Despite the recent market growth difficulties encountered by some segments of the communications industry, the long-term fundamental case for intelligent optical networks remains compelling.

As the mechanics of intelligence are embedded in associated systems, not the optics, the relation and scaling of these to the optical components is critical.

RRX Network is conducting an ongoing investigation from the abstract requirements to design particularities with a running comparison to emerging areas of artificial intelligence.

Clearly the lack of satisfactory solutions to self supervision issues in artificial intelligence is a key consideration.

Although network awareness is not explicitly considered in the preliminary vision we are committed to embedding preliminary components of Bayesian network awareness in associated systems, in an appropriate and scalable manner.

Preliminary Vision

Network architecture is basicly a set of abstract principles for design. Optical networks are a subset of all networks and as such their architecture should likely meet all standard abstract requirements, as far as possible:

• accountability
• cost effectiveness
  
• distributed management
  
• ease of attachment
  
• heterogeneity
  
• interconnection
  
• robustness
  
• scalability

The choice of particular abstract level requirements, is the most critical step.

The legacy network assumes:

• variable serialized data sequences
  
• emdedded control plane mechanics
  
• adaptive end nodes (polite back off congestion control)
  
• hierarchical multiplexing (edge to core model)
  
• routing updates based on embedded control parsing
  
• routing control based on updated state tables per transit interface
  
• routing adaptivity based on rapid convergence (sometimes problematic)

In our vision next generation intelligent optical network architecture should preferably:

• accommodate burst multiplex operations.
  
• allow flexible effective multi-provider/multi-homing solutions.
  
• allow the data flow owner to specify intermediate carriers.
  
• encapsulate legacy functionality while providing new functionality.
  
• optimize the disposition of components in a cost effective manner.
  
• optionally support AAA within domains.
  
• reduce susceptibility to DoS attacks.
  
• scale easily to very large models.
  
• support open-ended extensibility in terms of heterogeneity.
  
• support selective transparency.
  
• support very large object models.
  
• support virtual memory.

Specifically, as the active and passive optical components, are not themselves intelligent, the additional intelligence added, by the network context, might include:

• discovery
  
• error handling/reporting
  
• naming
  
• provisioning
  
• routing
  
• security

The intelligent architecture context, apparently, implies that design should specify:

• How communication is grouped or aggregated.
  
• How data flow and control plane resources are divided across data flows and how agent systems re-act and interact, i.e., fairness and congestion.
  
• How data flows, aggregates and their agents are modularized/distributed.
  
• How differing data flow/aggregate QoS is arranged, requested and achieved.
  
• How inter-domain provisioning/routing is requested and achieved.
  
• How naming, addressing, and routing functions interact.
  
• How network domain management boundaries are drawn and interact.
  
• Where and how remembered state is maintained and how it is removed.
  
• Where security boundaries are drawn and how they are checked and enforced.
  
• Which entities are named and how?