Routing and Wavelength Assignment VERSUS Wavelength Converter Placement in All-Optical WDM Networks

1
Routing and Wavelength Assignment VERSUS
Wavelength Converter Placement in All-Optical
WDM Networks

• Xiaowen Chu Bo Li
• Department of computer Science
• Hong Kong Univ. of Science Technology
• Clear Water Bay, Kowloon, Hong Kong

2
Outline

• The all-optical backbone
• Circuit-Switched wavelength-routed WDM Networks
• Two issues
• Routing and Wavelength Assignment (RWA)
• Sparse Conversion and Converter Placement Problem
• Argument they should be considered jointly!
• Given an RWA algorithm, how to solve the
  converter placement problem?
• Given a converter placement scheme, how to design
  an RWA algorithm?

3
Background

• Wavelength Division Multiplexing

4
Background

• WDM in different network environments
• Long-haul Point-to-Point link
• LAN (Broadcast and Select)
• MAN and WAN our focus!
• Services
• Lightpath circuit switching (near future) our
  focus!
• Datagram packet switching (long-term solution ?
  processing and buffering

5
Background

• Wavelength-routed WDM networks
• Circuit switching
• Physical topology
• A set of routing nodes connected by fiber links.
• Lightpath
• A lightpath has to be setup before the data
  transmission.
• Wavelength router (Optical Cross-connect)
• No O-E-O conversion All-Optical.

6
Background (Reconfiguration)

• An example of wavelength-routed network

7
Background Wavelength Router
Architecture of a wavelength router
8
Outline

• The optical backbone
• Circuit-Switched wavelength-routed WDM Networks
• Two issues
• Routing and Wavelength Assignment (RWA)
• Sparse Conversion and Converter Placement Problem
• Argument they should be considered jointly!
• Given an RWA algorithm, how to solve the
  converter placement problem?
• Given a converter placement scheme, how to design
  an RWA algorithm?

9
RWA Routing and Wavelength Assignment

• Definition
• Given network topology, a set of end-to-end
  lightpath requests (using one wavelength
  end-to-end)
• Problem Determine routes and wavelengths for the
  requests to achieve good performance.
• Offline RWA
• The entire set of requests are given in advance.
• Online RWA
• A sequence of lightpath requests arrives over
  time and each lightpath has a random holding
  time.
• Minimizing the Overall Blocking Probability is
  the Objective

10
RWA Routing and Wavelength Assignment

• RWA Two sub-problems
• Routing
• Wavelength Assignment
• Routing
• Static routing
• Routing DOESNT depend on the network state
• Dynamic routing
• Routing DOES depend on the network state

11
Online RWA Static Routing - SPR

• SPR Shortest Path Routing
• A single route is provided for each pair node.
• The routes do not change (fixed path).
• Pro
• Simple, short setup time
• Con
• Poor performance
• Low utilization of fiber resources link traffics
  are not balanced

12
Online RWA Static Routing - FAR

• FAR Fixed-Alternate Routing
• A set of routes is provided for each pair node.
• Route selection routes are tried sequentially
  according to a prior-determined order.
• Pro
• Improve the utilization of fiber links and the
  overall blocking performance.
• Fault tolerance.
• Con
• It does not consider the network state.

13
Online RWA Dynamic Routing

• Dynamic routing algorithms
• A set of routes (or all the possible routes) is
  provided for each pair node.
• Route selection is dynamic based on the
  information of network status, such as free
  wavelengths distribution, route length, etc.
• Pro.
• Better blocking performance.
• Fault tolerance.
• Con.
• Longer setup delay and complicated control.

14
Online RWA Wavelength Assignment

• Least-used wavelength assignment
• the wavelength used in the smallest number of
  links
• Most-used wavelength assignment
• the wavelength used in the greatest number of
  links
• Random wavelength assignment
• choose the wavelength randomly
• First-fit wavelength assignment
• All the wavelength are labeled in order
• choose the wavelength with the smallest label

15
Online RWA Wavelength Assignment

• Performance Comparison
• Most-used the best, but require wavelength
  information of all the links
• First-fit very good, only require wavelength
  information of the chosen route
• Random bad, but easy for analysis
• Least-used the worst

16
Outline

• The optical backbone
• Circuit-Switched wavelength-routed WDM Networks
• Two issues
• Routing and Wavelength Assignment (RWA)
• Sparse Conversion and Converter Placement Problem
• Argument they should be considered jointly!
• Given an RWA algorithm, how to solve the
  converter placement problem?
• Given a converter placement scheme, how to design
  an RWA algorithm?

17
Wavelength Conversion

• Wavelength Continuity Constraint A lightpath
  should use only a single wavelength over its
  physical links

18
Wavelength Conversion

• WCR Wavelength Convertible Router
• A wavelength router with conversion capability is
  called a wavelength convertible router.
• However, wavelength conversion is expected to be
  very expensive.
• Sparse Wavelength Conversion having WCRs at a
  small fraction of nodes is typically sufficient
  for a desired performance compared to full
  wavelength conversion
• Full Wavelength Conversion all nodes are WCRs

19
Sparse Wavelength Conversion
Blocking Prob. vs Number of WCRs in 8-node ring
network
20
Sparse Wavelength Conversion

• Converter Placement Problem
• Place K WCRs in an arbitrary mesh network with N
  nodes, so as to minimize the average blocking
  probability.
• Difficulties
• Blocking performance depends on lots of factors
  physical topology, traffic load, RWA, conversion
  capability, etc.
• Pure simulations for large networks are slow and
  impractical.
• There doesnt exist simple and accurate
  analytical models for calculating blocking
  performance.

21
Wavelength Converter Placement Converter
placement in Mesh Network

• Approaches
• Intuitive heuristic using simulation
• Enumeration based on an analytical model
• Heuristics based on an analytical model

22
Wavelength Converter Placement Converter
placement in Mesh Network

• Intuitive heuristic using simulation
• Step 1 Obtain some traffic parameters of the
  given network by simulation, assuming all the
  nodes are WCRs
• Nodal degree
• Number of lightpaths passing through a node and
  the average path length
• Number of wavelength conversions performed at a
  node
• Step 2 Place WCRs at the following nodes
• Nodes with high transit traffic
• Nodes which lie on the long lightpaths
• Nodes which perform a lot of conversions

23
Wavelength Converter Placement Converter
placement in Mesh Network

• Enumeration based on analytical models
• Step 1 Develop an efficient analytical model to
  calculate the blocking probability a trade-off
  between efficiency and accuracy.
• Step 2 Enumerate all the possible ways of
  wavelength converters placement and choose the
  one with minimum blocking probability
  times of calculation
• Problems
• Not scalable time consuming for large networks
• The accuracy of the result depends on the
  analytical model.

24
Outline

• The optical backbone
• Circuit-Switched wavelength-routed WDM Networks
• Two issues
• Routing and Wavelength Assignment (RWA)
• Sparse Conversion and Converter Placement Problem
• Argument they should be considered jointly!
• Given an RWA algorithm, how to design a converter
  placement scheme?
• Given a converter placement scheme, how to design
  an RWA algorithm?

25
Re-Examinations of Converter Placement Problem

• Most existing research on wavelength converter
  placement assumes the Shortest Path Routing (SPR)
  
• Our initial thinking
• How about FAR and dynamic routing?
• Investigated in our research
• Is there a general solution?
• For future investigation

26
Re-Examinations of Converter Placement Problem

• Converter placement with FAR
• We proposed the Minimum Blocking Probability
  First (MBPF) algorithm which is a heuristic based
  on an analytical model.
• The analytical model is used to calculate the
  blocking probability for the FAR algorithm.
• MBPF Place WCRs one by one.
• Each time we place a WCR on the node which can
  minimize the overall blocking probability the
  most significantly.

27
Re-Examinations of Converter Placement Problem
The 14-node NSFNET topology
28
Re-Examinations of Converter Placement Problem
Blocking Prob. Vs. Traffic Load in NSFNET using
FAR-FF algorithm
29
Re-Examinations of Converter Placement Problem

• LLR Least-Loaded Routing
• a novel dynamic routing algorithm
• A set of routes is pre-designed for each pair
  node.
• Route selection based on the number of free
  wavelengths of all the candidate routes the
  route with the largest number of free wavelengths
  is selected.

30
Re-Examinations of Converter Placement Problem

• WMSL Weighted Maximum Segment Length
• Place the WCRs one by one.
• Each time place a WCR at the most important node
• The node that can decrease the sum of maximum
  segment length of all the routes the most
  significantly is considered to be the most
  important.

31
Re-Examinations of Converter Placement Problem
Blocking Prob. Vs. Traffic Load in NSFNET using
LLR-FF algorithm
32
Outline

• The optical backbone
• Circuit-Switched wavelength-routed WDM Networks
• Two issues
• Routing and Wavelength Assignment (RWA)
• Sparse Conversion and Converter Placement Problem
• Argument they should be considered jointly!
• Given an RWA algorithm, how to solve the
  converter placement problem?
• Given the conversion capability, how to design an
  RWA algorithm?

33
Re-Examination of RWA

• Our focus dynamic routing
• Traditional dynamic routing schemes may not work
  well in the presence of wavelength conversion.
• Two factors for route selection
• Free wavelength distribution
• Route length
• Length of the route has always been ignored!
• Without wavelength conversion, these issues are
  largely correlated (wavelength continuance)
• Short length routes tend to have more wavelengths

34
Re-Examination of RWA WLCR

• We proposed the WLCR algorithm
• Weighted Least-Congestion Routing
• Route selection
• Assign a weight to each candidate route.
• Choose the route with the maximum weight value.
• How to choose the weight function?
• Free wavelengths F(R)
• Route length H(R)
• Our weight function

35
Re-Examination of RWA WLCR

• Comparison of WLCR and LLR under sparse
  conversion or full conversion
• Link utilization
• LLR gt WLCR gt FAR gt SPR
• Average route length
• LLR gtgt WLCR ? FAR gt SPR
• WLCR can result a whole lot better blocking
  performance!

36
Re-Examination of RWA WLCR
Blocking Prob. vs. traffic load in 8-node Ring
network with full conversion
37
Re-Examination of RWA WLCR
Blocking Prob. vs. traffic load in NSFNET with 5
WCRs
38
Conclusions and TBDs

• Converter Placement Problem
• MBPF for FAR
• WMSL for LLR
• Is there a general scheme for all the RWA
  algorithm?
• Routing Problem
• Dynamic routing WLCR
• How to design the weight function?
• How to incorporate the conversion information?

39
Q A