Question Answer

Dynamic Connection Network in Computer Architecture

A dynamic connection network (also called a dynamic interconnection network) is a switching-based fabric that connects multiple processors, memories, or I/O modules and can reconfigure its paths on demand. Instead of fixed links, it uses switches to establish temporary routes between any input and output, allowing multiple transfers to occur concurrently.

Key Characteristics

• Reconfigurable: Paths are set up per request and released after use.
• Concurrent transfers: Several non-conflicting connections can exist at the same time.
• Arbitration needed: Resolve conflicts when multiple inputs target the same output.
• Switching styles: Can be circuit-switched or packet-switched.
• Used in multiprocessors and high-speed memory systems for scalable bandwidth.

Type 1: Crossbar Switch

A crossbar is an N × N switching network with a grid of crosspoints that can directly connect any input to any output. Each crosspoint acts like a controllable switch.

• Structure: N inputs, N outputs, and N² crosspoints.
• Operation: If output j is free, input i can be connected to output j by closing one crosspoint.

Basic steps during a transfer:

1. Request: An input requests a specific output.
2. Arbitrate: If multiple inputs request the same output, an arbiter selects one (e.g., round-robin).
3. Connect: The selected crosspoint is enabled to form a path.
4. Transfer and release: Data moves; the path is freed after completion.

Advantages:

• Non-blocking: Any free input-output pair can connect independent of other connections.
• High throughput and low latency under load.

Limitations:

• High cost and area: O(N²) crosspoints and complex wiring.
• Power and physical scalability challenges as N grows.

Type 2: Multistage Interconnection Network (Omega/Banyan)

A multistage interconnection network (MIN) like the Omega or Banyan network uses several stages of small 2×2 switches with fixed inter-stage wiring patterns (e.g., perfect shuffle). It provides many possible routes using far fewer switches than a crossbar.

• Structure: log₂N stages, each containing N/2 simple 2×2 switches.
• Routing: Each stage forwards based on one bit of the destination address.

Typical 2×2 switch control rule:

At each 2×2 switch:
- control = current destination bit (MSB to LSB across stages)
- control = 0 → go straight
- control = 1 → cross over

Example flow (N = 8):

Inputs → [Stage 1: 2×2 switches] → shuffle → [Stage 2] → shuffle → [Stage 3] → Outputs
Each stage uses one of the 3 destination bits to set straight/cross.

Advantages:

• Lower cost: O(N log N) switches; easier to scale than a crossbar.
• Moderate latency: Paths are log₂N switches long.

Limitations:

• Blocking: Two connections may need the same internal link; both cannot pass simultaneously.
• Requires contention resolution and buffering/scheduling to maintain performance.

Quick Comparison

• Crossbar: Non-blocking and fast, but expensive and power-hungry for large N.
• Multistage (Omega/Banyan): Cost-efficient and scalable, but blocking and needs careful routing/arbitration.

Other examples of dynamic networks include shared bus systems (very low cost, but low concurrency) and Clos networks (multi-stage, often rearrangeably non-blocking). Choosing a type depends on cost, scale, and target bandwidth.