MultiProtocol Label Switching

1
Multi-Protocol Label Switching
University of Southern Queensland
2
MPLS

• Packets have an additional field
• A stack of labels
• Labels lie between IP and link layer
• Routers use the labels instead of the IP destn
  field
• Routing is faster because labels are local to
  each router.

3
Label Switching
Label Switching Table of next label/hop for each
label

• Label Switching
• Table of next label/hop for each label

Label Switching Table of next label/hop for each
label

• Label table per incoming port.
• label -gt (new label, next link), OR
• label -gt pop

4
Label Switched Path (LSP)

• A label switched path is like a tunnel.
• At the end of an LSP, the label is popped.
• If there is another label, use this.
• If there is no label, use the router's RT, as
  usual.

5
Advantages of MPLS

• Much smaller tables
• forwarding rather than routing
• LSP's can have special treatment.
• LSP's might use a special technology, eg ATM,
  WDM.
• LSP's keep traffic classes separate.

6
Disadvantages of MPLS

• Additional complexity
• LSP's must be set up and managed.

7
Example Campus Router/Switch

• When a tcp connection is set up, create an entry
  in a forwarding table in the switch, one table
  for each incoming port.
• In this way, packets can be forwarded by the
  switch without the need for routing.
• No need for a label in this case.
• To generalise this idea, use labels.

8
History

• ATM was perceived to have an efficiency advantage
  due to the use of forwarding rather than
  routing.
• E.g. major Telco's have used ATM transport
  network, running IP over the top, for cost
  reasons.
• MPLS can achieve similar efficiency to ATM.

9
Present

• MPLS is currently proposed as a means to provide
  network wide VLAN's and/or VPN.
• MPLS provides virtual isolation of traffic
  without real segregation.
• (Segregation of traffic is very inefficient).

10
MPLS and WDM

• Wavelengths can also be treated as tunnels across
  the Internet.
• In this instance, the wavelength is operating
  like the label.
• Similarly, SDH paths can be regarded as LSP's.

11
Optimization / Efficiency

• Using an end-to-end LSP is very efficient as
  regards routing.
• But, if an LSP carries very little traffic,
  utilization of the path could be very low.
• If bandwidth must be reserved for the LSP, this
  will be inefficient.

12
How to Optimize

• Manage bandwidth reserved for each LSP
• If LSP bandwidth falls below a certain level,
  close it down.
• If a path through a higher layer network carries
  more than a certain threshold level of traffic,
  set up an LSP, or wavelength, or SDH path, to
  carry it.

13
Example

• Suppose I start a download of a large video from
  across the Internet.
• The mechanisms of dynamic GMPLS (Generalised
  MPLS, ie MPLS including WDM and SDH) will respond
  to this by setting up a wavelength, end-to-end.
• The efficiency of this way of handling this
  traffic is fantastic.
• In addition, if an LSP is set up, peformance will
  be better (less buffering, less congestion).

14
Why a label stack?

• In many cases, the stack will have just one
  entry.
• In some cases, the label will be implied rather
  than explicit.
• However, it can happen that one link in an LSP is
  really another LSP. In this case, we need the
  stack.
• The top label is the only one referenced, till
  that LSP ends. Then the label is popped off and
  the LSP underneath is followed.

15
Performance Differentiation

• It is conceivable that LSP's can also be given
  special, or different treatment, from each
  other.
• E.g. a wavelength will have no buffering and
  minimal delay. The same is true of an SDH path.
• An LSP through the Internet might not really have
  protection, however. This should not be assumed.