Internet Traffic Engineering

1
Internet Traffic Engineering

• Motivation
• The Fish problem, congested links.
• Two properties of IP routing
• Destination based
• Local optimization
• TE optimizing performance of operational
  networks.
• Reduce hot spots
• Improve resource utilization
• Approach carefully select the route for the
  flows.

2

• TE solutions
• Need mechanisms to explicitly manage the traffic
  inside a network
• The overlay model
• IP on top of ATM (or other link layer
  technologies)
• Use ATM VCs to construct a virtual network
  between edge nodes.
• Traffic distribution is managed by carefully
  mapping VCs to the physical network topology.
• MPLS explicit route
• Work well with BGP (BGP NEXT HOP information)

3

• Overlay model An example

A
B
G
F
E
H
D
C
4

• Overlay Model
• Some problems
• N square problem. a large ISP can be 20 PoPs with
  each has 10 edge nodes, how many explicit routes
  do we need to manage?
• Too many neighbors also increase the load on IP
  routers.
• Peer model
• Use native IP routing, but manipulate the link
  weights on the network.

5

• TE optimization objectives
• Different algorithms need to be developed for
  different optimization objectives
• Minimizing congestion and pack losses in the
  network
• Improving link utilization
• Minimizing the total delay experienced by packets
• Increasing the number of customers

6

• The architecture of a typical TE system

Route Computation
Demand Estimation
Data Repository
Graphical user interface
Topology and State discovery
Network Interface
7

• Topology and State Discovery
• Static information usually can be determined
  when the network is configured.
• Dynamic information TE need accurate up-to-date
  information about traffic and capacity on each
  link, such information must be maintained in some
  way.
• OSPF based schemes
• OSPF is widely deployed for IP routing
• LSAs in OSPF may not carry sufficient information
  for TE.
• Local and remote interface IP addresses (to
  distinguish between multiple links)
• Traffic engineering metrics
• Maximum bandwidth
• Maximum reservable bandwdith
• Unreserved bandwidth
• Resource classes.