then classified packets are put into different queues as required. The tail drop mechanism is used as the only packet drop policy when congestion occurs. The queue length can be set to 0, meaning this queue is infinite in length. Packets in this queue are never dropped according to the tail drop policy only if the memory is available. The FIFO is used in the queue internally. In queue scheduling, packets from the queue of high priority are scheduled first. PQ has also obvious advantages and obvious disadvantages as follows: Advantages: Packets from the queue of high priority are provided with higher bandwidth, lower latency, and less jitter. Disadvantages: Packets from the queue of low priority are not scheduled in time so that they keep "starving." To solve the "starving" problem of PQ, the CQ queue scheduling mechanism comes into being. The mechanism contains 17 queues numbered from 0 to 16. Queue 0 is provided with the highest priority. The router handles queues 1 to 16 only after the packets from queue 0 are all sent. Because of this feature, queue 0 is usually used as the system queue. 19

26 ... HUAWEI NetEngine5000E Core Router 2 Congestion Avoidance and Management CQ uses the Round Robin algorithm. The advantage is that all queues can be served. WFQ The weighted fair queuing (WFQ) is a complex queuing algorithm. With this algorithm, services of the same priority are processed in fair manner. Services of different priorities are weighted before being processed. The number of queues can be preset, ranging from 16 to WFQ handles packets fairly in terms of the bandwidth and latency. Meanwhile, the weight of different types of packets is also concerned. The weight is determined by the IP precedence carried in the IP header. The WFQ organizes packets into different classes dynamically according to the quintet (IP address, destination address, source port number, destination port number, and protocol ID) or according to the ToS value in packets. Packets of the same IP address, destination address, source port number, destination port number, and protocol ID, or the same ToS belong to one flow. Each flow maps one queue. This process is called the Hash queuing. Queuing is carried out automatically based on the Hash algorithm: different types of traffic are placed into different queues. When packets leave a queue, WFQ gives a share of bandwidth for the queue according to the precedence of the traffic. The smaller the priority is, the less bandwidth the packets obtain. The higher the priority, the more bandwidth the packets obtain. In this way, services of the equal priority are treated fairly. Meanwhile, services of different priority are given different weights. Figure 2-4 shows the WFQ queuing principle. Figure 2-4 WFQ queuing principle Queuing Queue 1 Classifying Queue 2 Packets out Packets to be sent from this interface Queue N Scheduling Suppose on the current interface there are eight flows, mapping the priorities of 0, 1, 2, 3, 4, 5, 6, and 7 respectively. The total bandwidth quotas are the sum of the priority level number + 1. That is, = 36 The bandwidth proportion of each flow is: (Own priority level number + 1)/(sum of all priority number + 1). That is, the shares of bandwidth of all flows are 1/36, 2/36, 3/36, 4/36, 5/36, 5/36, 6/36, 7/36, and 8/36. Another example: Suppose the on the current interface there are four flows. Three of the flows have a priority level of 4 each and one of them has a priority level of 5. (4 + 1) * 3 + (5 + 1) = 21 The proportion of bandwidth given to the three flows with the priority of 4 is 5/21 each. The proportion of bandwidth given to the one flow with the priority of 5 is 6/21. So, WFQ apply weights to services of different priority on the fair basis. The weight is determined by the IP precedence carried in the packet header. The WFQ has the following features: 20

27 2 Congestion Avoidance and Management Packets are classified according to the quintet. The classification cannot be customized. The WFQ drop policy is used. This is an improvement of the tail drop policy. The WFQ is based on flows. Each flow occupies one queue. One interface supports a maximum of 4096 queues. Queues are applied with different WFQ scheduling policies. Inside a queue, the FIFO algorithm is used. Another queue scheduling mechanism is called the class-based weighted fair queuing (CBWFQ). The CBWFQ is similar to the CQ. In the CBWFQ mechanism, each queue is reserved with a minimum bandwidth. The classification of packets for the CBWFQ is similar to that of the CQ. Different from the CQ, you can configure the actual percentage of bandwidth for each flow in the CBWFQ rather than bytes. Implementation of Congestion Management In queue configuration, you do not need to care about what scheduling algorithms are used. You just need to care about the exterior traffic features (expressed with parameters) of the service carried in a queue, for example, the bandwidth to be guaranteed, the bandwidth at the peak time, and the proportion for taking up the remaining bandwidth. A scheduling algorithm is chosen according to the configured traffic parameters so that user's configuration is guaranteed. The queue scheduling for the device consists of two stages: the traffic rate limit and queue scheduling on the interface. The following are configurable parameters for queues: Peak burst size: The peak burst size (PBS) is set to provide guaranteed peak bandwidth. PQ algorithm: It is a scheduling algorithm based on priority. WFQ algorithm: It is a scheduling algorithm based on weight. In interface queue scheduling, PQ or WFQ algorithm is used. The advantages of interface queue scheduling are: Real-time services sensitive to latency is provided with guaranteed quality; services of high priority can be provided with prioritized bandwidth. Traffic flows of different priorities can be assigned with different bandwidth according the weight. In the DiffServ model, the router reserves eight service queues for each interface. These queues map the service types of BE, AF1 to AF4, EF, CS6, and CS7. By default, AF1 to AF4 and BE queues are configured to the WFQ scheduling scheme; bandwidth is distributed proportionally according to the preset weight. The EF, CS6, and CS7 queues are configured to the PQ scheduling scheme by default. This scheduling is based on absolute priorities. PQ scheduling is used in services sensitive to latency. 21

28 2 Congestion Avoidance and Management 2.4 Application Networking and Application of Congestion Avoidance Figure 2-5 Typical networking for congestion avoidance Server /24 Telephone /24 GE1/0/ /24 RouterA RouterB GE2/0/ /24 Network PC /24 PC /24 As shown in Figure 2-5, devices Server, Telephone, PC1 and PC2 all send data to the network through Router A. The data sent from Server is of critical traffic class; the data sent from Telephone is of voice services; the data from PC1 and PC2 is of normal services. Because the rate of the inbound interface GE 1/0/0 on Router A is greater than that of the outbound interface GE 2/0/0, congestion may occur on GE 2/0/0. When network congestion occurs, the data sent by Server and Telephone must be transmitted first. Users PC1 and PC2 allow a little delay to the transmission of their data but they also require bandwidth guarantee because they are VIP users. Therefore, Router A must discard packets based on the priority of the packets when the network congestion intensifies. Networking and Application of Congestion Management Figure 2-6 Typical example of congestion management /16 S0 S1 RouterA RouterB As shown in Figure 2-6, Router A in the LAN of a company sends packets through the S0 interface to Router B in the WAN. Because the bandwidth of the WAN is less than that of the LAN, the S0 interface on Router A is easy to be congested. The queuing technologies need to be used to manage and control the congested interface. First classify packets to be sent from the S0 interface and place them into many different queues. 22

29 2 Congestion Avoidance and Management Then process the queues respectively according to the priorities. Packets of high priorities are handled first. Take the PQ scheduling as an example. You can set the router as follows: The packets that conform to the rules in ACL1 enter the top queue. The packets that arrive at the S1 interface enter the normal queue. The default queue is the middle queue. After setting the maximum length of the queues, apply the PQ rule group 1 to the S0 interface. This setting results in the fact that the PQ scheduling processes services with different ways. Packets of high priority can be forwarded normally; congestion on the interface is released. 23

30 3 Port Queue 3 Port Queue About This Chapter 3.1 Introduction to Port Queue 3.2 References 3.3 Enhancement 3.4 Principles 3.5 Applications 24

31 3 Port Queue 3.1 Introduction to Port Queue Definition Purpose Port queue (PQ) is a technology used to guarantee the bandwidth for the multi-user multi-service environment in the Differentiated Service (DiffServ) model through a queue scheduling mechanism. DiffServ is a type of the class-based Quality of Service (QoS) technology. DiffServ provides differentiated services for different service flows. Therefore, in the DiffServ scheme, service flows need to be classified based on service requirements first. Traffic classification is the prerequisite and basis of Hierarchical QoS (HQoS). For details of DiffServ, see the Traffic Classification. With the rapid development of network equipment, the capacity of a single port increases and access users grow in number. In this case, new problems pop up in applications of the traditional QoS. Traditional traffic management schedules traffic based on the bandwidth of ports. As a result, traffic management is sensitive to the class of services rather than users, which is applicable to traffic at the network core side rather than traffic at the service access side. It is of great difficulty for traditional traffic management to simultaneously control multiple services of many users. To solve the problems and provide better QoS, a QoS technology that can carry out queue scheduling based on service priorities and control user traffic is in urgent demand. Combined with the DiffServ scheme, HQoS adopts five levels of scheduling. HQoS enables the equipment to perform policy-based control over internal resources with the existing hardware. It can both provide the quality assurance for the advanced users and reduce the total cost of the network construction. 3.2 References None. 3.3 Enhancement None. 3.4 Principles QoS Port Queue A queue is a method of storing packets during the forwarding process. When the rate of traffic exceeds the bandwidth on a port or the bandwidth set for the traffic, packets are placed into the queues in the buffer. The time and sequence for packets leaving related queues and the scheduling of packets in various queues are determined by scheduling policies. 25

32 3 Port Queue Each port has eight downstream queues (include CS7, CS6, EF, AF1, AF2, AF3, AF4, BE)on the NE5000E. The PQ or WFQ scheduling is adopted for port queue scheduling. The PQ or WFQ scheduling can ensure the real-time service that is sensitive to delay, guarantee the bandwidth for the service with a higher priority, and allocate different bandwidths for the flows of different priorities based on the configured weights. Packets of control protocols, such as packets of routing protocols, have the highest priority. These packets are placed in the queues of CS6 and CS7. Packets for device management, such as CLI packets, SNMP packets, and SSH packets, are placed in the AF queue Queue Scheduling Technology RR The queue scheduling mechanism is a very important technology in QoS. When the congestion occurs, a proper queue scheduling mechanism can provide packets of a certain type with proper QoS features such as the bandwidth, delay, and jitter. The queue scheduling mechanism works only when the congestion occurs. The commonly used queue scheduling technologies include Weighted Fair Queuing (WFQ), Weighted Round Robin (WRR), and PQ. RR, short for Round Robin, is a simple scheduling method used to schedule multiple queues. Figure 3-1 Schematic diagram of RR WRR RR schedules multiple queues by polling the queues in circular order. If the queue scheduled through RR is not empty, the scheduler takes one packet away from the queue. If the queue is empty, the queue is skipped and the scheduler does not wait. Compared with RR, WRR can set the weights of queues. During the WRR scheduling, the scheduling chance obtained by a queue is in direct proportion to the weight of the queue. 26

33 3 Port Queue During the WRR scheduling, different amounts of packets are scheduled from the queues according to the weights of the queues. Each time a queue is scheduled, a packet is taken away from the queue. In a round of the WRR scheduling, the queues with the larger weights are scheduled many times, as shown in Figure 3-2. Figure 3-2 Schematic diagram of WRR During the WRR scheduling, the empty queue is skipped, and the period of circular scheduling is shortened. Therefore, when there is a small volume of traffic in a queue, the remaining bandwidth of the queue is used by other queues according to a certain proportion. WFQ+PQ PQ is based on the absolute priority. WFQ is used to assign bandwidth to queues taking part in the scheduling according to the weights of the queues. If the weight of a queue is 0, it indicates that there is no limitation on the bandwidth of the queue. That is, PQ is configured for the queue. Therefore, the technology is called WFQ +PQ. In WFQ, unused bandwidth is reassigned. WFQ can thus ensure the minimum bandwidth of queues according to the weights of the queues. Comparison Between the Scheduling Technologies Scheduli ng Algorith ms Complexity Delay/Jitter Fairness RR It is easy to implement. In case of low-speed scheduling, delay and jitter intensify. The packets in the queues are scheduled according to the length of packets. 27

34 3 Port Queue Scheduli ng Algorith ms Complexity Delay/Jitter Fairness WRR It is easy to implement. In case of low-speed scheduling, delay and jitter intensify. Packet length affects the implementation of packet scheduleing. WFQ It is complex to implement. The delay is controlled properly and the jitter is low. The queues are scheduled fairly at the granularity of bytes. PQ It is easy to implement. The delay is controlled properly and the jitter is low for queues of high priority. The queues of high priority are processed ahead of the queues of low priority. The queues of low priority are processed only after the queues of high priority are sent out. Therefore, important services are processed ahead of other services. The bandwidths are not allocated rationally. When a large amount of high-priority traffic is generated, lowpriority traffic is allocated insufficient or even no bandwidth Queue Scheduling Port Queue Scheduling NOTE The NE5000E can only support configure the scheduling policy on the outbound. The inbound class queue with the specified CoS is scheduled as the default scheduling policy. To ensure the allocation of available bandwidth, the upstream port queue scheduling is classified into the following levels: RR scheduling for traffic with the same CoS WFQ scheduling for multicast traffic with the same CoS and unicast traffic with the same CoS PQ+WFQ scheduling for traffic with different CoSs The downstream port queue scheduling is classified into the following levels: 28

35 3 Port Queue WFQ scheduling for traffic with different CoSs on a port WRR scheduling for traffic between ports PQ shaping and port queue shaping Basic Principles of Congestion Avoidance Congestion avoidance is a traffic control mechanism used to discard packets according to the queue status when the network is congested. Through congestion avoidance, the QoS of traffic is improved when the network is congested. The traditional solution adopted by congestion avoidance is tail drop. That is, all arriving packets are discarded when the network is congested. If a large number of packets from a TCP connection are discarded, the TCP connection times out and enters the slow start state. Then, the TCP connection sends less packets. When packets from multiple TCP connections are discarded in a queue, these TCP connections enter the congestion avoidance and slow start state at the same time, which is referred to as global TCP synchronization. Thus, these TCP connections simultaneously send less packets to the queue so that the rate of incoming packets is smaller than the rate of outgoing packets, which reduces the bandwidth utilization. To avoid the preceding problems, packet must be discarded before the queue is congested. WRED is a congestion avoidance mechanism used to discard packets to prevent queues from being congested. WRED discards at probabilities increasing packets that may cause congestion. Thus, the bandwidth consumed by outbound interfaces of TCP connections is reduced slowly, which does not cause global TCP synchronization of a large number of TCP connections. This also reduces the average queue length and shortens the delay for sending traffic. Congestion Avoidance of Port Queues The downstream PQ supports the WRED and tail drop mechanisms. 3.5 Applications Figure 3-3 Schematic diagram of queue scheduling /16 S0 S1 RouterA RouterB As shown in Figure 3-3, when Router A on the LAN sends data packets to Router B on the WAN through the interface S0, the interface S0 on Router A is congested. This is because the bandwidth of a WAN is usually less than the bandwidth of a LAN. To manage and control the congested interface, you need to adopt queue scheduling. That is, classify all the packets sent from S0 and send them to various queues. These packets in the queues can be processed according to the priorities of the queues. Packets in the queue of a higher priority can be processed first. 29

36 3 Port Queue Take PQ as an example. You can send the packets that match ACL1 to the top queue and send the incoming packets that pass through S1 to the normal queue. Packets are sent to the middle queue by default. After the maximum lengths of various queues are set, PQ rule 1 is applied on S0. In this manner, different services are processed differently to ensure the normal forwarding of packets of a higher priority and manage the interface congestion. 30

37 4 Traffic Classification 4 Traffic Classification About This Chapter 4.1 Introduction to Traffic Classification 4.2 References 4.3 Enhancement 4.4 Principles 4.5 Applications 31

38 4 Traffic Classification 4.1 Introduction to Traffic Classification Definition Purpose Based on certain rules defined according to certain information contained in packets, traffic classification classifies the packets, and then implements different QoS policies for the packets matching different rules. Based on the classification rules, traffic classification is classified into the two types: simple traffic classification and complex traffic classification. Simple traffic classification Simple traffic classification refers to classifying data packets based on multiple priorities or service classes. If the first three bits (IP priority) of the ToS field in the header of IP packets are used to mark the packets, the packets can be classified into a maximum of eight classes. If the Differentiated Services Code Point (DSCP), the first six bits of the ToS field, is used to mark packets, the packets can be classified into a maximum of 64 classes. After the packets are classified, QoS features can be applied to different classifiers to implement classifier-based congestion management and traffic shaping. The network administrator can set BA policies for packets, including the BA policies based on IP preferences or the DSCP values of the IP packets, the EXP values of the MPLS packets, and the 802.1p values of the VLAN packets. Complex traffic classification MF refers to classifying packets based on the information such as the quintuple (source IP address, source port number, protocol number, destination IP address, destination port number) and the TCP SYN (the general basis for traffic classification is limited to the header information of encapsulated packets, and packet contents are seldom used as the standards for classification). MF is configured at the edge of a network by default. The network administrator can flexibly configure classification rules on the edge node. There is no restriction on classification rules. The classification result can be a narrow range defined by a quintuple (including source IP address, source port number, protocol number, destination IP address, and destination port number) or all packets destined for a network segment. BA can retrieve data according to the priority fields of packets, search for entries, and implement traffic behaviors, which does not affect the forwarding performance. MF needs to extract packet information, calculate the key value, search for matching rules and match the packets with the rules, obtain data according to the index, and implement traffic behaviors, which affects the forwarding performance. Traffic classification provides differentiated services for the traffic of users in the DiffServ domain. Due to the characteristics of the traffic model and service model on the IP network, the Internet backbone network needs to provide services for thousands of service flows at the same time. As a result, the solution of reserving bandwidth for each traffic flow cannot be implemented on the Internet backbone, which seriously restricts the IntServ application on the practical network. The IntServ application is also restricted by other factors such as the large-scale deployment of the RSVP signaling, the interworking between the devices of different manufacturers, and the 32

39 4 Traffic Classification management (including authentication and accounting) based on services. Therefore, the IntServ has not been used for the commercial purpose since DiffServ, as an improvement of IntServ, is a class-based QoS technology. On the ingress of the network, DiffServ is used to implement traffic classification and traffic control based on service requirements and set the ToS fields of the packets. DiffServ is also used to differentiate the communication types based on the values of the ToS fields in packets and provide QoS services including resource allocation, queue scheduling, and packet discarding policies, which are called Per Hop Behaviors (PHBs). All the nodes in the DiffServ domain carry out PHBs based on the DSCP fields of packets. The DiffServ model classifies services into limited classes, which improves the service scalability. The DiffServe model provides different services for different types of traffic. Therefore, in the DiffServ scheme, the service traffic needs to be classified based on service requirements, which is the prerequisite and basis for the differentiated service. 4.2 References The references of this feature are as follows: Document Description Remarks RFC 2597 Assured Forwarding PHB - RFC2598 Expedited Forwarding PHB - RFC 2697 A Single Rate Three Color Marker - RFC 2698 A Two Rate Three Color Marker - FRC2474 Definition of the Differentiated Services Field (DS Field) in the IPv4 and IPv6 Headers Enhancement None. 4.4 Principles Simple Traffic Classification Simple traffic classification implements the mapping between the internal priority and the external priority. Packets are classified based on the DSCP values of IP packets, the EXP values of MPLS packets, and the 802.1p values of VLAN packets. Then, the mappings between the priorities of the packets on different networks are created. The DiffServ domain consists of the connected DiffServ nodes that adopt the same service policies and PHB set. The traffic policies are bound to the DS domain (or 802.1p) through commands on interfaces to implement simple traffic classification. Simple traffic classification can be classified into two types: upstream simple traffic classification and downstream simple traffic classification. 33

40 4 Traffic Classification Upstream simple traffic classification: Based on the DSCP values of IP packets, EXP values of MPLS packets, and 802.1p values of VLAN packets, the packets are classified into eight service classes (CS7, CS6, EF, AF4 to AF1, and BE) and marked in three colors (green, yellow, and red). If the service classes of packets is EF, BE, CS6, or CS7, the packets can be marked only green. Upstream simple traffic classification is used to differentiate services such as voice, video, and data services. During congestion management and queue scheduling, different services enter different queues. Thus, different scheduling solutions are implemented. For example, voice services can enter the PQ queue of a higher priority to minimize the delay. If upstream simple traffic classification is not implemented, the service class of all the packets is BE. Downstream simple traffic classification: Based on the service classes (CS7, CS6, EF, AF4 to AF1, and BE) and three colors (green, yellow, and red), the DSCP values of IP packets, the EXP values of MPLS packets, or the 802.1p values of VLAN packets are re-set. The downstream simple traffic classification implements the re-marking function, that is, remarking the DSCP values of IP packets, the EXP values of MPLS packets, or the 802.1p values of VLAN packets. Figure 4-1 Mappings of upstream and downstream simple traffic classification Upstream Mapping Based on the DSCP Values of IP Packets According to the mappings between the DSCP values of IP packets and internal priorities, the scheduling priority and color of the packets are specified on routers based on the DSCP values of the packets, which ensures the proper scheduling of the packets. 34

41 4 Traffic Classification Table 4-1 Default mappings between DSCP values of IP packets and service classes DSCP Service Color 00 BE Green 01 BE Green 02 BE Green 03 BE Green 04 BE Green 05 BE Green 06 BE Green 07 BE Green 08 AF1 Green 09 BE Green 10 AF1 Green 11 BE Green 12 AF1 Yellow 13 BE Green 14 AF1 Red 15 BE Green 16 AF2 Green 17 BE Green 18 AF2 Green 19 BE Green 20 AF2 Yellow 21 BE Green 22 AF2 Red 23 BE Green 24 AF3 Green 25 BE Green 26 AF3 Green 27 BE Green 28 AF3 Yelloe 35

42 4 Traffic Classification DSCP Service Color 29 BE Green 30 AF3 Red 31 BE Green 32 AF4 Green 33 BE Green 34 AF4 Green 35 BE Green 36 AF4 Yellow 37 BE Green 38 AF4 Red 39 BE Green 40 EF Green 41 BE Green 42 BE Green 43 BE Green 44 BE Green 45 BE Green 46 EF Green 47 BE Green 48 CS6 Green 49 BE Green 50 BE Green 51 BE Green 52 BE Green 53 BE Green 54 BE Green 55 BE Green 56 CS7 Green 57 BE Green 58 BE Green 36

43 4 Traffic Classification DSCP Service Color 59 BE Green 60 BE Green 61 BE Green 62 BE Green 63 BE Green Upstream Mappings Based on the 802.1p Values of VLAN Packets According to the mappings between the 802.1p values of VLAN packets and the internal priorities, the scheduling priority and color of the packets are specified on routers based on the 802.1p values of the packets, which ensures the proper scheduling of the packets. Table 4-2 Default mappings between 802.1p values of VLAN packets and service classes 802.1p Service Color 0 BE Green 1 AF1 Green 2 AF2 Green 3 AF3 Green 4 AF4 Green 5 EF Green 6 CS6 Green 7 CS7 Green Upstream Mappings Based on the EXP Values of MPLS Packets According to the mapping between the EXP values of MPLS packets and internal priorities, the scheduling priority and color of the packets are specified on routers according to the EXP values of the packets, which ensures the proper scheduling of the packets. Table 4-3 Default mappings between EXP values of MPLS packets and service classes Exp Service Color 0 BE Green 1 AF1 Green 2 AF2 Green 37

44 4 Traffic Classification Exp Service Color 3 AF3 Green 4 AF4 Green 5 EF Green 6 CS6 Green 7 CS7 Green Priority Mapping of Outgoing Packets A router searches for the internal scheduling priority and color of packets based on the DSCP values, 802.1p values, or EXP values of the packets on the outbound interface. After performing the internal scheduling, the router adds the priority fields such as DSCP values, 802.1p values, and EXP values to the packets to be sent according to the internal scheduling priority and color of the packets Simple Traffic Classification Simple traffic classification implements the mappings between the internal priorities and the external priorities. Packets are classified based on the DSCP values of IP packets, the EXP values of MPLS packets, and the 802.1p values of VLAN packets. Then, the mappings between the priorities of the packets on different networks are created. The DiffServ domain consists of the connected DiffServ nodes that adopt the same service policies and PHB set. The traffic policies are bound to the DS domain (or 802.1p) through related commands on interfaces to implement simple traffic classification. Simple traffic classification can be classified into two types: upstream simple traffic classification and downstream simple traffic classification. Upstream simple traffic classification: Based on the DSCP values of IP packets, EXP values of MPLS packets, and 802.1p values of VLAN packets, the packets are classified into eight service classes (CS7, CS6, EF, AF4 to AF1, and BE) and marked with three colors (green, yellow, and red). If the service class of packets is EF, BE, CS6, or CS7, the packets can be marked only green. Upstream simple traffic classification is used to differentiate services such as voice, video, and data services. During congestion management and queue scheduling, different services enter different queues. Thus, different scheduling solutions are implemented. For example, voice services can enter the PQ queue of a higher priority to minimize the delay. If upstream simple traffic classification is not implemented, the service class of all the packets is BE. Downstream simple traffic classification: Based on the service classes (CS7, CS6, EF, AF4 to AF1, and BE) and three colors (green, yellow, and red), the DSCP values of IP packets, the EXP values of MPLS packets, or the 802.1p values of VLAN packets are re-set. 38

45 4 Traffic Classification Figure 4-2 Mappings of upstream and downstream simple traffic classification Upstream Mapping Based on the DSCP Values of IP Packets According to the mappings between the DSCP values of IP packets and internal priorities, the scheduling priority and color of the packets are specified on routers based on the DSCP values of the packets, which ensures the proper scheduling of the packets. Table 4-4 Default mappings between DSCP values of IP packets and service classes DSCP Service Color 00 BE Green 01 BE Green 02 BE Green 03 BE Green 04 BE Green 05 BE Green 06 BE Green 07 BE Green 08 AF1 Green 09 BE Green 39

46 4 Traffic Classification DSCP Service Color 10 AF1 Green 11 BE Green 12 AF1 Yellow 13 BE Green 14 AF1 Red 15 BE Green 16 AF2 Green 17 BE Green 18 AF2 Green 19 BE Green 20 AF2 Yellow 21 BE Green 22 AF2 Red 23 BE Green 24 AF3 Green 25 BE Green 26 AF3 Green 27 BE Green 28 AF3 Yelloe 29 BE Green 30 AF3 Red 31 BE Green 32 AF4 Green 33 BE Green 34 AF4 Green 35 BE Green 36 AF4 Yellow 37 BE Green 38 AF4 Red 39 BE Green 40

47 4 Traffic Classification DSCP Service Color 40 EF Green 41 BE Green 42 BE Green 43 BE Green 44 BE Green 45 BE Green 46 EF Green 47 BE Green 48 CS6 Green 49 BE Green 50 BE Green 51 BE Green 52 BE Green 53 BE Green 54 BE Green 55 BE Green 56 CS7 Green 57 BE Green 58 BE Green 59 BE Green 60 BE Green 61 BE Green 62 BE Green 63 BE Green Upstream Mappings Based on the 802.1p Values of VLAN Packets According to the mappings between the 802.1p values of VLAN packets and the internal priorities, the scheduling priority and color of the packets are specified on routers based on the 802.1p values of the packets, which ensures the proper scheduling of the packets. 41

48 4 Traffic Classification Table 4-5 Default mappings between 802.1p values of VLAN packets and service classes 802.1p Service Color 0 BE Green 1 AF1 Green 2 AF2 Green 3 AF3 Green 4 AF4 Green 5 EF Green 6 CS6 Green 7 CS7 Green Upstream Mappings Based on the EXP Values of MPLS Packets According to the mappings between the EXP values contained in MPLS packets and internal priorities, the scheduling priority and color of the packets are specified on routers according to the EXP values contained in the packets, which ensures the proper scheduling of the packets. Table 4-6 Default mappings between EXP values of MPLS packets and service classes EXP Service Color 0 BE Green 1 AF1 Green 2 AF2 Green 3 AF3 Green 4 AF4 Green 5 EF Green 6 CS6 Green 7 CS7 Green Priority Mapping of Outgoing Packets A router searches for the internal scheduling priority and color of packets based on the DSCP values, 802.1p values, or EXP values of the packets on the outbound interface. After performing the internal scheduling, the router adds the priority fields such as DSCP values, 802.1p values, and EXP values to the packets to be sent according to the internal scheduling priority and color of the packets. 42

49 4 Traffic Classification Complex Traffic Classification Complex traffic classification refers to classifying packets based on certain characteristics of the packets and then performing the pre-defined forwarding action on the packets of different classes. Configuring Traffic Classifiers A classifier is a set of defined conditions for classifying packets. Multiple matching rules can be defined in a classifier. The default relationship between these rules is "OR". That is, the corresponding behaviors can be implemented for the packets when the packets match any one of the rules. The relationship between these rules can be set through the parameter operator. The NE5000E supports the following rules for complex traffic classification: Rules based on Layer 3 or Layer 4 information DSCP TCP SYN Flag IP precedence of IP packets Source IP address of IPv6 packets Destination IP address of IPv6 packets IPv6 next header Rules based on Layer 2 information Source MAC address Destination MAC address 802.1p Rules based on MPLS EXP Complex traffic classification can be applied in the inbound and outbound directions of an interface. NOTE Complex traffic classification cannot be configured for the downstream packets on the LPUE. Configuring Traffic Behaviors Traffic classification is performed to provide differentiated services. Therefore, traffic classification is useful only after it is associated with traffic control actions or resource distribution actions. The traffic behaviors are as follows (these behaviors can be combined): Deny/Permit Deny/Permit is the simplest traffic control action. It enables the device to control the network traffic by permitting packets to pass through or denying packets. Traffic policing As one of the traffic behaviors, traffic policing is also called CAR. Through CAR, operators can set the maximum traffic for various services from the network edge and control the usage of network resources, which ensures the QoS of the entire network. Operators sign the service level agreements (SLAs) for cooperation. An SLA contains the parameters such as the Committed Information Rate (CIR), Peak Information Rate (PIR), Committed Burst 43

50 4 Traffic Classification Size (CBS), and Peak Burst Size (PBS) of various service traffic. The device performs such behaviors as Pass, Drop, or Re-mark the priorities of packets for the traffic exceeding the promised limit. Re-mark The re-mark action refers to marking service traffic with classes according to the SLA and results of traffic classification. Currently, the related RFC protocol defines eight types of standard services: EF, AF1 to AF4, CS6, CS7, and BE and confirms the requirements for implementing these services by defining the PHBs of the services, that is, the requirements for processing these services by the device. EF traffic requires short delay, little jitter, and low packet loss ratio, and corresponds to real-time services such as video services, voice services, and video conferences. AF traffic requires shorter delay, low packet loss ratio, and high reliability, and corresponds to services that have high requirements for data reliability such as e-business and enterprise VPNs. BE traffic has no requirement for the information rate and delay, and corresponds to traditional Internet services. The device can re-mark the DSCP values, IP precedence, 802.1p values, or EXP values of related packets and specify the service classes, that is, EF, AF1 to AF4, CS6, CS7, and BE, of packets. Redirect The redirect action indicates that the device does not forward packets according to the original destination addresses of the packets but forwards the packets to a specified next hop or Label Switched Path (LSP). In this manner, policy-based routing is implemented. Currently, the redirect action is valid only for Layer 3 packet forwarding. The device can implement multiple types of the redirect action. IPv4/IPv6 strong redirection If a user specifies the next-hop IP address and outbound interface of a packet, devices do not need to search the FIB table before forwarding the packet. The devices can directly send the packet to the outbound interface specified by the user. The packet can be sent after being encapsulated with the ARP information on the outbound interface. When the outbound interface is Down, the packet is discarded and is not forwarded according to the original destination address. IPv4/IPv6 weak redirection When a user specifies the next-hop IP address of a packet but does not specify the outbound interface of the packet, devices search the FIB table according to the nexthop IP address configured by the user before forwarding the packet. If the path specified by the user is available, the devices forward the packet along the path. If the path specified by the user is unavailable, the devices forward the packet according to the original destination address of the packet. IPv4/IPv6 multiple next hop strong redirection A user can specify multiple next hops and outbound interfaces. A maximum of 16 next hops and a maximum of 16 outbound interface can be specified. Packets are forwarded along an available path selected according to the configured next hops. If the current path is unavailable, packets are forwarded along another available path that is automatically selected. If all the interfaces are invalid, the packets are discarded. IPv4/IPv6 multiple next hop weak redirection A user can specify a maximum of 16 next hops. Packets are forwarded along an available path selected according to the configured next hops. If the current path is unavailable, packets are forwarded along another available path that is automatically selected. If all the interfaces are invalid, the packets are forwarded to the original destination. Redirecting packets to IPv4 L3VPNs 44

51 4 Traffic Classification Traffic from a VPN host can enter other VPNs based on a traffic policy. The 32-bit source IP address, quintuple information, or source network segment can be specified in the policy. Thus, packets with the source address can be redirected to one or more VPNs. Redirecting packets to IPv4 LSP Security Configuring Traffic Policies If the specific LSP does not exist or the LSP is Down, packets will be discarded. Security refers to performing such measures as Unicast Reverse Path Forwarding (URPF), port mirroring, or traffic sampling over packets. Security actions are not QoS measures but can be used together with QoS actions to improve the security of the network. A traffic policy is an integrated QoS policy formed by associating traffic classification with QoS behaviors. A traffic policy can be applied to interfaces, devices, or user-specific service policies, hence applying traffic classification and behaviors defined in the traffic policy. The traffic policy supports two attributes, that is, the shared attribute and the non-shared attribute. The shared attribute indicates that different interfaces on the same interface board use the same traffic policy and share a set of traffic classification and traffic behavior entries. The non-shared attribute indicates that different interfaces on the same interface board use the same traffic policy but use multiple sets of traffic classification and traffic behavior entries generated according to interfaces and VLANs. When two interfaces on an interface board use the same traffic policy, the two interface share a set of rules and behaviors if the attribute of the traffic policy is shared. If CAR is set, the traffic on the two interfaces is limited at the same time. The two interface use two sets of rules and behaviors if the attribute of the traffic policy is nonshared. The rules are the same but the behaviors are different. If CAR is set, the traffic on the two interfaces is limited independently. The device supports the dynamic modification of rules of a traffic policy but does not support the dynamic modification of the shared attribute or the non-shared attribute of a traffic policy. After applying a traffic policy on an interface, you can dynamically add, delete, or change the rules and behaviors of the traffic policy, but you cannot change the shared attribute of the traffic policy. You can change the shared attribute of the traffic policy only after disabling the traffic policy on the interface. 45

52 4 Traffic Classification Process of Complex Traffic Classification Figure 4-3 Process of complex traffic classification Packet forwarding based on the interface, VLAN, or VSI Is complex traffic classification enabled? No Yes Contruct the Key value. Search the table of complex traffic classification. GID SIP DIP Rule key... Rule mask SIP_mask DIP_mask... packet GID SIP DIP... Do the packets match the traffic policy? No Yes Implement the behavior specified in the traffic policy. Perform other processing. Figure 4-3 shows the basic process of implementing complex traffic classification for packets. When the AND operation is performed between the masks of the rules of a traffic policy and the source IP addresses and destination IP addresses of packets, the packets match the policy if the value obtained through the AND operation is the same as the value defined in the rules. Then, the behavior associated with the traffic policy is implemented for the packets. 4.5 Applications Mapping Instances of Simple Traffic Classification Priority mapping of VLAN packets 46

53 /24 HUAWEI NetEngine5000E Core Router 4 Traffic Classification Figure 4-4 Priority mapping of VLAN packets GE1/0/ /24 RouterA GE4/0/ /24 VLAN 10 VLAN Network GE2/0/ /24 VLAN 10 RouterB GE3/0/ /24 As shown in Figure 4-4, Router A and Router B connect to each other through a VLAN. When IP packets sent from Router A enter the VLAN, the priorities of the IP packets are mapped to the priorities of the VLAN frames according to the default mappings in the DiffServ domain. When packets from the VLAN reaches Router B, the priorities of the VLAN packets are mapped according to the priority mappings in the DiffServ domain set on Router B. Simple traffic classification applied to MPLS networks Figure 4-5 Simple traffic classification applied to MPLS networks POS1/0/0 POS2/0/0 POS1/0/0 POS2/0/0 POS1/0/0 RouterA RouterB RouterC POS2/0/0 As shown in Figure 4-5, MPLS neighbor relationships are established between Router A, Router B, and Router C. After reaching Router A, IP packets are forwarded through MPLS from Router A to Router C. After these MPLS packets reach Router C, Router C forwards them as IP packets. The mappings from IP DSCP values to MPLS EXP values are set on POS 1/0/0 of Router A, and the mappings from MPLS EXP values to IP DSCP values are set on POS 1/0/0 of Router C. Simple traffic classification is enabled on the two POS interfaces. In this manner, the DSCP value of the IP traffic can be changed to the EXP value of MPLS traffic on Router A, and the EXP value of MPLS traffic can be changed to the DSCP value of the IP traffic on Router C. 47

54 4 Traffic Classification Example for Complex Traffic Classification Figure 4-6 Complex traffic classification applied to networks As shown in Figure 4-6, assume that the bandwidth purchased by Company A is 200 Mbit/s, and that purchased by Company B is 400 Mbit/s. To ensure the bandwidth, you can configure complex traffic classification on the edge access node. The node can thus differentiate the traffic of Company A from that of Company B based on the IP addresses, and then carry out different traffic policing policies. 48