Advanced Traffic Management (qos) Concepts

319 1056_05F9_c2

1

© 1999, Cisco Systems, Inc.

Advanced Traffic Management and QoS Concepts Session 319

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

2

1

Introduction

• Traffic Management • Applications and Transports • So what Are the Issues for TCP Voice on IP Video (Broadcast and Teleconferencing) 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

3

Let's Talk about Traffic Management • Why it is a concern • What the guiding principles are • What tools are available • What can be accomplished using those tools • What cannot be accomplished 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

4

2

Why Traffic Management Is a Concern

• Needs of certain applications Mail? Web? Transaction processing?

• Opportunities with certain transports

319 1056_05F9_c2

5

© 1999, Cisco Systems, Inc.

Guiding Principles for Traffic Management • We want to achieve

• In a network that

Predictability Reliability Availability

Keeps intelligence at the edges Scales to necessary sizes and bandwidths Minimizes complexity Uses cost-effective technologies

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

6

3

What Tools Are Available for Traffic Management

• Traffic path control • Queue depth management • Queue rate management • Permission to use a link

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

7

How Well Will Traffic Management Do?

• We know we can do this: Management of latency Management of bandwidth

• What cannot be accomplished Creation of bandwidth that otherwise would not be there 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

8

4

Primarily a WAN IP Talk WAN Protocol Breakdown

• IP is the dominant internet protocol • TCP is the dominant data transport 95% of Internet traffic uses TCP

• Voice is a growing market

80% 70% 60% 50% 40% 30% 20% 10% 0%

But beware of hype

IP

1994

• Heterogeneous link layers Source: Gartner 319 1056_05F9_c2

1996

1998E

2000E

2002E

IP SNA IPX Others RFC 1490

Group Study, March 1997 9

© 1999, Cisco Systems, Inc.

Making Networks Predictable The Grail

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

10

5

This Is what You Need to Understand:

• TCP-based applications, voice, and video can be managed well with a little planning

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

11

Parekh and Gallagher’s Paper • INFOCOMM ’93 • One must have at most a predictable amount of traffic in the network • One must have predictable traffic delay in each network element • Given these, end-to-end delay of a host to host message is predictable 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

12

6

Definition of “Predictable”

• Does not mean “Fixed”, “Invariant”, or “Zero”

• Means that it has a Mean value Statistical distribution Upper bound 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

13

Predictable Amount of Traffic in the Network • The source must pace traffic initiation so that standing queues are bounded Queues form when arrival rate exceeds departure rate

• When congestion (too many messages in one queue) sets in: Sources must not increase their rate Ideally, sources decrease their rate 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

14

7

Examples of Source Predictability

• TCP will keep at most a certain amount of traffic in flight We say it is “elastic”—rate is proportional to latency

• Voice will send only and exactly as fast as the coding algorithm permits We say it is “inelastic” 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

15

Predictable Packet Treatment in Routers and Switches • Transit latency must be within limits acceptable to the application • Variation in transit latency must be within limits acceptable to the application • No stream may be locked out apart from administrative policy • Applicable policy must be observed 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

16

8

Examples of Unpredictability

• Queues change rapidly enough that the distribution cannot be described • Discards happen frequently enough that there is effectively no upper bound on delivery time

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

17

Examples of Predictability

• Classes of queues get sufficient service that ultimate arrival is timely and normal “Timely” is an application concept…

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

18

9

Quality of Service Issues in Traffic Management

• Predominantly TCP traffic • Some specific applications • Voice/video traffic

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

19

Managing TCP Traffic Moving Mountains of Data Without Incurring the World Wide Wait 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

20

10

Backbone Traffic Mix Transport Breakout

TCP Applications

Source: MCI/NSF OC-3MON via http://www.nlanr.net, 1998

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

21

TCP Technology Issues

• Single drops communicate from network to sending host “You need to slow down”

• Multiple drops in round trip trigger time-outs “Something bad happened out here” 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

22

11

Behavior of a TCP Sender • Sends as much as credit allows • Starts credit small

N N+ N+ 1 2 N+ 3

Avoid overloading network queues

• Increases credit exponentially To gauge network capability 319 1056_05F9_c2

23

© 1999, Cisco Systems, Inc.

Behavior of a TCP Receiver

• When in receipt of “next message,” schedules an ACK • When in receipt of something else, acknowledges all it can immediately 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

N N+ N+ 1 2 N+ 3

+1 kN Ac 1 + kN Ac +1 kN c A

24

12

Sender Response to ACK • If ACK acknowledges something Update credit and send

• If not, presume it indicates a lost packet Send first unacknowledged message right away Halve current credit Increase linearly to gauge network throughput 319 1056_05F9_c2

N N+ N+ 1 2 N+ 3

+1 kN Ac 1 + kN Ac +1 kN Ac

N+ 1

+4 kN Ac

25

© 1999, Cisco Systems, Inc.

Multiple Drops in TCP • In the event of multiple drops within the same session: Current TCPs wait for time-out Selective acknowledge may work around (but see INFOCOM ’98) New Reno “fast retransmit phase” takes several RTTs to recover 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

N N+ N+1 2 N+ 3 +1 N+ 4 ck N A +1 kN Ac +1 kN Ac N+ 1 +4 kN c A

N+ 4

World Wide Wait!

+5 kN c A 26

13

Remember Parekh and Gallagher

• One must have at most a predictable amount of traffic in the network • One must have predictable traffic delay in each network element • Given these, end-to-end delay of a host to host message is predictable 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

27

How Can We Make TCP in a Network Act Predictably? • Predictable amount of traffic in the network: Well-written TCP implementations manage their rates to the available bandwidth

• Router needs to Provide predictable treatment of packets Queue delay and drop characteristics 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

28

14

Fundamental FIFO Queue Management Technologies • Tail drop Network standard behavior Causes session synchronization when waves of traffic experience correlated drops

• Random Early Detection (RED) Random drops used to desynchronize TCP sessions and control rates 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

29

Session Synchronization • Session synchronization results from synchronized losses • Tail drop from waves of traffic synchronizes losses 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

30

15

Effect of Random Early Detection

Courtesy of Sean Doran, Ebone

RED Enabled

• One day, below 100% throughput Simple FIFO with tail drop

• Starting 10:00 second day, 100% throughput Random Early Detection enabled 319 1056_05F9_c2

31

© 1999, Cisco Systems, Inc.

Was that a Fluke?

Courtesy of Sean Doran, Ebone

RED Enabled

• No, here’s what happened that week… • Session synchronization reduced throughput until RED enabled 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

32

16

FIFO Traffic Timings 400

350

300

Mean Latency Correlates with Maximum Queue Depth

Ns RTT

250

200

150

100

50

0 Elapsed Time Mean RTT

319 1056_05F9_c2

Min RTT

Max RTT

STD DEV

33

© 1999, Cisco Systems, Inc.

RED Traffic Timings 400

350

Additional Capacity to Absorb Bursts

300

Ms RTT

250

200

Mean Latency Correlates with Minimum Drop Threshold

150

100

50

0 Elapsed Time Mean RTT

319 1056_05F9_c2

Min RTT

Max RTT

STD DEV

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

34

17

Therefore—TCP QoS Definition:

• Normally at most one drop per round trip • Mean variation in latency bounded by predictable network

319 1056_05F9_c2

35

© 1999, Cisco Systems, Inc.

TCP Flow Statistics

• >90% of sessions have ten packets each way or less Transaction mode (mail, small web page)

• >80% of all TCP traffic results from <10% of the sessions, in high rate bursts It is these that we worry about managing 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

36

18

An Interesting Common Fallacy about RED: • “RED means you will have more drops” Statement derives from observed statistics

• RED means that you will have Closer to 100% utilization of your line Less average delay per packet

• But queuing theory? As a line approaches 100% utilization, drops will increase, even though served load increases 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

37

TCP Traffic Management Issues

• Applications Often have site-specific policy associated with them Traffic often identifiable by port numbers

• Sites Generally identifiable by address prefix or interface traffic is received on 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

38

19

TCP Bandwidth Policy Questions to Answer

• Particular site or application wants at least a certain bandwidth • Particular site or application wants at most a certain bandwidth • Particular site or application wants to average about a certain bandwidth 319 1056_05F9_c2

39

© 1999, Cisco Systems, Inc.

This Is Where “Classes” Come in • Classes can be for: Voice Important application/site

Classifier

Unimportant application/site Assuring at least a rate

Queues

Interface

Limiting to a rate 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

40

20

Some Class of Traffic Wants at Least a Certain Bandwidth ICU Left UC Me

Right

Managed Link

U Betcha

• Example: Several organizations share cost of link Distribute bandwidth proportional to fiscal responsibility 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

41

Class of Traffic Wants at Most a Certain Bandwidth • Traffic shaping • Similar queuing technology to classbased weighted fair queuing • Rate assigned to Interface or sub-interface Frame Relay circuit ATM virtual channel (in hardware) 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

42

21

Examples of Rate Control • Frame Relay network

• Intranet exposure

Access rate exceeds PVC rate—limit rate to rate of PVC

64 KBPS

T-1

319 1056_05F9_c2

Limit rate of web surfing outside the company

43

© 1999, Cisco Systems, Inc.

Some Class of Traffic Wants to Average a Certain Bandwidth

• Service provider or large enterprise model • Designed for Cost containment Managed response to conflicting demands 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

44

22

Marking TCP Traffic at Edge • A useful technique: • Mark traffic at a network edge with simple classifier • This allows network to Do the right thing without having to fully classify everywhere Use more effective markings 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

45

Serving TCP Traffic with the Assured Service • Presumes service level agreement Flat rate for traffic meeting a rate/burst profile Usage charging for traffic out of profile

• Drop management (weighted RED) All traffic subject to loss Traffic out of profile much more subject to loss Enhances ISP traffic engineering (Good for service provider and consumer) 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

46

23

Structure of Presumed Service Level Agreement

Assured Service 70% 60%

Potentially dropped by WRED at bottleneck Usage pricing of overage 319 1056_05F9_c2

50% 40% 30% 20% 10% 0%

0 10 20 30 40 50 60 70 80 90

Usage

• Up to rate over interval is “in profile” • Traffic within profile gets some guarantees • Traffic out of profile has no guarantees

Time

47

© 1999, Cisco Systems, Inc.

Best Effort Service in Simple IP Networks

Line Congested? Drop at Some Rate!

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

48

24

Assured Service in Simple IP Networks

Line Congested and Packet Out of Profile? Drop at Profile Some Rate!

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

49

Best effort Service in an ATM-Based Network

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

50

25

Assured Service in an ATM-Based Network

319 1056_05F9_c2

51

© 1999, Cisco Systems, Inc.

So, for TCP

• Traffic can be contained to a rate in a manner consistent with good quality of service • Traffic can be managed well with a little foresight and planning

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

52

26

Convergence with Voice Networks “It’s about Internet Telephony!”

319 1056_05F9_c2

53

© 1999, Cisco Systems, Inc.

Again, the Premise:

• TCP-based applications, voice, voice and video can be managed well with a little planning

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

54

27

Changing Corporate Network Application Predominance Numbers in Percent 2%

100

2%

7%

Multimedia Dynamic WWW Static WWW FTP and Telnet Email and News Other

13%

7%

80

27% 28% 27%

60 15% 39%

40 20

39%

17% 8%

0

1996

12%

17%

17%

8%

14%

1998

2000

Source: The Yankee Group, 1996

319 1056_05F9_c2

55

© 1999, Cisco Systems, Inc.

Growth of IP Traffic • Email • Information search/access • Subscription services/“Push” • Conferencing/ multimedia

Rel. Bit Volume

Traffic Projections for Voice and Data

250

Data (IP)

200 150

Circuit Switched Voice

100

• Video/imaging “From 2000 on, 80% of Service Provider Profits Will Be Derived from IP-Based Services.” Source: CIMI Corp.

50

1997

1998

1999

2000

2001

Source: Multiple IXC Projections

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

56

28

High End IP Transport Alternatives IP or Voice over ATM

B-ISDN

IP over SONET/SDH

IP over Optical

Multiplexing, Protection and Management at Every Layer

IP or Voice ATM

IP or Voice

IP

Voice

SONET/SDH

ATM

SONET/SDH

IP

Optical

Optical

Optical

Optical

Lower Cost, Complexity and Overhead 319 1056_05F9_c2

57

© 1999, Cisco Systems, Inc.

H.323 Voice/Video • Voice Constant bit rate when sending Relatively small messages (44-170 bytes)

• Video Generally high variable bit rate Controlled by codec efficiency on picture Message size is generally the MTU 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

58

29

Video: Traffic Pattern

Key Frame

Key Frame

Delta Frames

319 1056_05F9_c2

59

© 1999, Cisco Systems, Inc.

Video: Effect of Delay

Key Frame

Key Frame

Delta Frames

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

60

30

Video: Playback Point Transmission Time

Typical Delivery Playback Point

Preferred Delivery Interval Application Buffers Data to Ensure Consistency

Unless it’s Too Late…

Distribution of Deliveries in Time 319 1056_05F9_c2

61

© 1999, Cisco Systems, Inc.

Synchronization of Voice and Video

• McGurk effect: voice can sound garbled to human ear when not synchronized with video • Therefore, we have to synchronize these

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

62

31

QoS Definition for Voice:

• Low loss rate • Low absolute delay in two-way situations Broadcast voice doesn’t have this problem…

• Low variation in delay 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

63

Key Issue for Voice QoS:

• Silent periods must not be randomly inserted or removed so as to make other sounds unintelligible • End to end delay must be comprehended by human listener

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

64

32

QoS Definition for Video:

• Low loss rate • Low absolute delay in two-way situations • Low variation in delay

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

65

Key Issues for Video:

• All packets that comprise a video frame must arrive during the same frame interval OK if it’s the last millisecond of that interval…

• Audio and video must be synchronized when shown to user 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

66

33

How Can We Make Internet Voice Act Predictably?

• Predictable amount of traffic in the network • Predictable treatment of packets in routers and switches • Planning to support these aspects results in a predictable network 319 1056_05F9_c2

67

© 1999, Cisco Systems, Inc.

Voice/Video Traffic Management Issues

• The fundamental problems with Voice/video traffic are It doesn’t slow down in response to delay or loss It requires minimal variation in delay

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

68

34

Predictable Amount of Traffic in the Network • The implication is that we have to control used capacity Capacity that individual calls consume “If you experience poor quality, use a more compact encoding or a lower frame rate” Capacity that total call volume can consume “If there isn’t capacity, refuse new calls” 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

69

Predictable Treatment of Packets in Routers and Switches

• We have to place voice in queues that give it high priority Maintain tight delay budgets Application of class-based WFQ

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

70

35

Planning for a Predictable Network • Enable CB—WFQ on all relevant links Configure voice queue with more bandwidth than traffic will need, or For low bandwidth, priority queue [12.0(6)T]

• Low speed links should use Link Fragmentation or FRF.12 RTP compression for voice

• Enable RSVP call negotiation “Refuse excess calls” 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

71

FRF.12, and Link Fragmentation and Interleaving

• Premise: Reducing voice packet size reduces session requirements on network So compress out IP, UDP, and RTP headers as much as possible

• Limits jitter on lower bandwidth links 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

72

36

Resource Reservation

• Current deployment • Current extensions • Extensions being developed

319 1056_05F9_c2

73

© 1999, Cisco Systems, Inc.

Current Deployment • RSVP version 1 Call control for individual sessions Deployed Cisco 11.2 Microsoft Windows ’98 (service pack) Microsoft Windows NT 2000

• Appropriate to edge networks 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

74

37

Current Extensions

• Policy management via COPS • LAN management via subnet bandwidth manager

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

75

Policy Management Via COPS • Local or central policy server can authorize decisions • Local policy: Simple policies

• Central policy server: Certificates, Complex policies 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

76

38

LAN Management Via Subnet Bandwidth Manager

• Subnet bandwidth manager is RSVP in a switch • Controls aggregate reservations on a LAN

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

77

Extensions Being Developed

• Rapid deployment of calls • Aggregate classification in edge networks • Aggregate classification and admission in service provider networks 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

78

39

Rapid Deployment of Calls

• Problem: need acknowledged reservation installation • Solution: acknowledge it…

319 1056_05F9_c2

79

© 1999, Cisco Systems, Inc.

Aggregate Classification in Edge Networks

PSTN

• Use differentiated services code points to identify traffic Rather than specific flows

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

PSTN

80

40

Aggregate Classification in Edge Networks • Reservation requested by host in the usual way (RFC 2205) • Flow classification and policing at first hop router • Flow admission along end to end path • Aggregate classification and policing at subsequent routers 319 1056_05F9_c2

81

© 1999, Cisco Systems, Inc.

Aggregate Classification and Admission Across Service Provider Networks • Voice/video calls Placed across aggregation domain boundary

• Aggregate reservations Placed from ingress to egress for DSCP used Use expedited forwarding service Limited rate of change 319 1056_05F9_c2

• Why? Otherwise, you don’t know that bandwidth exists on a path

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

82

41

Solving Voice/Video Issues Using the Expedited Forwarding Service • Rate control Application at source Reservation in network

• Jitter control WFQ’s priority queue (low speed) Statistically empty queue (CB-WFQ) 319 1056_05F9_c2

83

© 1999, Cisco Systems, Inc.

The Implications for Voice and Video • We can control call volume And therefore traffic volume

• We can scalably prioritize traffic in the system And therefore deliver on latency issues

• So, voice and video can be managed well with a little planning 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

84

42

Traffic Path Control What if IP Routing Isn’t Quite Good Enough for Your Traffic? 319 1056_05F9_c2

85

© 1999, Cisco Systems, Inc.

Traffic Engineering

• Historical approaches Load sharing Routing metrics

• A new one Label switching 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

86

43

Load Sharing

• Multipath routing Equal and unequal cost

• Multilink PPP

319 1056_05F9_c2

87

© 1999, Cisco Systems, Inc.

Routing

• Administrative metrics Designed to move traffic to statistically low volume links

• Load sensitive metrics Designed to move data away from congested links Tendency towards oscillation 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

88

44

Utility of These:

• While they basically work, they are Not deterministic, and Tend to be hard to predict

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

89

Multi-Protocol Label Switching

• MPLS traffic engineering VPNs and general engineering

• MPLS routing for resource reservation In the direction of QoS routing 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

90

45

Principles of Label Switching • Labeled paths: Multiple enumerated point to point relationships between pairs of routers Sets of pair-wise relationships create a labeled tunnel

• Conceptually similar to ATM VCs or Frame Relay DLCs, but Interface independent Used to model network layer constructs Variable length packets 319 1056_05F9_c2

91

© 1999, Cisco Systems, Inc.

Network Layer Constructs… • Types of traffic streams Destination routes Source-destination routes AS pairs BGP community pairs

Notice: Two Labels on One Interface, Distinguishing Routes

• Tunnels can create Any routing that meets engineering needs 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

92

46

Using Labeled Tunnels to Create Virtual Private Networks • Imagine edge network with private address space • Stretch labeled tunnels across the network • Now, do it again • Disjoint networks Same address space Separate routing 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

93

MPLS Traffic Engineering

• Same technology can drag specific routes around Several less-used paths vs a few denser paths…

• Initially seen as off-line engineering • Can use either LDP or RSVP to install routes 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

94

47

CoS in MPLS Networks • Class of Service Roughly similar to diff-serv code point Eight values, not sixty-four

• Implements similar drop/delay management within labeled tunnels • Therefore, MPLS networks have fundamental TCP QoS support 319 1056_05F9_c2

95

© 1999, Cisco Systems, Inc.

The Obvious Hole…

• Wouldn’t it be nice if engineered labeled tunnels could Have specific bandwidths guaranteed? Recover from network events quickly and automatically using reasonable if not optimal routes? 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

96

48

MPLS Routing for Resource Reservation • Use OSPF/IS-IS to distribute bandwidth availability information • Edge router does SPF calculation when needed • RSVP used to install labeled tunnel while checking for race events • CoS field used to identify traffic for queued rate support 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

97

Automated Reinstallation of Labeled Tunnels • RSVP tears down affected tunnels • Edge devices recalculate routes • RSVP used to re-install tunnels • Bandwidth checks result in retry 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

98

49

Q.E.D. MPLS

• Traffic engineering for network layer traffic can be managed well with a little planning

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

99

So, What Are You to Do about It? Here the Rubber Meets the Road

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

100

50

Plan Your Network for Predictability

• Network engineering • Assured forwarding service TCP

• Expedited forwarding service Voice, implies some form of admission 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

101

Network Engineering

• Capacity engineering Engineered IP routes?

• May involve traffic engineering Labeled tunnels?

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

102

51

Assured Forwarding Service • Designed for TCP Classes control rates for SLAs Drop controls trace effects back to sources

• Implement using Committed access rate, Weighted Random Early Detection, 319 1056_05F9_c2

Class-based weighted fair queuing © 1999, Cisco Systems, Inc.

103

Expedited Forwarding Service

• Appropriate to voice/video • Requires Under-subscribed traffic classes Reservation of bandwidth Policing 319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

104

52

Guiding Principles for Predictability

• One must have at most a predictable amount of traffic in the network • One must have predictable traffic delay in each network element • Given these, end-to-end delay of a host to host message is predictable 319 1056_05F9_c2

105

© 1999, Cisco Systems, Inc.

In Your Network…

• TCP-based applications, voice, and video—and your bandwidth—can be managed well with a little planning

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

106

53

Please Complete Your Evaluation Form Session 319

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

107

319 1056_05F9_c2

© 1999, Cisco Systems, Inc.

108

Copyright © 1998, Cisco Systems, Inc. All rights reserved. Printed in USA. Presentation_ID.scr

54