Impact of OXC Failures on Network Reliability

1
Impact of OXC Failures on Network
Reliability Zsolt Pándia, Andrea Fumagallia,
Marco Taccaa, Lena Wosinskab aOpNeAR Lab., Erik
Jonsson Shool of EECS, University of Texas at
Dallas bDepartment of Teleinformatics, Royal
Institute of Technology, Kista, Sweden
Node Equipment
Node Reliability Model
Purpose

• Reliability block diagram
• Built from two-state components (operating,
  failed) each having their independent reliability
  characteristics
• One component failure puts whole equipment into
  failed state

• Characterize influence of OXC equipment
  reliability on end-to-end connection provisioning
• Combine detailed component-based node reliability
  models with network level Differentiated
  Reliability
• Study of different networking scenarios with
  realistic data

• Node architectures
• Wavelength-selective optical cross-connect with
  electronic protection path (1)
• Wavelength selective optical cross-connect w/o
  protection (2)
• Clos architecture with one path for shared
  protection (3)
• Clos architecture w/o protection (4)
• Switching technologies
• InGaAsP/InP (laser-amplifier gate-switch arrays)
  (a)
• MEMS (MicroElectroMechanical Systems) (b)

Differentiated Reliability
Methodology
Network Reliability Model
Networking scenarios

• Basic idea provide just the necessary level of
  protection and thus improve resource sharing
• Introduction of node failures makes it impossible
  to survive every single failure
• Validation of single-failure approach is
  necessary (reason for missing curves
  requirements cannot be guaranteed by considering
  only single failures)

• Link reliability is a function of the length of
  cable span
• Three examined scenarios based on scaled versions
  of the same network
• European (continental size),
• National (15 size),
• Metropolitan (150 size)
• 4 wavelengths per fiber is assumed
• Each type of node equipment is tested in all the
  scenarios

• Network components (links and atomic nodes) also
  have two states
• Independent component failures
• Asymptotic unavailability is assigned to each
  component
• SPP protection scheme

• Simulation of WDM networks
• No wavelength conversion
• Dynamic traffic, different call arrival
  intensities (lambda)
• Arriving demands specify reliability requirements
• Resource allocation using the extension of
  DSPP-DiR
• Only single failure scenarios are considered for
  optimization with simulated annealing
• Centralized call admission control with a
  single-slot buffer to ensure that endpoints of
  served calls are distributed evenly
• Statistical analysis of results
• Measure of goodness call blocking probability

Continental scale
National scale
Metropolitan scale
Summary

• Provided reliability is higher than the
  guarantees (not all possible multiple failures
  disrupt each connection)
• Single failure approach seems to work well for
  metropolitan scale networks and may be applied on
  the national scale, as well
• In the continental scale link failures determine
  network reliability, while in the metropolitan
  scale node failures do
• OXC equipment selection does make a difference
  under certain circumstances

Maximum failure probability 0.005
Maximum failure probability 0.05
Maximum failure probability 0.0005