Why Switches Are the New Computing Backbone: Definitions, Architectures, and Market Trends

Market Overview

According to IDC, the global switch market reached 308 billion CNY in 2022 (up 17.0% YoY) and is projected to grow to 322 billion CNY in 2023 and 3.77 trillion CNY by 2027, with a CAGR of about 4.6% from 2022‑2027. China accounted for 591 billion CNY in 2022 (19.2% of the global market) and is expected to reach 645 billion CNY in 2023 and 878 billion CNY in 2027 (23.3% share).

1. Switch Definition and Classification

A switch (or “switch”) is a network device that forwards electrical or optical signals, providing dedicated point‑to‑point communication paths between any two connected nodes. Ethernet switches are the most common, but telephone voice switches and fiber switches also exist.

Based on OSI layers:

Layer‑2 switch: operates on MAC addresses; typical for access and aggregation layers.

Layer‑3 switch: routes based on IP addresses and protocols; used in core networks and sometimes provides Layer‑4 functions.

Layer‑4 switch: makes forwarding decisions using TCP/UDP port numbers, enabling load‑balancing for services such as HTTP, FTP, NFS, Telnet, etc.

Layer‑4‑and‑above (application‑level) switches: primarily deployed in data‑center environments.

Based on network hierarchy:

Access layer switch: high port density, usually 10/100 Mbps ports, provides uplink bandwidth.

Aggregation layer switch: modular chassis with management, optical, and high‑speed electrical modules, high backplane capacity.

Core layer switch: chassis‑based, supports multiple modules, provides backbone connectivity.

Based on application area:

Wide‑area switch: operates at the data‑link layer, connects LANs or high‑performance servers, mainly used by telecom operators.

Local‑area switch: connects end‑devices such as PCs and printers within a LAN.

2. Switch Composition

Hardware components include chassis, power supply, fan, backplane, management engine, system controller, switching module, and line cards. The chassis protects internal electronics; fans provide cooling; power supplies can be internal or external; the backplane interconnects the management engine, switching modules, and line cards.

Key sub‑components:

Management engine: provides a serial console for configuration and management.

System controller: controls power and fan operation.

Line card: hosts Ethernet interfaces for data transmission.

Switching module: contains high‑performance ASICs that forward frames between ports.

3. Switch Architecture

Industry‑standard architectures include Full‑Mesh, Crossbar, and CLOS, with most high‑end core switches adopting CLOS.

CLOS Board Designs

Non‑orthogonal (parallel) design: line cards run parallel to the switching fabric; used by Huawei. Drawbacks include signal interference from backplane wiring, limited bandwidth upgrades, and cooling challenges.

Orthogonal design: line cards connect perpendicularly to the switching module via the backplane; reduces signal loss but still limits bandwidth upgrades; used by Cisco.

Backplane‑less design: line cards connect directly to the switching module, eliminating backplane constraints and simplifying cooling.

Switching Module Mechanism

Data flows from line card A → backplane → switching module → ASIC → backplane → line card B. The ASIC learns MAC addresses, builds a MAC table, and forwards frames only to the appropriate port, reducing broadcast domains while still belonging to the same collision domain.

4. Technical Principles

Switches operate at OSI Layer 2 (Data Link). Each port forms an independent physical segment with its own full bandwidth, allowing simultaneous transmissions without contention. MAC learning creates a table that maps MAC addresses to ports, enabling selective forwarding and reducing broadcast traffic.

Switches contain a high‑bandwidth back‑bus and internal switching matrix. When a frame arrives, the control circuit looks up the destination MAC in the table; if found, the frame is sent to the specific port; otherwise, it is flooded. The device can also filter traffic based on IP tables, helping to segment networks and reduce collision domains.

5. Comparison with Hubs and Routers

Hubs simply repeat incoming signals to all ports; all devices receive every frame. Routers connect separate networks, each with its own IP address, and route packets between them, enabling inter‑network communication.

6. Performance Metrics

Key metrics include per‑port bandwidth, backplane capacity, latency, and forwarding rate. Images illustrating typical performance charts are omitted for brevity.

7. Application Scenarios

Switches are categorized by use case:

Commercial switches: SMB, campus, and data‑center switches.

Industrial switches: Designed for harsh environments (temperature extremes, electromagnetic interference, salt‑fog, vibration). They use robust TCP/IP stacks, support higher data rates (1 Gbps, 10 Gbps), and include intelligent alarms for monitoring.

8. Data‑Center Switch Topologies

Traditional Three‑Tier Architecture

Consists of access, aggregation, and core layers. Limitations include bandwidth waste due to STP blocking, large fault domains, and difficulty scaling to massive cloud‑scale networks.

Leaf‑Spine (Distributed Core) Architecture

Leaf switches aggregate server traffic and connect upward to spine switches, which interconnect all leaf nodes in a full‑mesh. Benefits:

Flat topology: short paths reduce latency and improve application performance.

Scalability: add spine nodes for more bandwidth, add leaf nodes for more access ports.

Low convergence ratio: can achieve 1:1 non‑blocking performance.

Edge traffic handling: leaf switches process IoT and sensor traffic at the edge, while spine provides low‑latency, high‑throughput backbone.

Multi‑cloud management: supports high‑performance, fault‑tolerant connections between multiple clouds.

9. Industrial Switch Definition and Certification

Industrial Ethernet switches are built for control‑system environments, offering wide‑temperature operation, strong electromagnetic compatibility, corrosion resistance, and IP‑rated enclosures (typically IP40 or higher). They support high‑speed Ethernet (1 Gbps, 10 Gbps) and include intelligent alarm functions to ensure reliable communication in hazardous settings.

10. Switch Industry Chain

Upstream: chips, components, optical modules, PCBs, power modules, and chassis parts.

Midstream: unmanaged switches, Layer‑2 managed switches, Layer‑3 managed switches, PoE switches, industrial switches, and data‑center switches.

Downstream: telecom operators, cloud service providers, data‑center operators, and other enterprise customers.