QoS in the Internet Proposals and Prospects

1
QoS in the Internet Proposals and Prospects
2
Introduction

• For Many Years the Internet was primarily used
  for networking research. File transfer and email
  were the most popular applications They do not
  need any performance guarantee from the
  underlying network.
• The current Internet cannot provide any resource
  guarantees the service is best effort
• New applications such as VoIP, video
  conferencing, e-commerce apps are sensitive to
  network performance (e.g., delay and bandwidth
  guarantees).

3
Current State of Internet

• The internet service did not change by much
• Uses best-effort service model
• No guarantee of timeliness or delivery
• No service discrimination
• Bandwidth and network congestion problems
• Unpredictable network response time

4
What is QoS

• The capability to provide resource assurance and
  service differentiation so that delay, jitter or
  loss sensitive applications can perform
  satisfactorily is referred to as quality of
  service (QoS).
• Can be provided through relative prioritization
  of resource allocation to different flows/packets
  in the network.

5
Resource Allocation

• Many problems in the Internet come down to the
  issue of resource allocation.
• Packets get delayed or dropped because network
  resource cannot meet the traffic demands.
• A network consists of shared resources
  bandwidth, buffers, serving traffic from
  competing users.
• To support QoS, the network must allocate
  resources and decide who should get how much
  resources.

6
Integrated Services

• Based on per flow resource reservation.
• Applications must make a reservation before
  transmitting traffic.
• Application characterizes its traffic and
  resource requirement.
• Network uses routing protocol to find a path.
• A reservation protocol is used to install the
  reservation state along that path.

7
Integrated Services (contd)

• At each hop admission control checks whether
  sufficient resources are available to accept the
  new reservation.
• Resource reservation enforced by packet
  classification and scheduling mechanisms.
• Two service models are introduced guaranteed
  service and controlled load.
• Guaranteed service provides deterministic worst
  case delay
• Controlled load provides less firm guarantee
  its close to a lightly loaded best-effort
  network.

8
Integrated Services (contd)

• Overhead of setting up reservation is high.
• Scalability problem Backbone will have a large
  number of flows and keeping flow information is
  not feasible.

9
Differentiated Service

• Users traffic is divided into a small number of
  forwarding classes.
• For each forwarding class, the amount of traffic
  that users can inject is limited at the edge of
  the network.
• Edge of a differentiated Services network
  responsible for mapping packets to their
  appropriate forwarding classes.
• Packet classification is usually based on service
  level agreement.

10
Differentiated Service (Contd)

• Nodes at the edge of the network perform traffic
  policing to ensure conformance.
• Non-conforming traffic may be dropped, delayed or
  marked with a different forwarding class.
• Forwarding class is directly encoded into the
  packet header.
• Interior nodes use this info. to differentiate
  the treatment of packets.
• Does not require resource reservation.

11
Differentiated Service (Contd)

• Forwarding classes apply to traffic aggregates
  rather than individual flows.
• No scalability problem.

12
Diffserv Architecture

• Ingress routers
• Police/shape traffic
• Set Differentiated Service Code Point (DSCP) in
  Diffserv (DS) field
• Core routers
• Implement Per Hop Behavior (PHB) for each DSCP
• Process packets based on DSCP

DS-2
DS-1
Ingress
Egress
Ingress
Egress
Edge router
Core router
13
Multiprotocol Label Switching (MPLS)

• A short fixed length label is encoded into the
  packet header and is used for packet forwarding.
• When a label switch router (LSR) receives an MPLS
  packet, it uses incoming label to find the next
  hop and the corresponding outgoing label.
• In MPLS, the path a packet traverses is called
  label switched path (LSP).
• Network protocol independent

14
Multiprotocol Label Switching (MPLS)

• Works alongside existing routing technologies and
  provide a mechanism for explicit control over
  routing paths.
• Used for traffic engineering, guaranteeing QoS
  and VPN.

15
Traffic Engineering

• The basic problem Given a network and traffic
  demands, how can traffic flows in the network be
  organized so that an optimization objective can
  be optimized.
• Typically optimal operating point is reached when
  traffic is evenly distributed leads to min
  queuing delay and packet loss.
• This cannot be achieved through destination based
  IP routing
• Advanced routing techniques constraint-based
  routing are used.

16
Traffic Engineering (contd)

• Routes are computed with multiple constraints and
  aims for efficient resource utilization.
• Appropriate path selection with uniform traffic
  distribution and the congestion avoidance
  techniques improve the network performance
• MPLS can be used to achieve this goal.

17
Internet QoS Overview

• Integrated services
• Differentiated Services
• MPLS
• Traffic Engineering

18
State information

• No State Vs. Soft State Vs. Hard State

19
Integrated Services

• Early 1990 IETF started Inegrated Services
  working group to standardize a new resource
  allocation architecture.
• Based on a per flow resource reservation.
• Goal is to preserve the datagram model of
  IP-based networks and at the same time support
  resource reservation for real-time applications.

20
QoS Router
Queue management
Policer
Per-flow Queue
Scheduler
Classifier
shaper
Policer
Per-flow Queue
Per-flow Queue
Scheduler
shaper
Per-flow Queue
21
Basic Approach

• A set of mechanisms and protocols is used for
  making explicit resource reservation.
• To receive performance guarantee from the
  network, resource reservation must be set up
  before the application can start transmitting
  packets.
• Sender starts the setup of a reservation by
  sending characteristics and resource requirement
  of the flow.
• The network can accept the new application flow
  only if sufficient resource is there.
• Once reservation is setup successfully, the
  application can start sending data packets.

22
Key Components
QoS routing agent
Admission control
Reservation setup agent
Resource reservation table
Control plane
Flow identification
Packet scheduler
Data plane
23
Key Component (contd)

• Control Plane sets up resource reservation.
• Data plane forwards data packets based on
  reservation state.
• To setup reservation, app first characterizes its
  traffic flow and specifies QoS requirements
  referred to as flow specification
• The reservation setup request is then sent to the
  network.

24
Key Component (contd)

• Router upon getting the request, interacts with
  QoS routing agent to find the next hop.
• It then coordinates with the admission control
  module to determine if there are sufficient
  resources to meet the requested resources.
• Once reservation set up is successful, the
  information for the reserved flow is installed
  into the resource reservation table.
• Info. in the resource reservation table is used
  to configure the flow identification module and
  the packet scheduling module in the data plane.

25
Route Selection

• IntServ does not specify any route selection of
  its own.
• It relies on existing routing protocols to
  forward its control packets further.
• Obviously a more efficient routing protocol which
  can find a path that is likely to have sufficient
  resources is desired.

26
Reservation Setup

• To setup reservation, a reservation set up
  protocol is needed that goes hop by hop along the
  path to install the reservation state in the
  routers.
• The reservation protocol must also deal with
  changes in the network topology.
• In IntServ, RSVP has been developed as the
  resource reservation protocol.

27
Admission Control

• In order to provide guaranteed resources for
  reserved flows, a network must monitor its
  resource usage and admit a new flow only if it
  has sufficient resource.
• It has two functions to determine if a new flow
  reservation can be set up based on the admission
  control policies and to monitor and measure the
  available resources.

28
Flow Identification

• Router must examine every incoming packet and
  decide whether the packet belongs to one of the
  reserved flows.
• IP flow is identified by src addr, dest addr,
  proto ID, src port, dst port five-tuple.
• These five fields of the incoming packet is
  compared against the five-tuple of all the flows
  in the reservation table for flow identification.

29
Packet Scheduling

• Packet scheduler responsible for resource
  allocation
• Directly affects delay, jitter and packet loss
• Primary task is to select a packet to transmit
  when outgoing link is ready such that the QoS
  promised to flows is provided

30
Service Models

• Describe interface between the network and its
  users.
• IntServ has standardized two basic service
  models
• Guaranteed service
• Controlled load service

31
Guaranteed Service

• Provides guaranteed bandwidth and strict bounds
  for delay.
• Intended for apps that require highest assurance
  on bw and delay mission critical apps,
  intolerant playback apps.
• Can be viewed as a virtual circuit with
  guaranteed bw.
• Provides bounds on maximal queuing delay.

32
Controlled load service

• Strict bw assurance and delay bound comes at a
  price resources have to be reserved for the
  worst case.
• For some apps a service model with less strict
  guarantees and lower cost would better serve
  their needs.
• End-to-end behavior somewhat vague.
• A very high percentage of packets will be
  successfully delivered by the network to the
  receivers.
• The transit delay experienced by a very high
  percentage of packets will not greatly exceed min
  delay.

33
RSVP

• A resource reservation protocol defined under
  IntServ.
• Used by hosts to communicate service requirements
  to the network and by routers in the network to
  establish reservation state along a path

34
Protocol Overview (Contd)

• RESV must follow the exact same reverse path
  upstream.
• They create reservation state in each node along
  the paths
• After receiving RESV msg sender can start sending
  data packets.

35
DiffServ

• Differentiated Services (DiffServ) is proposed by
  IETF as a scalable QoS solution for the next
  generation Internet.
• Developed for relatively simple, coarse methods
  of providing different levels of service for
  Internet traffic.
• Divides traffic into a small number of classes
  and allocates resources on a per class basis.
• Core of a diffserv network distinguishes between
  small number of forwarding classes rather than
  individual flows.

36
DiffServ (cont.)

• Complex per-flow classification and scheduling
  used in intServ (causes scalability) not needed.
• Operates on a per-hop behavior (PHB) basis
• Classifies packets by marking the headers
  Routers discriminate packets based on their
  markings
• Packet marking is done on the basis of a service
  level agreement (SLA) between the host and the
  ISP
• Provides service assurances but no QoS guarantee

37
Basic Approach

• Traffic is divided into a small number of groups
  called forwarding classes
• Forwarding class that a packet belongs to is
  encoded into a field in the IP packet header.
• Each forwarding class represents a predefined
  forwarding treatment in terms of drop priority
  and bandwidth allocation.

38
Basic Approach (cont.)

• Achieves scalability by implementing traffic
  classification and conditioning functions at
  network boundary nodes
• Classification involves mapping packets to
  different forwarding classes.
• Conditioning checking whether traffic flows
  meet the service agreement and dropping/remarking
  non-conformant packets.
• Interior nodes forward packets based solely on
  the forwarding class.

39
Per Hop Behavior (PHB)

• Forwarding treatments at a node
• Each PHB is represented by a 6-bit value called
  DSCP
• All packets with the same code points are
  referred to as a behavior aggregate (BA) and they
  receive the same forwarding treatment.
• May describe forwarding behavior in either
  relative or absolute terms
• Minimal bw for BA absolute term
• Allocate bw proportionally relative
• Typically implemented by means of buffer
  management and packet scheduling.

40
Differentiated Services Field

• Uses 6 bits in the IP header to encode forwarding
  treatment
• These 6 bits are those out of the IP TOS field (8
  bits long)
• DiffServ redefines existing IP TOS field to
  indicate forwarding behavior
• Replacement field, called DS field supersedes
  existing definition of TOS
• First 6 bits used as DSCP to encode the PHB,
  remaining 2 bits are currently unused (CU).

41
Differentiated Service (DS) Field
0
5
6
7
DS Field
0
4
8
16
19
31
Version
HLen
TOS
Length
Identification
Flags
Fragment offset
IP header
TTL
Protocol
Header checksum
Source address
Destination address
Data

• DS filed reuse the first 6 bits from the former
  Type of Service (TOS) byte to determine the PHB

42
Assured Forwarding (AF)

• The basic idea came from RIO scheme
• In RIO scheme packets are marked as in or out
• During congestion, out packets are dropped first
  in/out bit indicates drop priorities
• AF standard extended the basic in or out marking
  in RIO into four forwarding classes and within
  each forwarding class, three drop precedences

43
Assured Forwarding (AF) (cont.)

• Customers can subscribe to the service built with
  AF forwarding class and their packets will be
  marked with appropriate AF DSCPs.
• Drop priorities within each forwarding class are
  used to select which packets to drop during
  congestion
• When backlogged packets from an AF forwarding
  class exceed a specified threshold, packets with
  highest drop priority is dropped first, then
  packets with lower drop priority

44
Mechanism for assured service - summary
45
Expedited Forwarding (EF)

• Proposed to characterize a forwarding treatment
  similar to that of a simple priority queuing.
• Forwarding treatment of traffic aggregate must
  equal or exceed a configurable rate
• Should receive this rate independent of load of
  other traffic passing through the node
• Provides low delay and low loss service

46
EF implementation

• Several queuing mechanisms can be used to
  implement EF PHB
• Priority queuing with token bucket
• Priority of EF traffic should be highest in the
  system
• Token bucket is used to limit the total amount of
  EF traffic so that other traffic will not starve
• WFQ can be used such that weight assigned to EF
  traffic has relative priority than other traffic

47
DiffServ Summary
48
Functionality at DiffServ Routers
49
Proportional QoS

• Using a proportional QoS model, we not only
  guarantee that a higher priority class receives
  better service, but we also quantify the
  differentiation between different classes

50
Proportional Differentiation

• Definition
• If qi is the QoS metric of interest, and si is
  the differentiation factor for class i, we have

For example Given two classes 1 and 2, and
the QoS metric is packet loss rate, s11
s22, the packet loss rate of class 2 should
be twice that of the loss rate of class 1.
51
Proportional Differentiation

• Pros
• Controllable
• Differentiation level between service classes can
• be controlled by network operator
• Predictable
• Performance of higher classes is consistently
• better than the performance of lower Class even
• in short time scale

52
Proportionally differentiated packet delay

• Waiting Time Priority (WTP) Scheduling

One packet need to be scheduled
Class 0
Class 1
On-line priority measurement is done
Class N
53
Proportionally differentiated packet delay
Waiting Time Priority (WTP) Scheduling
54
Proportionally differentiated packet delay

• Wait Time Priority (WTP) Scheduling
• Suppose class i is backlogged at time t, and that
  wi(t) is the head waiting time of class i at t
• We have normalized head waiting time of class i
  at t as
• When a packet need to be scheduled, a backlogged
  class j is selected for

55
Performance
Proportional average packet delay
56
Proportionally differentiated loss rate

• Buffer Management

On-line priority measurement is done
Class 0
Class 1
One packet arrives
Class 2
Total buffer size 20
57
Proportionally differentiated loss rate

• Buffer Management

Class 0
Class 1
Class 0 has the lowest priority
Class 2
Total buffer size 20
58
Proportionally differentiated loss rate

• Proportional Loss Rate (PLR) dropper
• Suppose there are two counters for each class i,
  counter ai records packet arrival history of
  class i, counter di records packet drop history
  of class i
• We have normalized packet loss rate of class i
  as
• When a packet needs to be dropped, a backlogged
  class j is selected for

59
Performance
Proportional packet loss rate
60
Architecture I
Parameters
Timer
Waiting time recorder
Dropper
Scheduler
Packet arrival
Drop one packet
Schedule a packet
61
MPLS
62
Why MPLS?

• MPLS stands for Multi-Protocol Label Switching
• Goals
• Bring the speed of layer 2 switching to layer 3
• May no longer be perceived as the main benefit
  Layer 3 switches
• Resolve the problems of IP over ATM, in
  particular
• Complexity of control and management
• Scalability issues
• Support multiple layer 2 technologies

63
Basic Idea

• MPLS is a hybrid model adopted by IETF to
  incorporate best properties in both packet
  routing circuit switching

MPLS
ATM Switch
IP Router
64
Basic Idea (Cont.)

• Packets are switched, not routed, based on labels
• Labels are filled in the packet header
• Basic operation
• Ingress LER (Label Edge Router) pushes a label in
  front of the IP header
• LSR (Label Switch Router) does label swapping
• Egress LER removes the label
• The key establish the forwarding table
• Link state routing protocols
• Exchange network topology information for path
  selection
• OSPF-TE, IS-IS-TE
• Signaling/Label distribution protocols
• Set up LSPs (Label Switched Path)
• LDP, RSVP-TE, CR-LDP

65
MPLS Operation
66
Label switching technologies

• Main objectives a forwarding technique
• Improve Internet forwarding technology
  performance
• As a result, scale the (WAN) Internet
  infrastructure
• Several pre-standard industry flavors
• "Tag switching", "IP switching", "Fast IP", ...
• IETF Standardization
• First drafts produced mid 97

67
Label Switching

• Uses concept of edge and core where
• conventional routing done at the edge
• switching
• (i.e.make forwarding decision not based on
  destination address)
• done in the core

Conventional Routing at edge
Switching in core
68
Labels
7423

• In core, forwarding decision based on logical
  forwarding references
• Logical forwarding reference called Label (1)

H
7423
R
3
7423
R
7
R
7423
label
69
Label Switching Routers
7423

• In the core nodes capable of switching on labels
  called
• Label Switching Routers (LSR)
• LSR is a specific device
• neither regular router
• nor regular ATM switch

Specific device
H
7423
LSR
3
7423
LSR
7
7423
Label Switching Domain
70
Label Switching rationale

• Select output port at "hardware speed"
• Simple look-up
• Using an Index
• short
• of fixed length
• in fixed position
• at beginning of packet
• with no internal structure

Label
3
71
Relationships with QoS

• Entry may also contain information about what
  resources all the pkts carrying this label may
  use
• e.g.
• outgoing queue
• drop reference level

Label
3
Outgoing port
0
1
2
3
4
5
6
7
72
Label local significance

• No reason index (labels) to forwarding tables be
  the same in all nodes
• Label to be agreed by two communicating LSRs
• Have strictly local significance, as
• X.25 virtual channels numbers
• ATM VCIs and VPIs

LSR
LSR
3
7
2
73
Label Switched Path (LSP)

• Label Switched Path (LPS)
• concatenation of labels
• constitutes, from node to node, the path followed
  by all pkts carrying those labels

LSR
LSR
3
7
2
74
FEC and Label Switched Paths

• Forwarding Equivalence Classes
• Pkts following an LSP form an FEC
• a set of pkts forwarded in the same manner
• An FEC maps to a label

Example of Label and FEC binding
2
5
LSR
LSR
LSR

• FEC
• All packets forwarded in the same manner by that
  router
• Form a Class (FEC)
• carry same label 3 on this section

75
Further FEC partition

• Further FEC partition possible (but not
  mandatory)
• If packets have different QoS requirements

Example of further partition of an FEC
label x
LSR
LSR
label y
76
Label swapping

• Label swapping in a core Label Switching Network
  (LSN) is similar to forwarding over ATM or X.25
  virtual circuits
• replace incoming label with outgoing label for
  next hop
• Implement QoS/priority as appropriate
• forward to output port
• In addition, LSRs must check if pkt has reached
  final destination (egress point)

77
Core MPLS Switching Routers

• MPLS switches may be
• Routers enhanced with MPLS functionality
• If IP, can be
• IPv4 router
• IPv6 router
• ATM switches (1)
• supporting MPLS protocols
• implementing regular layer 3 routing
• called ATM-LSR or LS-ATM

Core switching
R
R
R
MPLS-capable Routers
Core switching
ATM
ATM
ATM
R
R
R
MPLS-capable and routing-capable ATM switch
78
Where to code the label?

• Labeling a pkt either
• overwriting an existing field in a layer 2
  header, or
• inserting an extra header between layer 2 and
  layer 3 headers - the MPLS header

Layer 2 header
Layer 3 header
Label
label set in existing field
Layer 2 header
Layer 3 header
MPLS header
Label
label inserted
79
Example of layer 2 label
Core switching
ATM
ATM

• When Label Switching Routers are ATM-LSR ...
• then, ...
• Label carried in the VPI/VCI field (28 bits)

ATM
R
R
R
Layer 2 header
Layer 3 header
Label
80
Traffic Engineering MPLS
81
Traffic Engineering

• Concerned with the performance optimization of
  operational networks
• Main objective is to reduce congestion hot spots
  and improve resource utilization across the
  network through carefully managing the traffic
  distribution inside the network
• Cost savings that results in more efficient use
  of resources (e.g. bw) helps to reduce overall
  cost of operation for service providers.
• IP routing is based on destination and used
  simple metrics such as hop count
• IP routing can lead to poor resource utilization

82
The Fish Problem
D
A
G
F
C
Tail
Head
B
E
83
The Fish Problem (cont.)

• There are two paths from A and B to G.
• But only one of the two paths (shortest path)
  will be used for traffic
• Leads to unbalanced traffic distribution
• Problem caused by two properties of IP routing
• IP routing is destination based. Thus for each
  destination network there is typically only one
  path in the routing table traffic distribution
  tends to be unbalanced

84
The Fish Problem (cont.)

• Decision making in current routing is based on
  local optimization any node simply selects a
  path that is best from its own perspective. It
  does not take into account the overall system
  objective and have a global view of the network
  in terms of traffic distribution

85
Optimization Objectives

• The main aim of TE is to improve network
  performance through optimization of resource
  utilization in the network.
• Common optimization objectives are
• Minimizing congestion and packet losses in the
  network
• Improving link utilization
• Minimizing total delay experienced by packets
• Increasing number of customers with the current
  assets

86
Optimization Objectives (cont.)

• ISPs would like to avoid hot spots in the network
• Mathematically means minimize the maximum link
  utilization
• Means lower total delay and loss
• Leaves more space for future traffic growth since
  available bandwidth is maximized

87
Constraint-Based Routing

• Conventional IP routing is based on an algorithm
  that optimizes a particular scalar metric
• With constraint based routing path is optimal
  w.r.t. some scalar metric, at the same time it
  does violate a set of constraints
• Performance constraint
• Path with certain minimum available bw
• Administrative constraints
• Path that excludes certain links in the network

88
Constraint-Based Routing (cont.)

• Plain IP routing cannot support constraint based
  routing
• Constraint-based routing requires path
  calculation at the source
• Because different source may have different
  constraints for a path to the same destination
• Constraints associated with a particular source
  router are only known to that router
• In plain IP routing paths are computed in a
  distributed fashion by every router does not
  take into account constraints of different
  sources

89
Constraint-Based Routing (cont.)

• When a path is determined by the source,
  forwarding along such a path cannot be provided
  using the destination-based IP forwarding
• Path computation at the source needs to have
  information about attributes associated with
  individual links (e.g. link utilization).
• There is no mechanism to distribute this
  information in the network through plain IP
  routing
• IP routing protocol can be augmented to support
  these functionality