Differentiated Reliability in Distributed Survivable Optical Mesh Networks

1
Differentiated Reliability in Distributed
Survivable Optical Mesh Networks Dhruvish Shah,
Dr. Marco Tacca and Dr. Andrea Fumagalli Optical
Networks Advanced Research (OpNeAR) Lab
Node Architecture
Network Architecture
Distributed Control Plane

• Advantages
• No Central Controller
• Less signaling compared to centralized
  counterpart
• Less set-up delay
• Limitations
• More connection blocking
• Complex Node Architecture

Distributed Control Plane

• IP Over WDM Mesh Network with Distributed Control
  Plane
• Multiple wavelengths and fibers per link
• Per source node queue for fairness in blocking
• On-Line Optimized Routing and Wavelength
  Assignment (RWA) based on Differentiated
  Reliability (DiR) and Shared Path Protection (SPP)

Simulation Scenario
Network Protocols (Contd.)
Network Protocols

• Berkeley Network Simulator (NS2)
• 19 Node European Topology with 39 bidirectional
• links
• Poisson arrivals, uniformly distributed among
  nodes,
• exponential demand lifetime
• One fiber in each direction of propagation
• Maximum Conditional Failure Probability (MCFP)
• of 0.03 assuming single link failure scenario
  allowing
• single unprotected link on working lightpath

• RWA
• Source computes lightpath resources
• Optimize network resources with Simulated
  Annealing
• Linear (LB) and Matrix DPM approach

• CRANKBACK
• Contentions cause lightpath set-up failure
• Reattempt to set-up lightpath until successful
• Avoid using the resource which was contended
• for successive RWA computations
• Aggressive reattempts
• Ensures fairness for longer lightpaths

• RSVP and Modified OSPF
• Destination initiated reservation
• Event driven node based OSPF for data
  synchronization

Summary
Comparison of DSPP-DiR with DSPP
Probability Distribution
Distributed and Centralized Control Planes

• With only one unprotected link the
• DSPP-DIR scheme reduces blocking
• dramatically
• As traffic becomes faster, due to lack of
• faster synch blocking increases
• The Distributed Control Plane performs
• almost same as its Centralized counterpart
• at light load, blocking increases more
• rapidly at higher load
• Single buffer queue at each source node
• and CRANKBACK ensure fairness among
• nodes