2. Jump to first page DiffServ Overview University of British Columbia Cpsc 527 Advanced Computer Communications Lecture 7b DiffServ Overview Instructor: Dr. Son T. Vuong Email: vuong@cs.ubc.ca The World Connected

3. Spring 2010 – Dr. Son VuongCpsc 527 2 Outline n Overview of QoS and DiffServ n Our research on DiffServ u A Class of Protocols u Dynamic (Fair) Distribution DiffServ n Conclusions

4. Spring 2010 – Dr. Son VuongCpsc 527 3 What is the Service on Internet n A "Service" defines some significant characteristics of packet transmission in one direction across a set of one or more paths within a network. n These characteristics may be specified in quantitative or statistical terms of: u throughput, delay, jitter, loss.

5. Spring 2010 – Dr. Son VuongCpsc 527 4 Current Problems with the Internet n Best-effort service has strained its current infrastructure capabilities. n A challenge is to design routers that can provide service differentiation for various types of traffic. n Two solutions proposed in IETF: u Integrated service (IntServ): individual flows with specific QoS requirements, e.g. RSVP flows u Differentiated services (DiffServ): aggregate of flows and per-hop behavior.

6. Spring 2010 – Dr. Son VuongCpsc 527 5 What is QoS and why we need it? n QoS has many different measures: u Delay, delay variance (jitter), rate u Fairness, flow isolation u Reliability, e.g. drop rate n QoS is needed in packet networks to support multimedia applications: IP Telephony, mobile web-access, Video teleconferencing, net-PC applications, Grid services, etc.

7. Spring 2010 – Dr. Son VuongCpsc 527 6 Application QoS requirements n Different applications=different QoS requirements. u Delay-sensitive real-time applications: Voice, Interactive Video. u Delay-tolerant real-time applications: Streaming Video. u Elastic applications that tolerate a wide range of delay variations: FTP, Telnet and e-mail.

8. Spring 2010 – Dr. Son VuongCpsc 527 7 How to Provide QoS in a Best-Effort Internet? n More bandwidth does not solve the problem it only moves it to the core of the network: u IP router technology has a limit to its computational capabilities. u Also, fairness is not solved when bandwidth is increased. n Solution: packet scheduling and resource reservation.

9. Spring 2010 – Dr. Son VuongCpsc 527 8 Improving QOS in IP Networks n IETF groups are working on proposals to provide better QOS control in IP networks, i.e., going beyond best effort to provide some assurance for QOS n Work in Progress includes RSVP, Differentiated Services, and Integrated Services n Simple model for sharing and congestion studies:

10. Spring 2010 – Dr. Son VuongCpsc 527 9 Principles for QOS Guarantees n Consider a phone application at 1Mbps and an FTP application sharing a 1.5 Mbps link. u bursts of FTP can congest the router and cause audio packets to be dropped. u want to give priority to audio over FTP n PRINCIPLE 1: Marking of packets is needed for router to distinguish between different classes; and new router policy to treat packets accordingly

11. Spring 2010 – Dr. Son VuongCpsc 527 10 Principles for QOS Guarantees (more) n Applications misbehave (audio sends packets at a rate higher than 1Mbps assumed above); n PRINCIPLE 2: provide protection (isolation) for one class from other classes n Require Policing Mechanisms to ensure sources adhere to bandwidth requirements; Marking and Policing need to be done at the edges:

12. Spring 2010 – Dr. Son VuongCpsc 527 11 Principles for QOS Guarantees (more) n Alternative to Marking and Policing: allocate a set portion of bandwidth to each application flow; can lead to inefficient use of bandwidth if one of the flows does not use its allocation n PRINCIPLE 3: While providing isolation, it is desirable to use resources as efficiently as possible

13. Spring 2010 – Dr. Son VuongCpsc 527 12 Principles for QOS Guarantees (more) n Cannot support traffic beyond link capacity n PRINCIPLE 4: Need a Call Admission Process; application flow declares its needs, network may block call if it cannot satisfy the needs

14. Spring 2010 – Dr. Son VuongCpsc 527 13 Summary

15. Spring 2010 – Dr. Son VuongCpsc 527 14 Scheduling And Policing Mechanisms n Scheduling: choosing the next packet for transmission on a link can be done following a number of policies; n FIFO: in order of arrival to the queue; packets that arrive to a full buffer are either discarded, or a discard policy is used to determine which packet to discard among the arrival and those already queued

16. Spring 2010 – Dr. Son VuongCpsc 527 15 Scheduling Policies n Priority Queuing: classes have different priorities; class may depend on explicit marking or other header info, eg IP source or destination, TCP Port numbers, etc. n Transmit a packet from the highest priority class with a non-empty queue n Preemptive and non-preemptive versions

17. Spring 2010 – Dr. Son VuongCpsc 527 16 Scheduling Policies (more) n Round Robin: scan class queues serving one from each class that has a non-empty queue

18. Spring 2010 – Dr. Son VuongCpsc 527 17 Scheduling Policies (more) n Weighted Fair Queuing: is a generalized Round Robin in which an attempt is made to provide a class with a differentiated amount of service over a given period of time

19. Spring 2010 – Dr. Son VuongCpsc 527 18 Policing Mechanisms n Three criteria: u (Long term) Average Rate (100 packets per sec or 6000 packets per min??), crucial aspect is the interval length u Peak Rate: e.g., 6000 p p minute Avg and 1500 p p sec Peak u (Max.) Burst Size: Max. number of packets sent consecutively, ie over a short period of time

20. Spring 2010 – Dr. Son VuongCpsc 527 19 Policing Mechanisms Token Bucket: limit input to specified Burst Size and Average Rate. n bucket can hold b tokens n tokens generated at rate r token/sec unless bucket full n over interval of length t: number of packets admitted less than or equal to (r t + b).

21. Spring 2010 – Dr. Son VuongCpsc 527 20 Policing Mechanisms (more) n token bucket, WFQ combine to provide guaranteed upper bound on delay, i.e., QoS guarantee! WFQ token rate, r bucket size, b per-flow rate, R D = b/R max arriving traffic

22. Spring 2010 – Dr. Son VuongCpsc 527 21 What Is Scheduling and Why Do We Need It? n In real networks, a large number of flows may require different guarantees from the network. n Scheduling determines the next eligible packet, among all the arriving packets, to be sent out in order to satisfy the required guarantees. n Attributes of a good scheduler: u Rate and delay guarantees. u Fairness. u Utilization. u Simplicity.

23. Spring 2010 – Dr. Son VuongCpsc 527 22 Scheduling (revisited) n Scheduling: per-flow or aggregated flows. n Per-flow (IntServ): u Provides exact delay and fairness guarantees. u Requires maintaining state information for all active sessions on a link. n Aggregation (DiffServ): u Delay and fairness are not as easy to guarantee. u No (or minimal amount of) states need to be maintained.

24. Spring 2010 – Dr. Son VuongCpsc 527 23 Types of Schedulers n Rate-controlled: Delay / jitter Earliest Due Date (D-EDD and J-EDD). n Non-rate-controlled: Generalized Processor Sharing (GPS) and Weighted Fair Queueing (WFQ).

25. Spring 2010 – Dr. Son VuongCpsc 527 24 How GPS Works n Every session has its own queue and share. n Weighted Round Robin (WRR) service order with an infinitesimal amount serviced from each session. n Empty queues are skipped when their turn arrives.

26. Spring 2010 – Dr. Son VuongCpsc 527 25 End-to-End Network Delay n Delay=Host Delay+Access Delay+Network Delay n Even if the network provided the required QoS without the host applications or access networks providing their guarantees then the end-to-end guarantees will not be met.

27. Spring 2010 – Dr. Son VuongCpsc 527 26 WFQ vs. GPS

28. Spring 2010 – Dr. Son VuongCpsc 527 27 Scheduling Policies (more) n Weighted Fair Queuing: is a generalized Round Robin in which an attempt is made to provide a class with a differentiated amount of service over a given period of time

29. Spring 2010 – Dr. Son VuongCpsc 527 28 Proactive Packet Discard n Congestion management by proactive packet discard u Before buffer full u Used on single FIFO queue or multiple queues for elastic traffic u E.g. Random Early Detection (RED)

30. Spring 2010 – Dr. Son VuongCpsc 527 29 Random Early Detection (RED) Motivation n Surges fill buffers and cause discards n On TCP this is a signal to enter slow start phase, reducing load u Lost packets need to be resent F Adds to load and delay u Global synchronization F Traffic burst fills queues so packets lost F Many TCP connections enter slow start F Traffic drops so network under utilized F Connections leave slow start at same time causing burst n Bigger buffers do not help n Try to anticipate onset of congestion and tell one connection to slow down

31. Spring 2010 – Dr. Son VuongCpsc 527 30 RED Design Goals n Congestion avoidance n Global synchronization avoidance u Current systems inform connections to back off implicitly by dropping packets n Avoidance of bias to bursty traffic u Discard arriving packets will do this n Bound on average queue length u Hence control on average delay

32. Spring 2010 – Dr. Son VuongCpsc 527 31 RED Algorithm – Overview Calculate average queue size avg if avg < TH min queue packet else if TH min  avg  Th max calculate probability P a with probability P a discard packet else with probability 1-P a queue packet else if avg  TH max discard packet

33. Spring 2010 – Dr. Son VuongCpsc 527 32 RED Buffer

34. Spring 2010 – Dr. Son VuongCpsc 527 33 RED Algorithm Detail

35. Spring 2010 – Dr. Son VuongCpsc 527 34 Architecture of DiffServ n DiffServ Domain n A Set of per-hop (forwarding) behaviors (PHB). n Packet classification functions. n Traffic conditioning functions including: u metering u marking u shaping u policing

36. Spring 2010 – Dr. Son VuongCpsc 527 35 DS Nodes n DS boundary Nodes u Classify the ingress packets. u Perform traffic conditioning according to a traffic conditioning agreement (TCA) n DS Interior Nodes u Perform limited traffic conditioning functions such as DS codepoint re-marking.

37. Spring 2010 – Dr. Son VuongCpsc 527 36 DS Definition n 000000 : Best - effort behavior n Pool Codepoint space Policy n 1 xxxxx0 Standards Action n 2 xxxx11 EXP/LU n 3 xxxx01 EXP/LU (*) n (*) May be used for future Standards Action allocations as necessary

38. Spring 2010 – Dr. Son VuongCpsc 527 37 Example of DiffServ Domains RRRRHHHHHHBB

39. Spring 2010 – Dr. Son VuongCpsc 527 38 Per-Hop Behaviors (PHB) n PHB is a description of forwarding behavior of a DS node for a particular DS behavior aggregate. n Via PHBs, a node allocates its resources (buffer, bandwidth, delay, loss) to behavior aggregates. n PHBs are implemented via buffer management and packet scheduling mechanisms. n A PHB is selected at a node by a mapping of the DS codepoint in a received packet.

40. Spring 2010 – Dr. Son VuongCpsc 527 39 Traffic Conditioning Specification n The TCS specifies detailed service parameters for each service level : F Service performance parameters such as expected throughput, drop probability, latency. F Traffic profiles which must be adhered to for the requested service to be provided, such as token bucket parameters. F Disposition of traffic submitted in excess of the specified profile. F Marking services provided. F Shaping services provided.

41. Spring 2010 – Dr. Son VuongCpsc 527 40 Packet Classifier and Traffic Conditioner n Classifier : select packets in a traffic stream based on packet header. n Meter: measures the temporal properties of the stream of packets selected by a traffic profile. n Marker: sets the DS field of a packet to a particular codepoint, add marked packet to a DS behavior aggregate. n Shaper: shapes the traffic stream to bring it into compliance with a traffic profile. n Dropper: discards some or all packets in a traffic stream to bring it into compliance with a traffic profile.

42. Spring 2010 – Dr. Son VuongCpsc 527 41 Service Level Specification (SLS) n In addition to TCS, the SLS specifies more general service : F Availability/Reliability F Encryption services. F Routing constraints. F Authentication mechanisms. F Mechanisms for service monitoring and auditing. F Responsibilities such as location of the equipment and functionality, action if the contract is broken, support capabilities. F Pricing and billing mechanisms.

43. Spring 2010 – Dr. Son VuongCpsc 527 42 Profile in Router n Profile ----> behavior of the router n There may be many profiles in the router at any time. n Profile contains : n Name n Attribute (in/out) n Policy Scope n Other information (Address, port, permission, month mask, time of day, direction …)

44. Spring 2010 – Dr. Son VuongCpsc 527 43 Example of Profile n Name Entry2 PolicyScope DataTraffic TimeOfDayRange 090000 to 170000 IncomingTOS 111 SourceAddressRange 139.24.2.12 to 139.24.2.255 Direction Outgoing MaxeRate 5000 OutgoingTOS 101

45. Spring 2010 – Dr. Son VuongCpsc 527 44 Examples of Services n Better than Best-Effort (BBE) Service : u This is a qualitative service which promises to carry specific web server traffic at a higher priority than best-effort traffic. n Premium Service : u Customer purchases a peak rate u Packets below that rate carried with no delay u Packets above that rate dropped

46. Spring 2010 – Dr. Son VuongCpsc 527 45 Our research on DiffServ n We focus on differentiated service since this aggregate service scales better than the per-flow integrated service. n We focus on developing a class of protocols including the Meter, Marker, Shaper at the Boundary nodes and the per-hop behavior of all interior nodes in a DS Domain. n For performance evaluation, the network simulator NS2 was used.

47. Spring 2010 – Dr. Son VuongCpsc 527 46 DiffServ and BestEffort Queues DiffServ queue Best-effort queue 

48. Spring 2010 – Dr. Son VuongCpsc 527 47 The simulation of DiffServ and BestEffort for TCP/IP traffic

49. Spring 2010 – Dr. Son VuongCpsc 527 48 The Dynamic Distribution DiffServ n The scheme unites DiffServ flows with Dynamic Distribution to the aggregate before sending them to the network. n Dynamic Distribution prevents any aggregate from taking all the bandwidth in a period of time, thus allows every aggregate a chance of passing through the router at all times. n Dynamic Distribution also guarantees Multimedia applications an upper bound on packet delay inside the DiffServ domain without the per-flow requirement.

50. Spring 2010 – Dr. Son VuongCpsc 527 49 Multiple Queues in DiffServ Router Figure 1. Multiple-Queues in a Router of the DiffServ Domain DiffServ queues Backlogged packets

51. Spring 2010 – Dr. Son VuongCpsc 527 50 The Scheme of Dynamic Distribution Differentiated Service n This scheme simplifies processing at routers in DiffServ domain (cf. the IntServ scheme) while maintaining QoS guarantee for every flow. n At Boundary Nodes, packets are classified by their flows, which may number in thousands at one node or tens of thousand in the whole domain n At Inside Nodes, packets classified by their aggregates, numbered in hundreds at a node.

52. Spring 2010 – Dr. Son VuongCpsc 527 51 The Scheme at Boundary Node n Classifier: selects packets in a traffic stream based on DiffServ byte in header. n Meter: The leaky-bucket meter scheme for every flow with its subscribed bandwidth. n Dropper: drops all out-profile packets. n Shaper: Fair-Queue scheme for every DiffServ flow depending on the subscribing bandwidth. n Marker: an aggregate marking scheme n PHB: exact Fair-Queue forwarding scheme.

53. Spring 2010 – Dr. Son VuongCpsc 527 52 The Scheme for Interior Nodes n Classifier: selects packets in a traffic stream based on DiffServ byte and places them in appropriate queues. n Shaper: time divided into equal slots, each further divided mini-slots. Each aggregate receives a number of mini slots based on calculation at the beginning of the slot. n PHB: Frame-based forwarding scheme with Dynamic Distribution of resources.

54. Spring 2010 – Dr. Son VuongCpsc 527 53 n The proposed Dynamic Distribution DiffServ scheme is flexible, simple and efficient n Guarantees delay bound without the per-flow complexity, suitable for multimedia traffic. n Future work: u Various types of multimedia application traffic such as video teleconferencing u Various performance measures, including delay, throughput and jitter u Multicasting and security DS services Summary