Deploying Advanced Path Computation Technologies for Optical Transport Networks

1
Deploying Advanced Path Computation Technologies
for Optical Transport Networks

• Daniel King Aria Networks

Takehiro Tsuritani KDDI Labs
2
Agenda

• Evolution of optical transport networks
• Why do we need advanced path computation
  capabilities?
• intelligent Virtual Network Topologies for PCE
  NMS
• Applicability of Path Computation Elements (PCE)
  to optical transparent networks
• Using a PCE with a Network Management System
  (NMS)
• Combined platform results
• Conclusions

3
Evolution of optical transport networks

• Optical transport networks are evolving
• Moving from ring to mesh topologies
• Bandwidth speeds are increasing exponentially
• Advanced recovery and protection mechanisms are
  available
• Need to minimize O/E/O conversions
• Management of optical transport networks
• Typically a proprietary Network Management System
  (NMS)
• Deployment of new services is a slow and a
  complicated process
• manually coordinated provisioning across the
  network
• risk of damage to existing services
• resource isolation (stranding) and orphaning
• New technologies and mechanisms are being
  developed via Internet Standards organizations.
  

4
Why do we need advanced path computation
capabilities?

• Sophisticated route planning for the deployment
  and the operation of the optical transport
  network is very complex
• Calculation of optical impairments
• Lightpath route parameters
• Wavelength continuity
• Attenuation (power loss)
• Amplified spontaneous emission (ASE)
• Chromatic dispersion (CD)Polarization mode
  dispersion (PMD), optical fiber non-Linear phase
  shift (NLPS) and filter concatenation effect.
• Consideration of network constraints
• Affinities (inclusion/exclusion)
• Link, Node, Shared Risk Link Group (SRLG)
• Maximum end-to-end IGP metric
• Maximum hop count
• Degree of paths disjointness (Link, Node, SRLG)

5
Aria Networksintelligent Virtual Network
Topologies (iVNT)
Cost Hops Delay OSNR PMD NLPS OXC/OEO Link
Protection Symmetrical ? Affinities
Customisation, Integration and Visualisation
Adapters
Scripting
Graphical Interface
Multilayer / Multidomain
Transport Technology
MPLS-TE
IP
Optical(WDM/GMPLS)
Ethernet(PBB-TE)
TDM(SONET/SDH)
Mobile
Generic, Multi-constraint based Routing Engine
DANI Machine Learning Evolutionary Computing
Model Builder
Model Executor
Distributed
Fault-tolerant
Genetic Algorithms
Spiking Neural Networks (SNNs)
Darwinian Neural Networks (DNNs)
Radial Basis Functions
Bayesian Nets
Genetic Neural Networks (WNNs GNNs)
Multidimensional Classifiers
Genetic Programs
6
iVNT Computation Engine (DANI)Algorithm
Selection, Evolution Execution
Selection
Evolution
Execution
Genetic Neural Networks
Spiking Neural Networks
Model Executor
Model Builder
Genetic Neural Networks Spiking Neural Networks
Genetic Algorithms Radial Basis
Functions Multidimensional Classifiers Bayesian
Nets Darwinian Neural NetworksGenetic
Programs Spiking Neural Networks Genetic Neural
Networks
Generic, Multi-constraint based Routing Engine
Cost Hops Delay OSNR PMD NLPS OXC/OEO Link
Protection Symmetrical ? Affinities
DANI Algorithm hosting platform Distributed
fault tolerant Supports Win, Solaris Linux
eDNA Algorithms represented as DNA Fast
processing and learning Parallel and distributed

7
iVNT Computation Engine (DANI)Algorithm
Selection, Evolution Execution
Algorithm Selection
ANN Classic 3 layer/Back Prop Artificial Neural
Network GNN ANN implementation with design
enhancements and interpretability features SNN
Implementation of new generation of pulsed neural
networks Dijkstra Solves the shortest path
problem for a graph with non negative edge path
costs, outputting a shortest path tree. DNN 3D
pulsed neural network with high computational
scalability and high degree of functional
parameterisation k-NN Classic Nearest Neighbour
j-NN Enhance k-NN with feature enhancements for
interpretability and data robustness Decision
Tree Representing of hierarchical rules that
lead to a class or value. Bayesian Net A maximum
likelihood classifier, using a priori weighting,
representing the probabilities that data belongs
to each class. SVM State Vector Machine are the
application of linear methods to very high
dimensional feature space. Steiner Compute the
minimum-Cost tree
8
iVNT Computation Engine (DANI)Algorithm
Selection, Evolution Execution
Algorithm Evolution

• eDNA - Genome (List of parameters to represent
  an algorithm - model)
• Parameters that define the variable parts of an
  algorithm.
• Parameters that define the variable parts of the
  optimisation process performed on a DANI client.

Tribe 1
Fittest eDNA
GNN
0.77
GNN
0.73
GNN
0.83
GNN
0.71
GNN
0.81
GNN
0.53
GNN
GNN
0.79
0.40
GNN
0.17
Fittest eDNA kept
GNN
0.83
9
iVNT Computation Engine (DANI)Algorithm
Selection, Evolution Execution
Model Executor
ObjectiveFunction
ConstraintWeights
CombinatorialProcess
Generate candidates Dijkstra, Steiner, etc.
Optimize All
Auto Generate Topology Services
C1 C2 C3 C4 Cn
Genetic Neural Networks Spiking Neural Networks
1000
C1, PN1, PL1
10
C2, PN2, PL2
PN1 PN2 PN3 PN4 PNnPNm
10,000
C3, PN3, PL3
1000
C4, PN4, PL4
750
Cn, PNn, PLn
Constraint-weighted objective functions
750
PNn, PLn
PL1 PL2 PL3 PL4 PLn PNm
Service Constraints C Node Properties
PN Link Properties PL
Automatically Build Internal Model
1. Automated and Continual Learning 2. Achieve
Necessary Fitness 3. Output the solution
10
PCE interworking with NMS

• Network management system
• NMS retrieves wavelength and link information
  and updates PCE
• NMS manages service requests
• NMS sends path computation requests to the PCE
• PCE path computation
• Topology information
• - TE link information
• - LSP information
• - Wavelength resource information
• Physical impairment parameters
• Explicit route provisioning over GMPLS-enabled
  transparent network

All-optically transport network
Optical PCE
Service Requests
NMS
11
Computing optical transport constraints

• Link performance and amplifier characteristics
  can be stored in the NMS
• Link OSNR
• Link PMD
• Link Nonlinearity

NMS
12
Computing optical transport constraints

• Wavelength availability information on each link
  is utilized as a network resource information
• The same wavelength should be selected along the
  path if there is no regenerator
• If such a wavelength is not available, then the
  insertion of regenerators including wavelength
  conversion functions needs to be considered

Resource Info
l1
l4
l2
l3
Link1
Link2
Link3
13
Computing optical transport constraints

• Using physical impairment constraints to compute
  a path via the PCE
• Optimum route selection
• Maximum OSNR
• Minimum PMD
• Combination of both constraints

N1
Link2
Link1
Link3
Link4
N2
N3
N4
Link6
Link5
N5
14
Actual results of optical path planning1.
Select the path with the maximum OSNR
R1OSNR24.2dB
Route1
L1
L2
R2OSNR21.2dB
R3OSNR22.8dB
N2
N1
L4
L3
Route2
L5
L6
Route3
15
Actual results of optical path planning2.
Select the path with the maximum OSNR
R2PMD2.6ps
Route1
L1
L2
R1PMD6.4ps
R3PMD4.8ps
N2
N1
L4
L3
Route2
L5
L6
Route3
16
Actual results of optical path planning3.
Consider OSNR PMD for path selection
L1
L2
N2
N1
L4
L3
Route2
L6
L5
R3 Limit OSNR22dB Limit PMD6ps
Route3
17
Actual results of optical path planning4.
Wavelength continuity

• We evaluated the capability to deal with a
  wavelength constraint.
• Wavelength continuity
• Wavelength conversion by 3R

Regenerator
N1
Link1
Link3
N3
N4
Link2
N2
18
Actual results of optical path planning4. Path
computation with regeneration
1) Computed path
l2
N1
N3
Link1
l2
N4
Link3
Link2
N2
Route information
19
Actual results of optical path planning5. Path
computation using regenerator node
1) Computed path
Wavelength conversion From l2 to l1 at Node 3
N1
N3
N4
2) Computed path
N2
20
Actual results of optical path planning5.
Japanese transparent optical network

• In order to evaluate the combined PCE and NMS
  platform in a realistic transparent mesh network
  model a nation-wide network was utilized

• 22 links (80km x N spans) and 15 nodes located in
  major cities in Japan
• Service policy
• Satisfy required OSNR
• Limit OSNR
• Limit PMD
• Limit NLPS
• Select the highest quality path
• Maximum OSNR
• Achieve the path computation result with a
  quality-optimum end-to-end path

21
Actual results of optical path planning5.
Japanese transparent optical network
OSNR17.5dB PMD3.3ps NLPS0.43
10Gbps service
Computed path with highest quality
N2
Calculated results of the transparent path
N1
Physical parameters of the link
Physical parameters of the node
22
Conclusion

• Combining the PCE and NMS technologies for
  optical transparent mesh networks, allows for
  sophisticated optical service path computation
  and network management.
• Greater computational capabilities are a key
  component to support the PCE and NMS
  architecture.
• A variety of algorithms are required to solve
  multiple problems.
• Consideration of multiple constraints
• Physical impairment constraints
• Wavelength continuity constraint
• Network constraints
• Advance path computation functions like Global
  Concurrent Optimization (GCO) and
  Point-to-Multipoint (P2MP) path computation can
  also be performed using this combined
  architecture.
• The combined PCE and NMS architecture can be
  applied to other transport and service
  technologies.