Switching Fabrics

Switching Fabric Classifications

Categorizing switching fabrics by technology, directionality, and combinatorial properties (blocking vs. non-blocking).

Introduction to Classification

A switching fabricis the internal hardware network within a switching node that physically connects input ports to output ports. Its design affects the node's throughput, latency, cost, and capabilities. Switching fabrics can be classified using several criteria that may overlap. A single switch can combine characteristics from all of the categories described here.

Classification by Technology

This classification describes technologies used in successive generations of switching equipment.

  • Electromechanical: An early form of switching that used moving parts such as rotary selectors and crossbar switches. These devices operated slowly, were noisy, and required frequent maintenance.
  • Electronic: Based on semiconductor components such as transistors and logic gates. These components increased speed and reliability while reducing equipment size, forming the basis of modern digital switching.
  • Optical: Switches signals in the optical domain, so the data path can avoid an optical-to-electrical conversion in each element. Technologies such as MEMS (micro-mirrors) physically direct light beams. They can be transparent to data format and rate, while capacity and switching time depend on the device design.

Classification by I/O Ports and Directionality

This category describes the physical layout and the flow of information through the fabric.

Port Configuration

  • Two-sided fabric: Has distinct sets of input and output ports. A connection is made from an input to an output.
  • One-sided fabric: Has a single set of ports. Each port serves as an input or an output according to the connection being established.

Number of I/O Ports (N1N_1 inputs, N2N_2 outputs)

  • Concentrator: More inputs than outputs (N1>N2)(N_1 > N_2). Aggregates traffic from many slower links onto fewer faster links.
  • Expander: More outputs than inputs (N2>N1)(N_2 > N_1). Distributes traffic from fewer inputs to a larger number of outputs.
  • Distributor: Equal numbers of inputs and outputs (N1=N2)(N_1 = N_2). This configuration is used to route traffic between peer devices.

Information Flow Direction

  • Unidirectional: Signal flow is strictly from inputs to outputs only.
  • Bidirectional: Signal flow can occur in both directions. A design may use two opposite unidirectional paths, often called a "four-wire" configuration, to create a full-duplex link.

Classification by Combinatorial Properties (Blocking)

This classification determines whether a new connection can be established in every admissible state of the fabric. A blocking state occurs when a free input port cannot be connected to a free output port because all available internal paths are occupied by existing connections.

  • Blocking Fabrics: Fabrics in which a blocking state can occur under particular traffic patterns. They are simpler and cheaper, but a new connection request can be rejected when no internal path is available.
  • Non-Blocking Fabrics: Fabrics designed to avoid blocking states. They include the following categories:
    • : A path can be found between any free input and free output without re-routing existing connections.
    • : A new connection is possible when earlier connections were routed with a specified path-selection policy. The policy avoids internal conflicts without rearranging existing paths.
    • : If a new connection is blocked, it can be established after changing the paths of one or more existing connections. This requires control-plane work and may disturb active traffic, so it is less desirable for real-time services.

Classification by Path Separation

This describes how different communication paths are separated as they traverse the switching fabric. The criterion is related to multiplexing techniques.

  • Space-Division: Each connection uses a separate path through the switching fabric, such as a specific set of crosspoints. This approach is used in electronic and optical switches.
  • Time-Division: Multiple connections share the same physical paths in different time slots. This is the basis of TDM switching.
  • Wavelength-Division: In optical systems, connections are separated by assigning different wavelengths of light to them. The signals can then share optical components.
  • Code-Division: Connections share the same time and frequency resources, while receivers distinguish them using assigned codes, often orthogonal codes as in CDMA.