Tài liệu IP over ATM doc

IP QoS IP over ATM-7 IP QoS and ATM IP QoS and ATM • Routers can be interconnected over an ATM backbone using different ATM services: – UBR – congestion management is virtually impossibl

Trang 1

10

IP over ATM

Overview

This module focuses on IP QoS mechanisms that can be used on ATM interfaces

It includes the following topics:

n Introduction to IP over ATM

Upon completion of this module, you will be able to perform the following tasks:

n List the requirements of IP QoS in combination with ATM QoS

n Describe the hardware and software requirements for advanced IP QoS mechanisms on ATM interfaces

Trang 2

10-2 IP QoS IP over ATM Copyright  2001, Cisco Systems, Inc

Introduction to IP over ATM

Objectives

Upon completion of this lesson, you will be able to perform the following tasks:

n Describe the QoS-related problems when using ATM networks

n Describe the hardware and software requirements for advanced IP QoS mechanisms on ATM interfaces

n Describe per-VC queuing

Trang 3

IP vs ATM Technology comparison

service parameters

ATM

• Connection oriented

• Per-connection (virtual circuit) QoS

• Large number of QoS traffic classes (CBR, VBR, UBR, ABR)

• Rich traffic parameters (PCR, MCR, SCR ) specified for each VC

The Internet Protocol (IP) is a routed protocol that is used to transmit data in packets It uses the best-effort delivery for individual packets without any flow control Transmission Control Protocol (TCP) is used with IP to provide a connection-oriented service

Asynchronous Transfer Mode (ATM), on the other hand, provides connections between endpoints in the ATM network The connections are called virtual circuits (VCs)

IP’s default best effort service can be supplemented by differentiated quality of service based on IP precedence or DSCP marking A QoS solution using IP precedence is limited to 8 classes, 2 of which are reserved and 1 should be used for the default best-effort class A QoS solution using DSCP scales up to 64 classes

ATM provides a wider range of services:

n Constant Bit Rate (CBR) is useful for delay-sensitive applications such as voice This service provides bandwidth and delay guarantees

n Variable Bit Rate—Real Time (VBR-RT) is useful for burstier delay-sensitive

Trang 4

IP’s IP precedence or DSCP are only used to mark packets They do not include any service parameters Servic e parameters depend on the QoS mechanism being deployed

ATM’s services also include various per-connection service parameters, such as:

n Sustained Cell Rate (SCR) for CBR, VBR and ABR services

n Minimum Cell Rate (MIR) for ABR

n Peak Cell Rate (PCR) for VBR, ABR and UBR services

n Maximum Burst Size (MBS) Both IP and ATM can implement Quality of Service (QoS) The decision on which technology to use for quality of service should be based on a number of factors, such as:

n Availability of ATM

n Interaction between ATM and IP

n Scalability options of the technology

n Performance limitations This module introduces the possibilities of combining IP QoS with ATM

Trang 5

Integrating IP and ATM Integrating IP and ATM

• Overlay model (ATM forum)

–ATM VC’s are manually established between pairs

of devices

–IP packets are sent across these VC’s

–ATM switches are not IP aware

There are two main approaches to integration of IP with/over ATM:

n The traditional way (overlay model) is to use individual permanent virtual

circuits (PVC) to establish point-to-point adjacencies between IP routers IP routing protocols are used to provide reachability across a network of ATM connections ATM has no knowledge of IP and cannot use IP information to optimize its links

n The newer approach (MPLS) is to make ATM switches IP aware ATM

switches run an IP routing protocol to establish virtual circuits

This module focuses on the QoS available with traditional permanent and switched virtual circuits (PVCs and SVCs)

The IP QoS- IP over MPLS module discusses QoS possibilities when using the peer model

Trang 6

IP QoS and ATM

• Routers can be interconnected over an ATM backbone using different ATM services:

– UBR – congestion management is virtually impossible because routers are allowed to transmit packets at line speed

– VBR – congestion management is easier, but it requires conservative setting of transmit rates

– CBR – similar to VBR from IP perspective

– ABR – pushes congestion back to the source, requires dynamic adjustment to available bandwidth

Achieving good quality of service for IP classes greatly depends on the type of ATM network and services used

n Using UBR, prevents routers from detecting congestion in the network It is therefore difficult to manage congestion based on IP precedence or DSCP The reason for this is because all packet drops happen on the congested link somewhere in the ATM network

n VBR makes it easier to push congestion back to the source where it can be managed by routers

n CBR is typically used for non-bursty delay sensitive traffic It is therefore more important to prevent congestion by correctly provisioning the class that is using CBR

n ABR is a good solution where bandwidth can be utilized to the maximum without having many drops in the ATM network

Trang 7

UBR Virtual Circuits

• Solution:

– Set CLP on the router based on IP information to minimize the effect of cell drops

No congestion Router allowed to send at full speed

Congestion

Random CLP marking

Unintelligent drops based on CLP

A solution using UBR can be improved in terms of IP QoS, by marking less important packets with the CLP bit for congestion control In case of congestion, the ATM switches will drop the less important packets to give more bandwidth for the higher-priority packets

The ATM FORUM also calls the UBR service category a “best effort” service, which requires neither tightly constrained delay nor delay variation In fact, UBR provides no specific quality of service or guarantee throughput whatsoever This traffic is therefore “at risk” since the network provides no performance guarantees for UBR traffic The Internet and Local Area Networks are examples of this type

of “best effort” delivery performance Examples of this are LAN emulation (LANE), IP over ATM, and non-mission-critical traffic

This solution is fairly limited, since it allows for only two classes on the IP layer where congestion should be managed

Trang 8

VBR Virtual Circuits

• Solution:

– Set CLP on the router based on IP information

– Use available IP QoS mechanisms to manage congestion at the source

Router is sending

at configured rate

Congestion is possible

Unintelligent random drops Congestion!

A solution using VBR is better at providing feedback to routers sending cells into the ATM network Congestion will occur on a router’s virtual circuit, where it can

be managed by using the QoS mechanisms available in the Cisco IOS software CLP marking can be used for less-important packets or for those packets above the Sustained Cell Rate (SCR) to improve the chances for higher-priority packets when congestion occurs in the ATM network

The rt-VBR service category supports time-sensitive applications, which also requires constrained delay and delay variation requirements, but which transmit at

a time varying rate constrained to a PCR, SCR, and MBS define a traffic contract

in terms of the worst-case source traffic pattern for which the network guarantees

a specified QOS Examples of such bursty, delay-variation-sensitive sources are voice and variable -bit-rate video

The nrt-VBR service category supports applications that have no constraints on delay and delay variations, but which still have variable -rate, bursty traffic characteristics This class of application expects a low Cell Loss Ratio (CLR) The traffic contract is the same as that for rt-VBR Applications include packet data transfers, terminal sessions, and file transfers Networks may statistically multiplex these VBR sources effectively

Trang 9

CBR and ABR Virtual Circuits

The CBR service category supports real-time applications requiring a fixed amount of capacity defined by the PCR CBR supports tightly constrained variations in delay Example applications are voice, constant-bit-rate video, and Circuit Emulation Services (CES) Normally, networks must allocate the peak rate

to these types of source

The ABR service category works in cooperation with sources that can change their transmission rate in response to rate-based network feedback used in the context of closed-loop flow control The aim of ABR service is to dynamically provide access to capacity currently not in use by other service categories to users who can adjust their transmission rate in response to feedback In exchange for this cooperation by the user, the network provides a service with very low loss Applications specify a maximum transmit-rate (PCR_ and the minimum required rate, called the Minimum Cell Rate (MCR) ABR service does not provide bounded delay variation; hence real-time applications are for ABR are LAN

Trang 10

Congestion Management in ATM

Networks

Congestion Management in ATM

Networks

• Congestion management on routers should

be performed on a per-VC basis

• Design options:

– Make sure there is no congestion in the ATM network (ABR, CBR, VBR) and use IP QoS mechanisms at the source ( CB-WFQ , WRED )

– Mark less important packets with the CLP bit in case there is congestion in the ATM network ( CB- Policing , CB-Marking )

– Use multiple parallel (per-CoS) virtual circuits with ATM QoS ( VC Bundling )

This module discusses three different approaches to designing QoS in IP networks

Trang 11

Per-VC Queuing

• Per-VC queuing is required in order to handle congestion on per-VC basis

• Per-VC queuing prevents head-of-line blocking by slow virtual circuits

ATM Port Adapter

Per-VC queuing with per-VC congestion management

One of the most important parts of implementing QoS is to make ATM virtual circuits appear as physical interfaces on routers; that is, each VC must have its

own queue (per-VC queuing) Per-VC queuing prevents one congested VC from slowing down other VCs (head-of-line blocking)

Per-VC queuing can then be supplemented by various IP QoS mechanisms, such as:

Trang 12

Answer the following questions:

1 What are the main differences between IP and ATM?

2 Which QoS services does ATM support?

3 How should congestion be handled when an ATM backbone is used?

4 Why is per-VC queuing so important?

Trang 13

Per-VC WRED

Objectives

Upon completion of this lesson, you will be able to perform the following tasks:

n Describe per-VC WRED

n Configure per-VC WRED

n Monitor and troubleshoot per-VC WRED

Trang 14

VC, resulting in differentiated IP services

A simple addition to best-effort service on ATM interfaces is Weighted Random Early Detection (WRED) WRED is most efficient when the majority of the traffic

is TCP (TCP reacts to random drops and slows down the transmission rate) With other protocols, packet sources may not respond or may resend dropped packets at the same rate Thus, dropping packets does not decrease congestion WRED treats non-IP traffic as precedence 0, the lowest precedence Therefore, non-IP traffic is more likely to be dropped than IP traffic

UBR would probably result in congestion somewhere in the ATM network, thus preventing any intelligent congestion management on the IP layer

Any other ATM service (CBR, VBR or ABR) will push congestion back to the source where WRED can be used to drop packets based on the IP precedence or DSCP value

Trang 15

Per-VC WRED : Intelligent

Per-VC queuing requires an Enhanced ATM Port Adapter that support up to 4096 cell queues Each virtual circuit is assigned a queue and the ATM scheduler forwards cells according to the ATM service and shaping parameters

The router (or VIP on Cisco 7x00 series routers) also assigns one queue per virtual circuit

Cell departure is shaped if ABR, VBR or CBR services are used, thus causing congestion in the frame queue if packet arrival is greater than the shaping rate in ATM Per-VC WRED can be used to manage congestion within individual queues (classes)

Trang 16

Configuring Per-VC WRED

• The following configuration steps are needed

to enable per-VC WRED:

– Create a Random-Detect-Group template with a WRED profile

– Apply the WRED template to an ATM interface or

to individual ATM VCs

– Verify and monitor the operation of per-VC WRED

Applying WRED to individual VCs is slightly different than applying WRED to interfaces A Random Detect Group must be created if non-default WRED profiles need to be used on VCs Standard WRED parameters (per-precedence minimum threshold, maximum threshold and maximum drop probability) are set in the random-detect-group configuration mode

Trang 17

random -detect -group name random -detect-group name

Router(config)#

• Creates a WRED template

Create and configure RED-group

exponential-weighting-constant exp exponential -weighting -constant exp

Router(cfg -red -group)#

• Defines WRED weighting constant

• Default: 9

precedence IP-prec min -threshold max-threshold prob -denominator precedence IP-prec min-threshold max-threshold prob-denominator

Router(cfg -red -group)#

• Defines RED profile for specified precedence

• Default: as with per-interface WRED

The random-detect-group global configuration command creates a WRED

profile and enters the red-group configuration mode WRED per-precedence profiles are configured in the red-group configuration mode, using similar commands as with per-interface WRED, except the commands are not preceded

by the random-detect keyword

Any class (IP precedence) can be configured with a RED profile different from

the default by using the precedence command in the red-group configuration

mode:

n Minimum threshold—When the average queue depth is above the minimum

threshold, RED starts dropping packets The rate of packet drop increases linearly as the average queue size increases, until the average queue size reaches the maximum threshold

n Maximum threshold—When the average queue size is above the maximum

threshold, all packets are dropped If the difference between the maximum threshold and the minimum threshold is too small, many packets might be dropped at once, resulting in global synchronization

n Mark probability denominator—This is the fraction of packets dropped

Trang 18

An exponential weighting constant N influences the calculation by weighing the two terms It therefore influences how the average queue size follows the current queue size, in the following way:

n A low value of N makes the current queue size more significant in the new average size calculation, therefore allowing larger bursts

n A high value of N makes the previous average queue size more significant in the new average size calculation, so that bursts influence the new value to a smaller degree

The default value is 9 and should suffice for most scenarios, except perhaps those involving extremely high-speed interfaces (such as OC12), where it can be increased slightly (to about 12) to allow more bursts

Note The default WRED parameter values are based on the best available data

Cisco recommends that you do not change the parameters from their default values unless you have determined that your applications will benefit from the changed values

Trang 19

Apply WRED group to

• Default: no WRED is used on the ATM PVC

The last step in the configuration of per-VC WRED is to attach a

random-detect-group to a virtual circuit The random-detect command is used in the VC

configuration mode to enable WRED If no random-detect-group is specified WRED will use the default WRED profiles

Trang 20

show queueing random-detect [interface intf [vc vpi vci ]]

show queueing random -detect [interface intf [vc vpi vci ]]

The show queuing random-detect command display WRED parameters and

statistics for a specific interface or virtual circuit There is only a single queue into which packets from all IP precedences are placed after dropping has taken place

The show queuing interface command displays per-VC queue parameters and

statistics The “Queuing strategy” reported by the command lists “random early detection (RED)” as the queuing mechanism The default minimum thresholds are spaced evenly between half and the entire maximum threshold Thresholds are specified in terms of packet count

Trang 21

WRED Case Study

• WRED is applied to a ATM PVCs in a network with the following IP precedence definitions

IP prec Meaning

0 High-loss best-effort traffic

1 Low-loss best-effort traffic

2 Premium traffic outside of the contract

3 Premium traffic in the contract

4 Unused

5 Voice-over-IP

6 Routing protocol traffic

7 Routing protocol traffic

• WRED queue length is 100 packets for PVCs with SCR > 10 Mbps and 40 packets for slower PVCs

The case study shows a QoS design where packets are classified into three user classes:

n Best-effort class

n Premium class

n Voice class The Best-effort and Premium classes use two IP precedence values to mark high-drop (out-of-contract) traffic and low-drop (within contract) traffic

IP precedence values 6 and 7 are reserved for control messages (for example, routing protocols) and should not be used for user traffic

The design lists these two additional requirements:

n Virtual circuits faster than 10Mbps should have queues that can hold up to 100 packets

n Slower virtual circuits can store up to 40 packets in the queue All virtual circuits should manage congestion by using WRED

Trang 22

Case Study WRED Profile

Average Queue Size

High drop Best-effort and Premium packets start being dropped when the average queue size reaches 10 or 15 respectively (25 or 37 on fast VCs) If the queue still grows the low-drop Best-effort packets start being dropped when the queue size reaches 20 (50 on fast VCs) High drop packets, of course, are more aggressively dropped than low-drop packets

Control packets, VoIP packets and packets of RSVP flows are only dropped in extreme situations when the average queue size is close to the maximum (40 for slow VCs and 100 for fast VCs)

Trang 23

random -detect -group slow-wred-profile precedence 0 10 25 10

precedence 1 20 40 10 precedence 2 15 25 10 precedence 3 25 40 10 precedence 4 1 10 10 precedence 5 35 40 10 precedence 6 30 40 10 precedence 7 30 40 10

The figure shows the configuration of WRED profiles used for slow VCs

Trang 24

random -detect -group fast-wred-profile precedence 0 25 62 10

precedence 1 50 100 10 precedence 2 37 62 10 precedence 3 62 100 10 precedence 5 87 100 10 precedence 4 1 10 10 precedence 6 75 100 10 precedence 7 75 100 10

The figure shows the configuration of WRED profiles used for fast VCs

Note This configuration simply uses scaled thresholds to support up to 100 packets

in the queue

Trang 25

! interface ATM11/0/0.100 point -to-point

ip address 17.1.1.1 255.255.255.252 atm pvc 100 0 100 aal5snap 17000 34000 10 inarp random-detect fast-wred-profile

! interface ATM11/0/0.101 point -to-point

ip address 17.1.1.5 255.255.255.252 atm pvc 101 5 101 aal5snap 2000 4000 10 inarp random-detect slow-wred-profile

interface ATM11/0/0

ip address 17.1.0.1 255.255.255.0 atm pvc 50 0 50 aal5snap 25000 50000 10 inarp random -detect fast -wred -profile

! interface ATM11/0/0.100 point-to-point

ip address 17.1.1.1 255.255.255.252 atm pvc 100 0 100 aal5snap 17000 34000 10 inarp random -detect fast -wred -profile

! interface ATM11/0/0.101 point-to-point

ip address 17.1.1.5 255.255.255.252 atm pvc 101 5 101 aal5snap 2000 4000 10 inarp random -detect slow -wred -profile

The figure shows the configuration of three virtual circuits Two are using the WRED profile for fast VCs and the third is using the WRED profile for slow VCs

Trang 26

Summary

Weighted Random Early Detection (WRED) is one of the IP QoS mechanisms that can be applied to individual virtual circuits

A Random Detect Group is used to configure a WRED profile that is attached to

individual VCs using the random-detect command in the VC configuration mode

Review Questions

Answer the following questions:

1 What are the benefits of per-VC WRED?

2 What are the configuration steps needed to enable per-VC WRED?

Trang 28

An IP precedence value or a range of IP precedence values are mapped to one virtual circuit Non-contiguous IP precedence ranges are not supported

Trang 29

VC Bundling Case Study

Control (routing) Voice VPN traffic Premium Internet Best-effort Internet

The figure illustrates a case study where there are four user classes and one class for control traffic

Routers perform classification based on IP precedence values:

n IP precedence 6 and 7 traffic is forwarded through the Control VC

n IP precedence 5 traffic is forwarded through the Voice VC

n IP precedence 4 traffic is forwarded through the VPN VC

n IP precedence 2 and 3 traffic is forwarded through the Premium VC

n IP precedence 0 and 1 traffic is forwarded through the Best-effort VC

Trang 30

Control (routing) Voice VPN traffic Premium Internet Best-effort Internet

VC Bundling Routing Adjacency

Whole bundle is treated as one routing adjacency and is covered by a single ATM map

Routing protocol packets are exchanged over control VC as they are sent with IP precedence 6

Each VC has its own HW queue in the router, managed with WRED

All five classes are separated in the ATM network and receive different quality of service Routers have to perform per-VC queuing to prevent head-of-line blocking All five virtual circuits, though, appear as one single point-to-point link on the IP layer

Trang 31

– less statistical multiplexing,

– more complex provisioning/engineering

VC Bundling provides an efficient utilization of QoS capabilities provided by

ATM IP classes are effectively isolated by being transported over different virtual circuits The drawbacks of this approach are:

n Less statistical multiplexing One class cannot use another class’s bandwidth (unless ABR is used)

n More complex provisioning Each IP adjacency, which normally requires one point-to-point virtual circuit, now requires multiple virtual circuits of different types and QoS

As much as IP QoS is simplified to classification and marking using IP precedence, ATM QoS is more complex because there are up to eight times more virtual circuits to be configured

Trang 32

Most Layer-2 technologies include some type of link management Keepalive frames are typically used as a last resort to determine if end-to-end connectivity works For example:

n HDLC and PPP use link-level keepalive frames to determine if the link is operational

n Frame Relay uses keepalive frames to determine if the link between a router and a switch is operational Frame Relay can also have end-to-end keepalive messages to determine if the virtual circuit is operational

n ATM uses two types of Operation Administration and Maintenance (OAM) cells to determine if link-level and end-to-end connectivity works

VC bundling is more complex since there are multiple parallel virtual circuits used for one single IP adjacency

The question is: what should happen if only one VC goes down?

Trang 33

Two ways of handling loss of VC in the bundle:

• The whole bundle is declared down

• Traffic from the lost VC is bumped onto another VC

• IP routing model does not allow the traffic for a single precedence value to be rerouted over another path

There are two possible ways of handling lost VCs:

n All VCs are declared inactive

n The traffic for the lost VC is rerouted onto another VC within the same bundle

IP forwarding decisions are based solely on the destination address and cannot reroute packets based on their IP precedence values

Trang 34

Keep All Graphics Inside This Box

VC Bumping

• VC bumping = possibility for a traffic mapped to VC X to be forwarded onto another VC Y, in case of failure of X

or explicit rules

protected

n VC bumping is one approach to handling lost VCs If one of the VCs goes

down the traffic from that VC is forwarded through another VC in the same bundle

n Implicit bumping is the default behavior where packets are forwarded

through the first available VC of a lower IP precedence value

n Explicit bumping requires manual configuration where the IP precedence of

a backup VC is set

Trang 35

The figure illustrates how routers automatically reroute Premium traffic to the first

VC with a lower IP precedence value (Best-effort in the example)

Trang 36

Rejects bumping

Some virtual circuits can be configured to reject bumped traffic

The figure illustrates how the Voice VC rejects bumped traffic (mixing delay

sensitive, well-provisioned traffic with other types of packets is not desired and should be prevented) Implicit bumping searches down the “ladder” for the first available VC (it has to be operational and accept bumped traffic)

Trang 37

Voice

VPN traffic

Premium Internet Best-effort Internet

Another approach is to explicitly set the backup VC

The Control VC in the figure was configured to use the Best-effort VC as backup

Trang 38

Bundle Failure Scenarios

• Problem: under default settings, the whole bundle is declared down if the lowest-precedence VC is lost

• Solution: be sure that the lowest-precedence VC is always bumped via explicit bumping rule

Precedence 0 traffic cannot be implicitly bumped

When a bundle is declared down, no traffic is forwarded out of the bundle, even if some VCs are still up

In this figure the VC used for IP precedence 0 does not have a lower-precedence

VC to be used as backup It is recommended to use explicit bumping for this VC

Trang 39

Some VCs have special QoS requirements that cannot be accommodated by any other VC

The Voice VC in the figure cannot be bumped to any other VC because the voice quality would no longer meet the requirements It is better to declare the entire bundle down and let the IP routing protocol find another path where guarantees can be met Classes that under no circumstances should be mixed with other

classes should reject bumped traffic (if a higher-precedence VC fails) and be protected (if their VC fails)

Trang 40

One group of VCs can be protected in a way where the bundle is declared down but only if all of the VCs in the group fail

Tiêu đề	IP over ATM
Trường học	Cisco Systems, Inc.
Chuyên ngành	Computer Networks
Thể loại	Document
Năm xuất bản	2001

Định dạng
Số trang	85
Dung lượng	4,19 MB