1. 1/57 Hansung University QoS over IP - Overview & Big Picture - 2011. 9 김 영 웅 yukim@hansung.ac.kr

2. 2/57 Hansung University  Overview  QoS Technologies for IP Networks  QoS Monitoring & Measurement Contents

3. 3/57 Hansung University  Introduction  QoS Requirements  QoS SchemeOverview

4. 4/57 Hansung University 서비스 사용자의 만족 정도를 결정하는 서비스 성능의 종합적인 효과 ITU-T 서비스를 사용하는 사용자의 형태, 특성, 요구 수준에 따른 사용자의 요구에 적응할 수 있는 네트웍의 성능지표 IETF 특정 응용 서비스 사용에 있어서 만족하는 정도 사용자 관점 어느 수준 이상의 서비스 품질을 보장하도록 제어가 가능한 자원요소. ( 측정 가능하고, 정량적으로 표현될 수 있는 값 ) 망 제공자 관점 QoS : What is it?

5. 5/57 Hansung University  Connection Level  Availability  Failure Notification Time  Failure Recovery Time  Packet Level  Throughput  Delay  Delay Variation (Jitter)  Loss QoS Parameters

6. 6/57 Hansung University VoIP - QoS Classes VoIP - QoS Classes QoS Class4 (Best)3 (High)2 (Medium)1 (Low) End-to-end Delay< 150 ms< 250 ms< 350 ms< 450 ms MOS4.2 ~ 5.03.8 ~ 4.22.9 ~ 3.82.0 ~ 2.9 Call Setup Time Direct IP addressing < 1.5s< 4s< 7s E.164 to IP addr < 2s< 5s< 10s E.164 to IP via clearing house or roaming < 3s< 8s< 15s

7. 7/57 Hansung University VoIP - Voice Quality (cont.)  ITU-T MOS  Mean Opinion Scores 음성품질의 기준을 사용자들의 평균 의견을 점수화한 값을 사용 사용자들이 느끼는 품질의 평균 값을 단지 만족감의 수치로 표현하기 때문에 주관적이 평가 값 RatingQuality LevelDistortion 5ExcellentImperceptible 4GoodJust perceptible but not annoying 3FairPerceptible and slightly annoying 2PoorAnnoying but not objectionable 1UnsatisfactoryVery annoying and objectionable  High Quality : MOS≥4.0 (Toll quality)  Communication Quality : MOS≥3.5

8. 8/57 Hansung University VoIP - Connection Quality  ETSI TS 101 024-9 ParameterObjectives Call Setup Failure Probability< 3 x 10 -5 Call Premature Disconnect Probability< 1.5 x 10 -5 Call Release Failure ProbabilityFFS Call Setup Delay 7500/8450 ms Mean/95% Call Release Delay 3500 ms Mean FFS : For Further Study

9. 9/57 Hansung University VoIP - Network Service Quality  ITU-T Y.1541 Class IPTD : IP packet transfer delay IPDV : IP packet delay variation IPLR : IP packet loss ratio IPER : IP packet error ratio Performanc e parameter QoS Class Class0Class1Class2Class3Class4 Class5 Unspecified IPTD100ms400ms 1sU IPDV50ms UUUU IPLR1x10 -3 U IPER1x10 -4 U VoIP 서비스를 위해 class1 이상의 수준을 만족

10. 10/57 Hansung University 서비스 분류 세부분류특 징특 징응용서비스 QoS 보장방법 IntServ Guaranteed Service (GS)  모든 연결에 대해서 일정 크기의 대역을 제공 ( 정량적 QoS 보장 )  단대단으로 확실한 지연을 보장하고 패킷 손실이 없도록 함 실시간 응용 ( 음성, 비디오, 오디오 등 ) 단대단 최대 대역 예약 RSVP 사용 Controlled Load Service (CLS)  정량적 QoS 는 보장하지 않음  한정된 정도의 패킷 지연이나 손실을 허용  최소한의 대역을 보장 비실시간 데이 터 통신 단대단 최소 대역 예약 DiffServ Expected Forwarding (EF)  각 노드에서 타 패킷의 영향을 받지 않고 우선적으로 전송되어 고정대역을 제공  상대적 우선 전송 VPN, VoIP 등 우선적으로 패킷 전송 Assured Forwarding (AF)  최저 대역을 보장  응용에 따라 서비스 품질 허용치 조절이 가능 Web, ftp, E- mail 등의 가변 속도 EF 가 사용하지 않 는 대역을 sub- class 별로 가중 할당 IETF QoS Class

11. 11/57 Hansung University 서비스 형태서비스 특징응용서비스 QoS 요구치 (ITU-T 기준 ) 비실시간 데이터 서비스  기존의 BE 클래스 E-mail, ftp, Low quality video 지연요구사항은 없음 Packet loss rate < 10 -3 실시간 음성 서비스  전용선과 같은 레벨의 QoS 상시 보장을 요구 VoIP, Interactive multimedia E2E delay < 150ms for 99.99% of packets, Packet loss rate < 10 -6 실시간 비디오 서비스  시간에 따라 전송률이 변화 TV 급 Video, Teleconference E2E delay < 250ms for 99.99% of packets, Packet loss rate < 10 -4 실시간 / 비실시간 복합 서비스  Intractive 성격이 강 한 응용서비스 www, online game, Streaming multimedia 계약한 최소 대역폭요구 Packet loss rate < 10 -6 QoS Requirements for Services

12. 12/57 Hansung University  Traffic classification  QoS 요구사항에 따라 분류 트래픽 QoS 에 대한 표현체계를 정의하는 기술  Signalling  사용자와 망간에 트래픽 /QoS 계약 및 QoS 요구전달, 자원확보를 위해 제어신 호를 전달하는 기술  Routing  망내 토폴로지 상태 정보 수집, 경로설정 정보를 망 내 노드간에 교환하는 기 술  Packet processing  차별화된 서비스 품질 요구사항에 따른 패킷 큐잉과 스케쥴링, 폭주시의 차별 화된 패킷처리 기술  Resource management  망 내 대역폭 할당 및 관리 기술 Functional Requirements for QoS Service

13. 13/57 Hansung University Classification & Mapping Metering & Marking Traffic Control & Policing/Shaping Scheduling Shaping & Flow Control Traffic Stream Packet Rewriting S/D IP Addr. Port Num. ToS DSCP WRED WFQ Mapping Traffic Stream QoS Scheme at Node

14. 14/57 Hansung University  What is Packet Classification  Classify packets into groups with the same QoS Metrics  Why Packet Classification is Required  Simplify QoS scheme by handling all the traffic with the same QoS requirement together  Criteria of Packet Classification  Network internal criteria : MAC Addr., IP Addr., Port Num., etc  Network external criteria : Subscriber, Service Type, etc  Ex) Gold, Silver, Bronze, (other or BE) Packet Classification

15. 15/57 Hansung University  What is Mapping  To correlate the priority of packets, priority of classes, or service priority among different layers, different protocols, or different service domains.  Why Mapping is Required  To maintain a consistent characteristics of the traffic pass through different layers, different protocols, or different service domains.  Type of Mapping  L2-L3 Mapping : CoS to ToS, CoS to DSCP  L3-L3 Mapping : ToS to DSCP, DSCP to ToS  L3-L2 Mapping : ToS to CoS, DSCP to CoSMapping

16. 16/57 Hansung University  What is Metering & Marking  Metering : to measure the incoming IP packets.  Marking : to mark packets according to the metering result  Why the Metering & Marking is Required  To verify that incoming packets conform to the traffic profile.  To provide a basis for differentiated services depending on the conformance. Metering & Marking (Traffic Conditioning)

17. 17/57 Hansung University  What is Traffic Control  The process to drop packets fairly to avoid the performance degradation due to the network congestion  Why Traffic Control is Required  To resolve the performance degradation due to buffer overflow  Congestion Management Tools (Queuing Algorithm) Traffic Control

18. 18/57 Hansung University  Policing  Drop the traffic over the bandwidth contracted  Implemented both in the ingress and egress side  Implemented using Token Bucket or Leaky Bucket  Shaping  Queue the traffic over the bandwidth contracted and transmit it in a specific rate  Usually implemented in the egress side  Traffic shaping algorithm Policing & Shaping

19. 19/57 Hansung University  What is Scheduling  The process to select a packet to be transmitted next from the multiple flows sharing a specific link or port  Function of Scheduling  To select a queue which stores an incoming packet  Policies to handle packets when buffer overflow occurs  Algorithms to provide services for diverse queueScheduling

20. 20/57 Hansung University  Integrated Services Model / RSVP (IntServ/RSVP)  Differentiated Services Model (DiffServ)  Multiprotocol Label Switching (MPLS)  Traffic Engineering QoS Technologies for IP Networks

21. 21/57 Hansung UniversityIntServ  Approach  Per-flow traffic handling at each hop  Per-flow resource reservation through signaling (RSVP)  Service Type  Guaranteed service for intolerant applications  Controlled load service for tolerant applications  Best-effort service  Protocol support  RSVP(Resource reservation protocol)  Algorithms  Scheduling  Admission control

22. 22/57 Hansung University IntServ (cont.)  Guaranteed Service : hard real-time  Guarantee a deterministic upper bound on delay for each packet in a session provided that the session does not send more than it specifies  Admission control based on worst-case analysis  Controlled Load Service : soft real-time  Provide a flow with a QoS closely approximating the QoS that the same flow would receive from an unloaded network  Measurement based admission control possible

23. 23/57 Hansung University IntServ Architecture  RSVP PATH message 전송 : traffic 특성 전송 (flowspec)  Path 는 IP routing protocol 이나 traffic engineering 에 의해 설정 Edge Router Core Router Host Host B Host A 1) 2) 3)  QoS 보장된 경로를 따라 데이터 전송  RSVP RESV message 전송 : 경로 내의 모든 노드간의 down-link 자원을 예약  Admission control  filterspec

24. 24/57 Hansung University Application RSVP daemon Policy Control Packet Scheduler Admission Control RSVP daemon Policy Control Classifier Packet Scheduler Routing Protocol Daemon RSVP data HOST ROUTER Classifier Admission Control IntServ Reference Model

25. 25/57 Hansung University Disadvantages of IntServ  Complex RSVP implementation  Large state machine  Support of mulicast  Poor scalability  Per-flow reservation and per-flow traffic handling not applicable to backbone core routers : too many flows Overload of classifiers Overload of scheduler (many number of queues) Overload of RSVP signaling daemon 패킷 흐름의 수가 적은 인트라넷 환경에서 적합한 모델

26. 26/57 Hansung UniversityDiffServ  Principle : Control traffic at the edge  Less complexity  Hop-by-hop instead of end-to-end  Fast processing on core router  Less state signaling, processing, storing  Small number of service class  At the ingress router (first router in the domain)  Packet classification into few classes  Packet marking with DSCP  Policing, shaping, dropping  Packet queuing and scheduling per DSCP class  At core and egress router  Packet queuing and scheduling per DSCP class

27. 27/57 Hansung University M M M R R R H H H H H Edge Router - Conditioning - Classifying - Marking Core Router - AF PHB - EF PHB DiffServ Architecture

28. 28/57 Hansung University Expedited Forwarding PHB  EF PHB  Called Premium service, Virtual Leased Line  Guaranteeing low loss rate, low latency, low jitter  Intended to send circuit replacement traffic across a DiffServ network  Suitable for IP telephony or video conferencing  Characteristics  Arrival rate  Configured rate Delivered with a high degree of assurance and with almost no distortion of the inter-packet timing  Arrival rate > Configured rate Unconditionally discarded

29. 29/57 Hansung University EF PHB Configuration M M M R R R H H H H H MeterMarker SLA Packet EF packet Edge Router Discarded Priority Que. EF packet Best-effort

30. 30/57 Hansung University Assured Forwarding PHB  Providing better service  Service Level Agreement (SLA) specify the amount of BW  Packet marked IN/OUT  Different drop rate for IN/OUT packet  Code 값 : XXXYY0 (XXX : Class 종류, YY : 폐기 우선 순위 )  4 개의 class 등급에 따라 패킷 전달 순서를 결정  3 개의 drop precedence 에 따라 congestion 발생시 패킷 폐기 순서를 결정

31. 31/57 Hansung University AF PHB (cont.) M M M R R R H H H H H MeterMarker SLA PacketsIN/OUT Edge Router 1 OUTIN RED IN/OUT

32. 32/57 Hansung UniversityMPLS  General operations  label creation and distribution  table creation at each router  label-switched path creation  label insertion/table lookup  packet forwarding/switching  Label removal  Characteristics  Used in core network to improve packet forwarding performance Label swapping without L3 lookup as LSRs except LERs  Improve network scalability (use of LSPs)  Integrated IP and ATM network  IP-VPN

33. 33/57 Hansung University MPLS Major Components  LERs (Label Edge Routers)  It located at the boundary of the MPLS network and its function is as signment and removal of labels as packet enter end leave the MPLS network respectively.  LSRs (Label Switching Routers)  It located at the core part of the MPLS network and its function perfo rm packet switching based on the label.  LDP (Label Distribution Protocol)  Maps IP address into MPLS labels  LSPs (Label-switched Paths)  A flow of MPLS packets with same label  Similar to VC in ATM network

34. 34/57 Hansung University  Existing ATM hardware  Ultra fast forwarding  IP Traffic Engineering  Constraint-based Routing  Virtual Private Networks  Controllable tunneling mechanism  Voice/Video on IP  Delay variation + QoS constraints Why MPLS

35. 35/57 Hansung University 47.1 47.2 47.3 1 2 3 1 2 1 2 3 3 Mapping: 0.40 Request: 47.1 Mapping: 0.50 Request: 47.1 MPLS Label Distribution

36. 36/57 Hansung University 47.1 47.2 47.3 1 2 3 1 2 1 2 3 3 IP 47.1.1.1 MPLS Label Switched Path (LSP) 1

37. 37/57 Hansung University Traffic Engineering  What is Traffic Engineering  To facilitate efficient and reliable network operations while simultaneously optimizing network resource utilization and performance (RFC 2702)  For efficient network  Constraint based routing To select routes that can meet certain QoS requirements To increase the utilization of the network  For reliable network  Traffic rerouting

38. 38/57 Hansung University Traffic Trunk  An abstraction of aggregated traffic flow that follows the same path(s) (within a service provider) between two end points  Same QoS requirements Traffic trunk R source destination Physical network R R R R R R R R Traffic Trunk attributes Resource attributes

39. 39/57 Hansung University Resource Attributes  Resource attributes parameters  Bandwidth  link attributes  TE-specific link metrics 등  Broadcast the resource attributes to all routers Traffic trunk R source destination Physical network R R R R R R R R (Priority=3, bandwidth= 50) Resource attributes BW(3)=100 BW(3)=30 BW(3)=80 BW(3)=30 BW(3)=80 BW(3)=50 BW(3)=20 100 60 70 50

40. 40/57 Hansung University Constraint based Routing  Input (constraint)  traffic trunk attributes  link resource attributes  topology state information  Output  explicit path(a list of routers) that satisfies the constraints  Route computation  Prune the links that do not satisfy the constraints  Select the shortest path among the paths

41. 41/57 Hansung University Constraint based Routing (cont.) R R R R R R R BW(3)=100 BW(3)=80 BW(3)=30 BW(3)=50 BW(3)=20 BW(3)=60 BW(3)=100 BW(3)=70 BW(3)=50 R R Priority=3 bandwidth= 50 X X X

42. 42/57 Hansung University QMS (Hanaro)  Internet Speed  Download/Upload 평균, 최대, 최소, 표준편차  RTT 평균, 최대, 최소, 표준편차  Packet Loss SLA 에 적용하지 않음

43. 43/57 Hansung University QMS (Hanaro)  SLA Speed  상품별 최저 download 보장 속도 기준 마련 (Pro: 1M, Mid: 0.7M, Lit: 0.5M)  60 분간 10 회 이상 측정, 60% 이상 초 ; 저속도 미달시 보상 ( 기본서비스 요금의 30% 감면 )  고객이 측정하여 보상 신청 유명무실한 SLA

44. 44/57 Hansung University QoS – Why not deployed yet?  Weak economy  Requirement on customer side not very strong  Primitive QoS tools  Negative impact on packet forwarding performance on deployed routers  Accounting, Billing, Pricing still open  Inter-Provider, Inter-Domain issues

45. 45/57 Hansung University  QoS Monitoring  QoS Measurement  QoS Solutions QMS

46. 46/57 Hansung University  QoS Monitoring 기술의 분류  Traffic Measurement  Traffic Analysis  Traffic Presentation  Monitoring 범위에 따른 분류  Protocol Monitoring  Network Monitoring  End-to-End QoS Monitoring  Monitoring 방식에 따른 분류  Passive 방식 Protocol Monitoring or Network Monitoring  Active 방식 End-to-End QoS Monitoring QoS Monitoring 기술

47. 47/57 Hansung University QoS Monitoring 기술 (cont.)  Protocol Monitoring  특정 QoS 구조를 가지는 Network Domain 에서 사용되고 있는 Protocol 이 제대로 작동되고 있는 지를 점검하는 기술  특정 QoS 관리 구조의 완결성을 확인할 수 있는 기준  예 ) RSVP Protocol Monitoring Tool – RSVP Diagnostics Tool  Network Monitoring  Network Resource 의 가용정도 및 소비정도를 측정하여 전반적인 Status 를 Monitoring 함으로써 Network Engineering 과 관리에 도움을 줌  기존의 SNMP 기반의 Network 관리 Tool 을 활용하여 QoS 보장 관련 기술 용 MIB 을 구현함으로써 Monitoring 가능  예 ) Real-time Traffic Flow Measurement (RTFM) 구조 Tool

48. 48/57 Hansung University QoS Monitoring 기술 (cont.)  End-to-End QoS Monitoring  이질적인 QoS 관리 구조를 가지는 Network Domain 간의 QoS 보장의 제공 여부를 점검하는 데 중요한 기능을 함  원하는 Traffic 생성을 원하는 시점에 원하는 양 만큼을 할 수 있는 Tool 과 그 결과를 Monitoring 할 수 있는 Tool 이 필요  정확한 Timing, 다양한 종류의 Test Traffic 으로 정확한 Monitoring 이 상당 한 어려움으로 신중한 사전 설계와 구조의 정의가 요구됨

49. 49/57 Hansung University QoS Measurement  Measurement Infrastructure Network node Active monitor Passive monitor User Control information collector User Active monitor User User traffic Measurement packets Copied traffic Splitter Report packets

50. 50/57 Hansung University Active vs. Passive Measurement  Active Measurement  Method : Inject measurement traffic into the network  Metric : RTT, One-way delay, Packet loss, Topology  Tools : Ping, Traceroute, NIMI, Surveyor, PingER, AMP  Passive Measurement  Method : Do not inject traffic but observe traffic  Metric : Link Utilization, Traffic Analysis  Tools : Tcpdump, MRTG, FlowScan, CoralReef

51. 51/57 Hansung University Passive Measurement  FlowScan  A Network Traffic Flow Reporting and Visualization Tool developed by Dave Plonka  CISCO router 서 보내는 flow data 를 분석하는 정보를 얻어내는 tool  Network 의 border traffic 을 측정하여 실시간에 거의 근접하게 (5 분 단위 ) 그래프를 생성  Flow  Packets with the same src ip & port, dst ip & port, protocol #  NetFlow – flow information exported by CISCO router Cisco IOS Release 12.0 and later releases 는 대부분 지원  FlowScan uses NetFlow data to analyze the traffic data.

52. 52/57 Hansung University CISCO ’ s NetFlow entry

53. 53/57 Hansung University FlowScan Graph

54. 54/57 Hansung University Active Measurement  Mechanism  Sending the measurement-packets(probes) into any network to guess the state and behavior of the network.  Metric (end-to-end)  RTT, One-way delay, Delay Jitter, Packet loss, Throughput  Tools  Skitter: CAIDA Group Solution  PingER: IEPM Group Solution  Surveyor: Advanced Network & Services Solution  RIPE: RIPE Network Coordination Centre (NCC) Solution  AMP: NLANR Solution  AMTv6: ETRI Solution

55. 55/57 Hansung University Comparison among Tools MetricAMTv6SurveyorRIPEPingERAMP Method 1way delay, loss, delay-jitter 1way delay, loss 2way delay, loss HostsDedicated SelectedDedicated Time sync.GPS NTP Scheduling Poisson Bursty (30m) Linear random about 1 st 15 sec of min Packet size 52 Bytes 72 Bytes(IPv6) 40 Bytes100 Bytes100 Bytes & 1000 Bytes 64 Bytes

56. 56/57 Hansung University Internet System Architecture Measurement System A Measurement System B Control System GPS Satellite GPS Receiver ② Measurement-packets ③ Measured data ① Measurement command Visualizer

57. 57/57 Hansung UniversityReferences  http://cnscenter.future.co.kr/ietf/qos.htmlhttp://cnscenter.future.co.kr/ietf/qos.html  IETF Working Group, RFC documents, Working group Drafts, Individual Drafts MPLS DiffServ IntServ/RSVP Traffic Engineering Resource Allocation Protocol