Internet QoS: A Big Picture

1
Internet QoS A Big Picture
2
Outline

• Introduction
• Integrated Services and RSVP
• Differentiated Services
• MPLS
• Traffic Engineering and Constraint-based Routing
• A Comparison of ATM Networks to router networks

3
Introduction

• Why QoS will be emerged?
• The Internet is only provides best-effort
  service.
• The internet will however be transformed into a
  commercial infrastructure.
• Thus, demands for services quality have rapidly
  developed
• What kinds of QoS Mechanisms is there?
• Integrated Service RSVP
• Differentiated Service
• MPLS
• Traffic engineering Constraint-based Routing
• In this paper, we describe how they differ from,
  related to, and work with each other to deliver
  QoS on the Internet.

4
Integrated Services and RSVP 1/2

• Propose two services
• Guaranteed service requiring fixed delay bound
• Controlled-load service requiring reliable and
  enhanced best-effort service
• Integrated Services is composed of
• Signaling protocol
• set up the paths and reserve resource. (e.g.
  RSVP)
• Admission control
• decides whether a request for resources can be
  granted.
• Classifier
• performs a multifield (MF) classification and put
  the packet in a specific queue based on it.
• Packet scheduler
• schedule the packet to meet its QoS requirement.

5
Integrated Services and RSVP 2/2

• RSVP signaling
• Problems
• Not scalable
• To maintain the amount of state information
• The requirement on routers is high
• To operate RSVP, admission control, MF
  classification, packet scheduling

6
Differentiated Services 1/4

• Differentiated Services is composed of
• DS field the field to contain the
  differentiated services class.
• PHB a base set of packet forwarding treatments
• SLA the agreement to specify the service
  classes supported and the amount of traffic
  allowed in each class.
• The progress of Differentiated Services
• At the ingress of the ISP networks
• Packets are classified, policed, and possibly
  shaped.
• DS field may be remarked by the SLA
• At the core router
• Forwarding each packet by PHB

7
Differentiated Services 2/4

• Assured Service
• Intended for customers that need reliable
  services from their service providers, even in
  times of network congestion.
• Packet is categorized by in and out at the
  ingress routers.
• fair traffic ? in packet
• excessive traffic ? out packet
• Core router
• Drop the out packet first, and then drop the
  in packet.

8
Differentiated Services 3/4

• Premium Service
• Deliver the premium packet before everything
• If the packet is premium ? the P-bit of a packet
  is set
• Premium Queue (PQ) is used to receive the premium
  packets.
• Admission Control
• Limited to a 10 percent of the bandwidth of input
  link
• The network administrators can guarantee that
  premium traffic will not starve the assured and
  best-effort traffic.
• Implementation
• Strict priority
• WFQ

9
Differentiated Services 4/4

• Example of End-to-End Service Delivery
• Delivery of Premium Service with a Dynamic SLA

10
MPLS 1/3

• MPLS network
• Label

11
MPLS 2/3

• MPLS network operation
• Existing routing protocol (e.g., OSPF, ISIS)
  establish reachability to destination networks.
• Label Distribution Protocol (LDP) established tag
  to destination network mapping.
• Ingress label switch router receives packets,
  performs layer 3 value-added services, and tags
  packets.
• Core LSR switches packets by using label swapping
• The penultimate hop to the destination removes
  the label and sends an unlabeled packet towards
  the last hop LER
• Egress LER switches an unlabeled packet based on
  its destination IP address.

12
MPLS 3/3

• Advantages
• Provides faster packet classification and
  forwarding
• Provides an efficient tunneling mechanism.
• Service Architecture based on MPLS

13
Traffic Engineering and Constraint-Based Routing
1/3

• What is Traffic Engineering?
• The progress of arranging how traffic flows
  through the network so that congestion caused by
  uneven network utilization can be avoided.
• What is Constraint-Based Routing?
• The tool for making the traffic engineering
  process automatic.
• To select routes that can meet certain QoS
  requirement
• To increase utilization of the network
• In order to do constraint-based routing, router
  needs to distribute new link state information
  and to compute routes based on such information.

14
Traffic Engineering and Constraint-Based Routing
2/3

• Constraint-Based Routing
• Distribution of Link State Information
• Approach is to extend the link state
  advertisements of protocols
• To reduce the frequency of link state
  advertisements
• When there are topology or significant band
  changes
• To use hold-down timer
• Route Computation
• Metrics
• monetary cost, hop count, bandwidth, reliability,
  delay, and jitter
• The routing table computation algorithms and the
  complexity of such algorithms depend on the
  metrics chosen for the routes.
• Pros and Cons
• Pros is to improve network utilization
• Cons is to increase overhead, routing table size,
  and other resource.

15
Traffic Engineering and Constraint-Based Routing
3/3

• The position of Constraint-based Routing in the
  QoS framework
• with Differentiated Services
• Constraint-based routing is to select the optimal
  routes for flows.
• It is not to replace differentiated services.
• Its result help differentiated services be better
  delivered.
• with RSVP
• RSVP reserves resources, but depends on
  constraint-based or dynamic routing to determine
  the path.
• with MPLS
• Constraint-based routing can better compute the
  routes for setting up LSPs.

16
A Comparison of ATM Networks to Router Networks

• ATM networks
• ATM networks gives the idea for router networks
  to make DiffServ and MPLS.
• ATM networks have the advantage, which includes
  faster data forwarding, QoS service.
• However, ATM networks have too overhead because
  of large cell header.
• Router networks can also provide QoS and traffic
  engineering with DiffServ and MPLS.

17
Theories and Models for Internet Quality of
Service
18
Outline

• Network Calculus
• Architectures for Scalable QoS Support
• Statistical Guarantees
• QoS Guarantees for TCP-Dominated Traffic
• Application-based QoS Control
• Challenges for the Future

19
Network Calculus

• Introductory Example The Shaper
• Arrival Curves
• a(t) is the amount of data that can be observed
  on the flow over any time windows of duration t.
• a(t) rt b (r rate, b burst)
• Shaper
• used to force a flow to satisfy some arrival
  curve constraint.
• stores incoming bits in a buffer and delivers
  them in such a way that the resulting output is
  s-smooth. (s is a token bucket constraint)
• Min-Plus Convolution
• (f?g)(t) inf0st(f(s)g(t-s))
• (f?g)?h f?(g?h) f?g?h
• f?g g?f
• The flow with a-smooth and R(t), defined as the
  number of bits observed from an arbitrary time
  origin up to time t.
• R R?a ? R R?a

20
Network Calculus

• Introductory Example The Shaper
• I/O Characterization of Shaper
• R is the output of the shaper. It must satisfy
  the constraint
• R R ? the output derives from the
  input after buffering
• R R?s ? it is s-smooth
• R R?s
• Consequences
• R?s (R?s)?a (R?a)?s R?s R
• Intserv and Service Curves
• Guaranteed Service
• Perform a reservation during a flow setup phase
• conform to an arrival curve of the form a(t)
  min(MCt, rtb) (T-SPEC) C the bit rate of
  link
  M the packet size
• All routers along the path accept if they are
  able to provide a service guarantee and enough
  buffer for loss-free operation.

21
Architectures for Scalable QoS Support

• Approach
• Class-based aggregate scheduling
• Adopted in DiffServ
• Dynamic packet state
• Control state information necessary for packet
  scheduling is carried in packet headers
• Core routers perform simple per-packet state
  update.
• As a result, using the dynamic packet state
  approach, per-flow end-to-end QoS guarantees
  similar to those provided by IntServ can be
  supported without per-flow management at core
  router

22
Dynamic Packet State

• What is Dynamic Packet State?
• Control state information is carried in data
  packets and updated at core routers for
  scheduling purposes.
• Virtual Time Reference System (VTRS)
• Scheduling framework-per hop end-to-end
  property characterization
• Three components
• packet state,edge traffic conditioning,reference/u
  pdate mechanism
• packet virtual time stamps-updated in core router
  using state in the packets-so core stateless
• edge traffic conditioning
• ensures packets are not injected into core at
  rate exceeding reserved rate.
• Packet state
• rate delay pair(rj,dj)
• virtual time stamp(wj,kgta1j,k)
• adjustment(dj)

23
Dynamic Packet State

• Conceptual Network Model (VTRS)

24
Dynamic Packet State

• Edge Conditioner and its effect (VTRS)

25
Dynamic Packet State

• Reference/Update mechanism (VTRS)
• Equipped in each core router
• The two properties of Virtual Time Stamp
• virtual spacing property wij,k1-wij,kgtLjk1/rj
  
• reality check aij,klt wij,k
• These two properties ensure that the end-to-end
  delay experienced by packets of a flow across the
  network core is bounded.
• Update Rule
• Wi1j,k Vij,k ? i p i wij,k dij,k ? i
  p i
• Core Stateless Packet Scheduling (VTRS)
• Using the notion of error term, delay guarantees
  can be characterized ? Not mandate scheduling
  mechanisms
• Stateless at core router

26
Dynamic Packet State

• Scalable Control Plane Operations
• Support for performance guarantees requires
  control and management of network resources.
• Approach
• Lightweight Signaling
• Conveys resource reservation to core routers.
• Aggregating a large number of individual RSVP
  requests
• Thus, reduce the number of request messages on
  backbone link
• Endpoint/Edge Admission Control
• Perform admission control based on measured
  resource availability information via probe
  packets.
• Eliminates the signaling protocols and QoS
  reservation messages.
• Centralized Bandwidth Broker
• Perform admission control, resource provisioning,
  and other policy decisions.

27
Statistical Guarantees

• Approach
• QoS guarantees may be given with some probability
  rather than on a deterministic basis.
• Model-Based Approaches
• A large body of work exists on computing loss and
  delay probabilities ? user flows satisfy some a
  priori traffic model.
• Better Than Poisson/MTU and Negligible Jitter
• The starting point for the analysis is the queue
  length.
• Approaches Based Only on Independence at Network
  Access
• Hoeffding Bounds
• Apply to the sum of a collection of independent,
  bounded random variables, assuming that the
  expectation of the sum is known.

28
QoS Guarantees for TCP-Dominated Traffic

• Elastic network calculus
• Traffic is transported by the Transmission
  Control Protocol(TCP)
• TCP congestion control follows an
  Additive-Increase-Multiplicative-Decrease(AIMD)
• Models for Expected Rate, Delay, and Loss of TCP
  Traffic
• Sending Rate
• P the average packet drop probability
• R the average round-trip time
• M the average size oif TCP packets
• a a small constant.

29
Application-based QoS Control

• This mechanisms rely on one or both of the
  following simple ideas
• the introduction of application-level routing and
  caching within the network
• the introduction of redundancy and quality
  adaptation to deal with end-to-end loss and delay
  variations.
• Application-Level Caching and Routing
• Cache is close to the clients ? low startup
  latency
• Application-level multicast algorithm is
  developed
• Redundancy and Quality Adaptation
• Priority encoding transmission (PET)
• different levels of protection were provided to
  I, P, and B frames in accordance to their
  importance to the application.
• This might of might not result in degraded
  quality as perceived by the application, because
  if all applications traversing the congested part
  of the network use this technique, then they will
  all observe higher loss rate.

30
Application-based QoS Control

• Adaptive Redundancy and Quality for Audio
  Applications
• Monitor network behavior (e.g. loss rate, delay
  jitter) this could result in periodic reports
  to the application or reports triggered by
  notable changes in network conditions.
• Increase/decrease redundancy level as a
  consequence of changes in network behavior this
  would include changes in encoding qualities.

31
Challenges for the Future

• The factors to hinder the deployment of Internet
  QoS
• Not technical But economic and political
• Challenges for the Future
• The tradeoff between complexity and efficiency in
  network models in open research area.
• The complexity of Internet QoS depends on the
  amount of computation needed to predict
  performance of new or existing traffic.
• Therefore, the management system for providing
  QoS guarantees in a sizable network is likely to
  be complex and expensive to build and to manage.