Quality of Service

1
Quality of Service
2
Quality of Service Requirements (1)
Arrival Offset Graph
Sampled Audio
Playout Point
Playout Buffer must be small for interactive
applications

• Real-time applications
• Interactive applications are sensitive to packet
  delays (telephone)
• Non-interactive applications can adapt to a wider
  range of packet delays (audio, video broadcasts)
• Guarantee of maximum delay is useful

3
Quality of Service Requirements (2)
Document is only useful when it is completely
received. This means average packet delay is
important, not maximum packet delay.
Document
Document

• Elastic applications
• Interactive data transfer (e.g. HTTP, FTP)
• Sensitive to the average delay, not to the
  distribution tail
• Bulk data transfer (e.g. mail and news delivery)
• Delay insensitive
• Best effort works well

4
Discussion

• What is the problem?
• Different applications have different delay,
  bandwidth, and jitter needs
• Some applications are very sensitive to changing
  network conditions the packet arrival time
  distribution is important
• Solutions
• Make applications adaptive
• Build more flexibility into network

5
RSVP Overview

• What is RSVP?
• Method for application to specify desired QoS to
  net
• Switch state establishment protocol (signaling)
• Multicast friendly, receiver-oriented
• Simplex reservations (single direction)
• Why run RSVP?
• Allows precise allocation of network resources
• Guarantees on quality of service
• Heterogeneous bandwidth support for multicast
• Scalable (?)

6
RSVP Design Criteria

• Heterogeneous receivers (multicast)
• Varying bandwidth needs
• Merging of resource reservations
• Dynamic membership
• Minimize control protocol overhead
• Soft state in routers
• Reservations timeout if not refreshed
  periodically
• Adapt to routing changes gracefully reestablish
  reservations

7
Protocol Independence

• RSVP designed to work with any protocol
• Protocol must provide QoS support
• Examples ATM, IP with Integrated Services
• Integrated Services
• Defines different levels of packet delivery
  services
• Defines method to communicate with applications
  Flowspec

8
Integrated Services Model

• Flow specification
• TSpec describes the flows traffic
  characteristics
• Rspec describes the resources requested from the
  network
• Routing
• Admission control
• Policy control
• Resource reservation
• Packet scheduling

9
RSVP Functional Diagram
Host
Router
RSVPD
D A T A
DATA
DATA
10
What is a flow?

• Equivalent packets by some classification
• RSVP Set of packets traversing a network element
  that are all covered by the same QoS request
• Packet classifier determines which packets belong
  to which flows
• IPv6 includes a flow label to ease classification
  
• Source/Dest IP and Source/Dest TCP port

11
Client Traffic Shaping

• Issue Need traffic shaping to meet allocated
  resources
• Source promises that data traffic will conform to
  a particular shape
• Why describe and shape traffic?
• Network knows what to expect, can manage traffic
  better
• Better admission control decisions
• Network can police flows
• Bursty traffic costly on scheduler, network

12
Traffic Shaping Example
Data Queue
Flow 1
Flow 2
Data Queue
13
Traffic Shapers

• Simple leaky bucket
• Isosynchronous flow regular intervals between
  packets
• Token bucket
• Bursty flow

14
Simple Leaky Bucket
Data
b
b bucket size r rate data is sent onto network
r

• Sends data at fixed intervals onto network
• Bursts bigger than b are discarded
• Traffic never injected faster than r
• Can be used with cells or datagrams

15
Token Bucket
r
b bucket size in tokens r rate tokens are
added to bucket
b
Data Queue
Data

• Sends bursty traffic onto network
• Bucket filled with tokens at rate r
• Data transmitted when enough tokens exist
• Allows bursts, but enforces upper bound

16
Restrictions on Reservations

• Admissions
• Is bandwidth available?
• Policy
• Service guarantees give preferential access to
  network bandwidth
• Permissions
• Pricing issues
• What are the policies of nodes on the path?
• Policy data represents a scaling and security
  issue

17
Resource Reservation Model

• Senders advertise using flowspecs
• RSVP daemons forward advertisements to receivers,
  update available bandwidth, minimum delay
• Receivers reservations use flowspec, filterspec
  combination (flow descriptor)
• Sender/receiver notified of changes
• Reservations are merged in multicast case

18
RSVP Service Types

• Controlled load
• Guaranteed service

19
Controlled Load Service

• Definition
• Service that gives a flow the QoS it would
  receive if the network was unloaded.
• Statistical guarantee
• No delay bounds
• Motivation
• Support delay sensitive applications
• Minimal functionality

20
Controlled Load Requirements

• Admission Control
• Ensure adequate resources are available
• Link bandwidth
• Computational power for processing flow
• Adequate buffer space to handle bursty traffic
• Operation
• Little or no average packet queuing delay
• Little or no congestion loss
• Time period significantly longer than burst time

21
Guaranteed Service

• Definition
• Service providing guaranteed delay and bandwidth
• Firm guarantee on end-to-end queuing delays
• Delay
• Two parts
• Fixed delay transmission delays, etc
• Queuing delay
• Queuing delay is a function of token bucket and
  data rate
• Often assumed that application has no control
  over delay
• Application can choose queuing sizes

22
RSVP UDP Reservation (1)
R3
R2
R4
R1
Host B 128.32.32.69
Host A 24.1.70.210
R5
23
RSVP UDP Reservation (2)
R3
R2
R4
PATH
R1
PATH
Host B 128.32.32.69
PATH
PATH
Host A 24.1.70.210
R5
24
RSVP vs DiffServ
RSVP

• Per-flow service state at every hop

• Scalability problems

• Focus on multicast

BB
BB
DiffServ

• Abstract/manage each clouds resources (BBs)
• Packets colored with behavior
• Focus on aggregates not flows
• Policing at edge to get services

25
DiffServ Overview

• Exploits edge/core distinction for scalability
• Applications contract for specific QoS profiles
• Policing at network periphery
• A few simple, differentiated per-hop forwarding
  behaviors (PHBs)
• Indicated in packet header
• Applied to PHB traffic aggregates
• PHBs policing rules range of services
• Clouds contract for aggregate QoS traffic
  profiles
• Policing at cloud-cloud boundary
• Supports simple, bilateral business agreements

26
DiffServ Architecture
Bandwidth Brokers (perform admissions control,
manage network resources, configure leaf and
edge devices)
Destination
Source
BB
BB
Core routers
Core routers
Ingress Edge Router (classify, police, mark
aggregates)
Egress Edge Router(shape aggregates)
Leaf Router (police, mark flows)
27
Example Service 1 Premium

• Contract leased line emulation at aspecified
  peak rate
• PHB forward me first (EF)
• Policing rule drop out-of-profile packets
• On egress, clouds must shape EF aggregates to
  mask induced burstiness

28
Example Service 2 Assured

• Contract network looks lightly-loaded for
  traffic within a specified rate and burst
  profile
• PHB ? drop me last (AF)
• Policing rule remark out-of-profile packets to
  have higher drop probability
• AF is actually a family of PHBs
• 4 independent AF classes
• 3 drop preference levels within each class
• Traffic within a class shares single queue
• On cloud egress, clouds may need to shape AF
  aggregates to mask induced burstiness

29
Example Service 3 CoS

• Contract better service relative to the
  schmucks who pay less
• PHB drop the lower classes first (AF)
• Policing rule drop or remark out-of-profile
  packets
• Olympic classes of BE service
• Gold
• Silver
• Bronze

30
DiffServ The Three Big QoS Problems

• Applications Framework supports a broad range of
  services depending on PHB and configuration of
  policers
• Scalability Simplicity of PHBs pushing
  smarts towards edge, lets core routers be
  simple, dumb, and fast, but still supports QoS!
• Interoperability
• PHBs suggest but do not imply implementations
• QoS through concatenations of simple bilateral
  SLAs
• Administratability also a big win

31
ATM ABR Service
RM cells
RM cells
H1
S1
S2
S3
H2
Source
V
irtual
V
irtual
Destination
destination
source
32
RSVP versus ATM (Q.2931)

• RSVP
• receiver generates reservation
• soft state (refresh/timeout)
• separate from route establishment
• QoS can change dynamically
• receiver heterogeneity
• ATM
• sender generates connection request
• hard state (explicit delete)
• concurrent with route establishment
• QoS is static for life of connection
• uniform QoS to all receivers

33
MPLS BUILT ON STANDARD IP
47.1
1
2
1
3
2
1
47.2
3
47.3
2

• Destination based forwarding tables as built by
  OSPF, IS-IS, RIP, etc.

34
IP FORWARDING USES BY HOP-BY-HOP CONTROL
47.1
1
IP 47.1.1.1
2
IP 47.1.1.1
1
3
2
IP 47.1.1.1
1
47.2
3
47.3
2
35
MPLS Label Distribution
1
47.1
3
3
2
1
1
2
47.3
3
47.2
2
36
Label Switched Path (LSP)
1
47.1
3
3
2
1
1
2
47.3
3
47.2
2
37
EXPLICITLY ROUTED OR ER-LSP
B
C
A
- ER-LSP follows route that source chooses. In
other words, the control message to establish the
LSP (label request) is source routed.
38
EXPLICITLY ROUTED LSP ER-LSP
1
47.1
3
3
2
1
1
2
47.3
3
47.2
2