Transcript Reliable multi-media services for NGNs

An Overview of QoS for
Multi-Service IP Networks
Peter Thompson
Chief Scientist
U4EA Technologies Ltd.
Internet Telephony Conference and EXPO East 2006
1
Overview
• Understand the reasons for performance
variability in IP networks
• Examine the techniques to control this in the
network core
• See why these techniques are less useful in the
access part of the network
• Explore how QoS can be achieved in parts of the
network where contention is likely
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
2
“Quality of Service” – one term, many meanings
Terminal
device
Customer
support
User Quality of Experience
Application
performance
Network
performance
Application
servers
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
3
Network performance: the problem
Every service is carried in a
stream of packets:
•Flexible: easy to add
services
•Efficient: streams share
network resources
Limited resources:
•Bandwidth
•Packet buffers
•Packet Service
Network
Element
Sharing resources
causes performance to
vary:
•Not all streams can see
an empty network
•Some streams’ delivery
will vary drastically
•Packet streams are
often bursty
•Bursts can overload
network resources
•Call this ‘contention’
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
4
Avoiding contention – vanilla IP
• Packets routed independently
• Congestion changes routing – upsets QoS
• Congestion point moves, causes route flap
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
5
Avoiding contention - MPLS
Label
switch
path
• MPLS gives more control
• Routing decision taken once per flow
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
6
Optimising route selection
Stream X: 5
Stream Y: 5
Stream Z: 6
A
10
12
8
10
B
Choice 1:
Choice 2:
X via A, Y via B
X via A, Y via A
Z cannot be routed
Z via B is OK
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
7
Allocating bandwidth
Bandwidth
Bursts where loss and/or
delay become excessive
Time
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
8
Allocating bandwidth
Bandwidth
Bursts for which loss and
delay are tolerable
Allocation above
average to get
acceptable
performance
Time
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
9
Contention strikes again!
Core
Edge
Access
• Plenty of routes and bandwidth in the core
network
• Less so in the network edge
• Not at all in the access
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
10
Contention management
Use resources
efficiently
Differentiate
performance for
multiple types of
service
Keep average
Network
load
below
Element
30-40%
Various mechanisms:
•Policing
•Shaping
•Queuing
•Scheduling
Maintain consistent
performance under
saturation
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
11
Effective contention management
Use resources
efficiently
Differentiated
performance for
multiple service
classes
Network
Element
Deep analysis of the
scheduling problem –
use the degrees of
freedom
Guarantee worstcase performance
under saturation
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
12
Edge-to-edge network performance
∆Q
Edge/access:
• Effective
contention
management
• QoS under
saturation
© U4EA Technologies 2006
∆Q
∆Q
Core:
End-to-end:
• Contention
• Performance is
avoidance
the sum of the
‘∆Q’
• Route control
• Bandwidth allocation
Internet Telephony Conference and EXPO East 2006
13
Thank you! Any questions?
Peter Thompson
U4EA Technologies Limited
[email protected]
www.u4eatech.com
© U4EA Technologies 2006
Internet Telephony Conference and EXPO East 2006
14