1. Q O S IN THE I NTERNET Better than best-effort Andreas Liaker Feroz Zahid

2. Agenda Quality of Service – What is it and Why it is needed? IntSrv, ST-II and RSVP Differentiated Services MPLS Constraint Based Routing and Traffic Engineering Difference between ST-II and RSVP

3. What is Quality of Service? Guarantees for the predictable results Unlike Best-effort service What Guarantees? Bandwidth Latency Robustness Methods Resource Reservation (IntServ) Resource Prioritization (Differentiated Services)

4. Integrated services (IntServ) and RSVP Basic Idea Network components (routers) reserve resources to provide Quality of Service to specific packet streams Types Guaranteed Most strict IP possible version of dedicated virtual circuit Controlled-Load Equivalent to best-effort service under unloaded conditions

5. Some Terminologies TermDescription Traffic ProfileA description of the properties of a traffic stream e.g. rate and burst size PHBExternally observable behavior of a packet Admission Control The decision process of whether to accept a request for resources or not ClassificationThe process of sorting packets based on the content of packet headers as per the rules defined MarkingThe process of setting the DS field in a packet PolicingThe process of handling out of profile traffic e.g. discarding excess packets ShapingThe process of delaying packets within traffic stream to cause it to conform to some defined traffic profile SchedulingThe process of deciding which packet to send first in a system of multiple queues

6. IntServ Components

7. RSVP Protocol Runs on top of the routing protocol Implementation should be available on sender, receiver, router Carries resource requests all the way through the network At each hop consults admission control sets up reservation and packet filter. If fails, inform sender Reservation Style Wildcard Fixed Filter Dynamic Filter

8. RSVP Protocol – In IP Stack ULPs IP Link layer modules ICMPIGMPRSVP IP service interface Link layer service interface

9. RSVP Messages Sender PATH message Traffic specification Receiver RECV message containing Reservation specification Guaranteed or Controlled Filter specification Type of packets

10. RSVP Signaling Mechanism PATH RESV Quality-bound Communication

11. ST-II IP v5 (4 first bit Header) Experimental protocol Serves as an adjunct to, not a replacement for, IPv4 ST-II Multicast distribution tree Unicast routing table ST-agent Connect (each hop) Allocate resources May reduce resource request. Receiver must Accept or Refuse Accept can reduce resource requests

12. ST-II ST-II Hole stream is treated homogeneous. Stream source must wait for all Accept/Refuse reply Must adapt to lower QoS, or reject group participation (Disconnect message) Receivers can be added or deleted using IP. Must reduce QoS or reject if needed Reliability and Robustness Manage Stream with hop-by-hop acknowledgment Hello Message to neighboring ST Agents Only service model supported is homogeneous point to multipoint simplex distribution tree

13. ST-II Protocol – In IP Stack ULPs ST-II Link layer modules SCMP ST-II service interface Link layer service interface

14. IntServ - Conclusions Pros Highest level service guarantees Granularity of resource allocation Feedback for QoS enabled applications Cons Scalability? Amount of state information increases proportionally with the number of flows Huge storage and processing overhead on routers Ubiquitous deployment is required (for Guaranteed Service)

15. Differentiated Services Basic Idea All complex functionality shift to the edge routers Applied on flow aggregates Services requirements are classified A predefined per-hop behavior (PHB) is applied to every service class Traffic is smoothed according to PHB applied Types Assured service Premium service Ordinary Best-effort service

16. Differentiated Services – Assured Service Defined in terms of user profile how much assured traffic is a user allowed to inject into the network Network provides a lower loss rate than best-effort In case of congestion best-effort packets are dropped first User sends no more assured traffic than its profile If it sends more, the excess traffic is converted to best-effort

17. End to End Service Delivery – Delivery of Assured Service with Static SLA 1. Host S sends a RSVP message to the local Bandwidth Broker (CN1-BB) requesting for Assured Service for its traffic. RSVP Granted 2. CN1-BB configure leaf router LR1 so that LR1 can set the A-bits of the packets of this flow. CN1-BB will also reply to the host S. 3. Host S sends packets to leaf router LR1.4. LR1 mark A-bits of the packet.5. Every router from LR1 (excl) to ER1 (incl) does a BA classification. Packets with the A- bit set are considered as in. 6. BR1 polices the traffic. If the in traffic exceeds its bit-rate, the excess packets’ A-bits will be reset. 7. All routers between boundary router BR1 and BR2 (incl) perform BA classifications and apply RIO on their AQs. 8. ER2 performs the same operations as BR1. 9. The packets are eventually delivered to host D.

18. Differentiated Service – Premium Service Provides the abstraction of a virtual pipe between an ingress and an egress router Network Guarantees that premium packets are not dropped and they experience low delay User does not send more than the size of the pipe If it sends more, excess traffic is delayed, and dropped when buffer overflows

19. Delivery of Premium Service with Dynamic SLA Phase I - Signaling 1. Host S sends RSVP PATH message to the local bandwidth broker CN1-BB. PATH 2. CN1-BB makes an admission control decision. Granted 3. If request is accepted, CN1-BB sends PATH message to ISP1-BB.4a. ISP1-BB makes an admission control decision. Granted 4b. If request is accepted, ISP1-BB sends PATH message to CN2-BB.5a. CN2-BB makes an admission control decision. Granted 5b. If request is accepted, CN2-BB set the classification and policing rules on router ER2 using LDAP or RSVP. RESV 5c. CN2-BB then sends RSVP RESV message to ISP1-BB. 6a. ISP1-BB configures classification and policing rules on router BR1, and the policing and reshaping rules on router ER2. 6b. ISP1-BB then sends RESV message to CN1-BB. 7a. CN1-BB will set classification and shaping rules on router LR1 and router ER1. 7b. CN1-BB will then set RESV message to host S.8. Host S may now start transmitting data packets.

20. Delivery of Premium Service with Dynamic SLA Phase II - Data Transmission 1. Host S sends packets to the leaf router LR1. 2. Leaf router LR1 performs a MF classification. If the traffic is non-conformant, LR1 will shape it. It will also set the P-bits of the packets. 3. Each intermediate router between leaf router LR1 and ER1 performs a BA classification, puts the packet in PQ and sends them out. 4. ER1 performs a BA classification and reshapes the traffic to make sure that the negotiated peak rate is not exceeded. 5. BR1 classifies and polices the premium traffic. Excess premium packets are dropped. 6. Intermediate router between BR1 and BR2 (incl) performs BA classification. BR2 also reshapes the premium traffic. ` 7. ER2 classifies and polices the premium traffic. Excess premium packets are dropped.8. The premium packets are delivered to host D.

21. Differentiated Services - Conclusions Pros Scalable Edge routers maintain per aggregate state Core routers maintain state only for a few traffic classes Easier implementation Incremental deployment Cons Provide weaker service than IntServ

22. Quality of Service Pyramid Not Strict Less Strict Very Strict

23. Multi Label Protocol Switching (MPLS) Basic Idea Header of the packet contains a label that is used to advance the packet toward its destination The label simplifies the forwarding decision a node must make for the packet A group of packets forwarded in the same manner are said to belong to the same Forwarding Equivalence Class (FEC)

24. Multi Label Protocol Switching (MPLS) Label Switched Paths (LSPs) Within an MPLS domain, a path is set up for a given packet to travel based on a Forwarding Equivalence Class (FEC) The LSP is set up prior to data transmission

25. Multi Label Protocol Switching (MPLS) MPLS improves packet forwarding performance Enhances and simplifies packet forwarding through routers Layer-2 switching Simplicity allows for easy implementation MPLS supports QoS for service differentiation Use traffic-engineered path set-up and support QoS guarantees Classification and QoS service are determined by the labels

26. Multi Label Protocol Switching (MPLS) Picture Source: http://itknowledgeexchange.techtarget.com/network-engineering-journey/how-mpls-works/

27. Constraint Based Routing and Traffic Engineering Traffic Engineering Process of arranging traffic flows so congestion caused by uneven network utilization could be avoided Constraint-Based Routing Compute routes that are subject to rules Multiple constraints possible

28. Traffic Engineering Network congestion Lack of network resources Uneven distribution of traffic Lack of network Resources All routers and links are overloaded Only solution is to add more resources Uneven traffic distribution Dynamic Routing protocols such as RIP and OSPF always select the shortest paths to forward packets Traffic Engineering can be utilized to: Avoid congestion Provide graceful degradation in case of congestion Picture Source: http://www.geoexpertsolutions.com/

29. Constraint Based Routing With DiffServ Select those routes that most likely are to satisfy QoS requirements Constraint-Based Routing with RSVP Select the path for RSVP messages Constraint-Based Routing with MPLS MPLS as a forwarding scheme Constraint-based routing as a routing scheme

30. Differences between ST-II and RSVP Static Analysis Self-limiting applications Support heterogeneous groups Support channel selection Dynamic Analysis Network dynamics Group membership dynamics

31. Self-Limiting Applications Multipoint-to-multipoint applications Application-level constraints Few simultaneous senders Audio conference Simulate 60 routers 82 Links 2-65 participants

32. Self-Limiting Applications

34. Supporting heterogeneous groups Global Scale InterNetwork different demand for QoS Wide-spread distribution services Cable-TV distribution Broadcasting of an audio/video lecture Multiple data streams/signal quality level to receiver. ST-II Entire stream as a homogeneous Maximum requested resource along all links. RSVP Support for heterogeneous reservations

35. Supporting heterogeneous groups Heterogeneous mix of receivers listening to an audio lecture. Sending the entire data stream on a single multicast tree most efficient. High quality – 64Kb/s Low quality – 16Kb/s Number Of Low Quality Receivers ST-II Resource Allocation (Kb/S) RSVP Resource Allocation (Kb/S) 02944 1029442656 2029442176 3029441600 40736

36. Link reservations for RSVP heterogeneous audio lecture (40 receivers).

37. Supporting channel selection Large multiparty conferences Unable to receive from all active participants simultaneously Select dynamically a subset of the sources Assured channel selection Independent reservation for each source (ST-II & RSVP) Dynamic Filter Reservation (RSVP) Non-assured Channel Selection (ST-II & RSVP)

38. Channel selection resource overhead Group Size Chosen Source (4 Res) Dynamic Filter (4-Res) Independent Streams (N-1 Reservations) Resource Allocation (KB/S) Overhead RatioResource Allocation (KB/S) Overhead Ratio 53200 1 1.00 10659287041,32120321,83 159728131841,36238082,45 2011840184321,56362243.06 2514400227201,58524803,64 3520160327041,62891524,42 4526368420481,591403525,32 6036416570241,572264326.22

39. Network Dynamics Reliability and robustness in the face of network dynamics. ST-II Reliable control message protocol and a Hello protocol RSVP Datagram control message protocol in combination with a soft state refresh mechanism Difficulty to compare because they rely heavily on timers. Compare the design philosophies Recovery ST-II Requires that the network be responsible for correctness RSVP leaves the final responsibility for maintaining reservations with the ends. “Fate-sharing”

40. Network Dynamics Overhead ST-II: ST agent periodically exchanging one Hello message with each active neighbor. RSVP: Path and Reservation refreshes RSVP incorporates a protocol overhead reduction mechanism (merging)

41. Group membership dynamics Global distribution of a conference Participants tuning into and leaving the conference. ST-II Overhead Connect and Accept message between source and receiver Overhead proportional to the number of downstream receivers HotSpots Bottleneck RSVP Overhead Assuming homogeneoues recievers. One protocol message on each link in each direction Heterogeneous. Splices and sufficient resource allocated for more demanding requests

42. Protocol overhead for independent group joins for audio lecture

43. Group membership dynamics Latency ST-II Setup/teardown – One roundtrip source and receiver. RSVP Setup. One initial delay for Path refresh One hop to end to end depending possibility for «spliced» RSVP teardown Can release the resource immediately

44. Some thoughts… Internet QoS: A Bigger Picture 2007 - TM Bohnert et al. New Generation Networks, Wireless Networks Net Neutrality Internet service providers and governments should treat all data on the Internet equally Talk: Concept of QoS in the Internet Geoff Huston from APNiC, September 2012 Why QoS? Operators believe that this will allow them to extort revenues from Content service providers Solution? Add more bandwidth Adaptive behavior in applications Picture Source: http://1.bp.blogspot.com/

45. Summary Integrated Services Not Scalable Difficult to implement even in corporate networks Differentiated Services No per flow resource reservations No assured outcome Can’t guarantee fixed service response Can’t measure performance Internet QoS still not experimented much!