C-LIGHT telephone TEL:+86 132 6656 7067    
Language
C-LIGHT search

ToR vs Middle-of-Row Architecture

By C-LIGHT Marketing 丨 Jul 10, 2025
Table of Contents

    Data center access switches can be installed directly inside server cabinets or consolidated into dedicated network racks serving an entire row. These physical deployment choices influence more than cable routing: they affect switch count, port utilization, rack-level power density, cooling requirements, maintenance procedures and the way network capacity is expanded.

    Top-of-Rack (ToR) architecture places access switches in individual server racks, keeping server-to-switch connections short. Middle-of-Row (MoR) architecture moves those switches into a dedicated cabinet near the center of a server row, allowing a shared switching system to serve multiple racks. Both can support modern Ethernet leaf-spine fabrics, but their physical infrastructure and operational requirements are different.

    1. What Is Top-of-Rack Architecture?

    In a ToR deployment, one or more access switches are installed near the top of each server cabinet. Servers, storage systems and other rack-mounted equipment connect directly to the switches in the same cabinet. The ToR switches then connect upstream to spine or aggregation switches through high-speed uplinks.

    Because server-to-switch links remain within the rack, ToR generally keeps horizontal cabling between server cabinets to a minimum. Short copper DAC assemblies, AOCs or other supported optical connections may be used for local links, depending on interface speed, reach and equipment compatibility.

    Typical ToR arrangement

    Rack 1ServersLocal ToR switch
    Rack 2ServersLocal ToR switch
    Rack 3ServersLocal ToR switch

    Each server cabinet has its own access-switch resources and connects to the upstream fabric.

    Typical MoR arrangement

    Rack 1Servers
    Rack 2Servers
    MoRShared switches
    Rack 3Servers
    Rack 4Servers

    A dedicated network cabinet near the row center serves multiple server cabinets through structured cabling.

    Simplified physical layouts. Actual switch counts, redundancy and rack positions depend on the deployment.

    2. What Is Middle-of-Row Architecture?

    Middle-of-Row architecture places access switches in one or more dedicated network cabinets around the center of a server row. Instead of installing a separate access switch in every server rack, the centralized switches provide connectivity to multiple cabinets through structured horizontal cabling.

    The MoR cabinet commonly functions as a horizontal distribution point. Patch panels and permanent cabling can connect server racks to the network equipment, while uplinks connect the access layer to the wider data center fabric.

    MoR shares the centralized switching concept used by End-of-Row (EoR) deployments. Its defining physical difference is switch location: MoR places the network cabinet near the middle of the row, whereas EoR locates it at one end. In a comparable row layout, this central position can reduce the longest server-to-switch cable route.

    MoR does not inherently require a different Ethernet topology or routing protocol. The access switches can still operate as leaf switches in a leaf-spine fabric, with the physical cabinet location determined by cabling and facility design.

    3. Main Architectural Differences

    Design FactorTop-of-Rack (ToR)Middle-of-Row (MoR)
    Switch locationInside or near the top of each server cabinetDedicated network cabinet near the row center
    Switch coverageUsually one rack, sometimes a small rack groupMultiple server racks in the same row or zone
    Server connectionsShort connections within the cabinetHorizontal structured cabling to the shared switch location
    Switch quantityGenerally more access switches across a large deploymentCan reduce the number of separate access-switch systems
    Port utilizationMay leave unused ports in partially populated racksPorts can be pooled across multiple server cabinets
    Rack spaceConsumes space in server cabinetsRequires dedicated network rack space
    Power and coolingSwitch heat and power are distributed among server racksSwitch heat and power are concentrated in network cabinets
    Cabling managementShort local cabling, with uplinks leaving each rackMore horizontal cabling and centralized patching
    Expansion modelCan add switching capacity rack by rackExpansion depends on shared switch capacity and cabling availability

    4. Server-to-Switch Cabling and Cable Distance

    Cable routing is one of the most visible differences between the two designs. ToR keeps server connections close to the access switch, often allowing short copper assemblies inside the same cabinet. Only uplinks and other inter-rack connections need to leave the rack.

    MoR requires connections from individual server cabinets to a shared network location. These horizontal runs need to be planned around overhead trays, underfloor pathways, patch panels, cable separation and service access. The central position can shorten the maximum route compared with placing the same shared switch at the end of the row, although actual cable lengths depend on the floor plan and routing paths.

    Cabling ConsiderationToRMoR
    Local server linksUsually short and contained within the rackRun from server cabinets toward the MoR cabinet
    Horizontal cable pathwaysGenerally fewer server-to-switch horizontal runsRequired for connections from multiple cabinets
    Patch panel planningCan be simpler at the access layerImportant for centralized cross-connects and moves
    Cable congestionMore localized rack cablingPotentially larger cable bundles near the network cabinet
    Future recablingOften easier to isolate to an individual rackMay involve shared pathways and centralized patching

    Neither architecture eliminates structured cabling requirements. ToR still needs organized uplink pathways and backbone connectivity, while MoR requires a carefully designed horizontal cabling system. In both cases, cable labeling, pathway capacity, bend radius, separation requirements and documentation are essential for long-term maintenance.

    5. Switch Port Utilization and Hardware Costs

    ToR provides dedicated access-switch capacity at each rack. This is convenient when racks have consistent server populations, but the number of switches can increase quickly as the facility grows. A rack with relatively few servers may still require its own switch pair to maintain the intended redundancy design, leaving some ports unused.

    MoR can pool access ports across several server racks. A centralized switch system may therefore achieve higher average port utilization when server populations are uneven. It can also reduce the number of separate switch management points.

    However, fewer switches do not automatically mean a lower total project cost. MoR may require additional horizontal cabling, patch panels, network cabinets and installation labor. Larger modular switches may also have different capital, support and maintenance costs from smaller fixed-configuration ToR devices.

    A realistic cost model should include:

    • Access-switch hardware, licenses and support contracts.

    • Number of active and reserved ports in each rack or row.

    • DAC, AOC, optical transceiver and fiber cabling requirements.

    • Structured cabling, patch panels, cable trays and installation labor.

    • Power distribution, cooling capacity and rack-space costs.

    • Spare capacity for future server additions and technology upgrades.

    6. Power Consumption and Cooling Requirements

    ToR distributes access-switch power consumption across server cabinets. This can simplify centralized network-rack requirements, but it adds electrical load and heat to racks that already contain servers, accelerators and storage equipment. High-density racks need sufficient power distribution and airflow to accommodate both computing and networking hardware.

    MoR moves much of the access-switch power and heat into dedicated network cabinets. This can free capacity in server racks, but the network location must be designed for concentrated equipment loads. Cooling, airflow direction, cable openings and cabinet service clearance need to be considered together.

    The thermal impact also depends on the switch generation and port configuration. A high-density 400G or 800G access switch can have substantially different cooling requirements from a lower-speed fixed switch. The physical placement decision should therefore use actual equipment power and thermal specifications rather than assumptions based only on rack location.

    7. Redundancy, Fault Isolation and Maintenance

    Both ToR and MoR can support redundant network paths, but their failure domains need careful planning.

    In a ToR design, redundant switches can be installed in each server cabinet. Servers connect to separate switches, and the switches use independent uplinks where the network design supports them. A failure can often be isolated to a single rack, although shared power, cabling or upstream dependencies can still create wider outages.

    In MoR, multiple access switches or modular switching systems can provide redundant connectivity for the row. However, centralizing many connections in one cabinet creates a location that requires particular attention to power feeds, cooling, physical access and cable-path diversity.

    Operational AreaToR DesignMoR Design
    Fault isolationOften easier to associate access failures with an individual rackCentralized failures may affect multiple connected racks
    Switch maintenanceMaintenance is distributed across many cabinetsAccess equipment is concentrated in dedicated network cabinets
    Power redundancyNeeds rack-level planning for redundant switch powerNeeds independent power arrangements for centralized switches
    Cable troubleshootingShort local connections can simplify tracingRequires accurate records of horizontal runs and patching
    Physical accessNetwork work takes place in server cabinetsDedicated network cabinets provide a separate maintenance area

    8. DAC, AOC and Optical Transceiver Selection

    Physical switch placement influences the connection media used between servers and access switches. ToR often creates short server-to-switch distances, making passive DAC a practical option when the interface, cable length and equipment compatibility permit it. AOC or optical transceiver-based connections may be appropriate when greater reach, cable routing flexibility or higher-density optical connectivity is required.

    MoR typically introduces longer horizontal connections between server racks and the centralized switching cabinet. The media choice depends on actual distance, data rate, interface type, fiber infrastructure, connector configuration and the supported channel specifications. Copper should not be selected solely because two cabinets are in the same row, and optical links should not be assumed necessary for every MoR connection.

    Connectivity OptionTypical ApplicationKey Selection Factors
    Passive DACShort connections, commonly within a rackSupported cable length, electrical interface and host compatibility
    AECShort-reach copper links requiring active signal conditioningPower, reach, latency, host support and cable specifications
    AOCOptical links with integrated transceivers at both endsReach, data rate, connector type and equipment compatibility
    Pluggable optical transceiversSwitch-to-switch or server-to-switch fiber linksWavelength, fiber type, reach, interface and optical budget

    For 100G, 400G and 800G deployments, the optical interface must match the switch port and the selected transmission medium. Engineers should verify the module form factor, lane configuration, wavelength, connector, fiber type, transmission distance and host compatibility before deployment.

    Relevant connectivity options are available through the C-LIGHT optical transceiver portfolio and the C-LIGHT product range, including optical and direct-attach connectivity categories.

    9. ToR and MoR in 100G, 400G and 800G Data Centers

    Higher-speed networking changes the economics of access-layer deployment. As server bandwidth increases, each rack may require more high-speed ports, higher-capacity uplinks and more attention to oversubscription. The number of connections and the physical reach of each link become important inputs to the ToR or MoR decision.

    In a ToR design, each rack can be upgraded independently as server bandwidth requirements increase. This offers flexibility for phased deployments, but it may also mean purchasing and managing more high-speed switches.

    MoR can consolidate high-density switching resources and allocate ports across several cabinets. This may be useful when server racks have uneven capacity requirements or when centralized network equipment is preferred. The challenge is ensuring that horizontal cabling, patching and switch capacity can support the required bandwidth without creating operational bottlenecks.

    For AI clusters and other bandwidth-intensive workloads, the physical placement model should be evaluated alongside the logical fabric design. Leaf-spine connectivity, uplink capacity, congestion management, path redundancy and application traffic patterns remain critical. Moving a switch from a server rack to a MoR cabinet does not, by itself, increase network bandwidth or reduce the number of logical switching stages.

    10. When Should You Choose ToR or MoR?

    ToR is often suitable when:

    • Server racks are deployed and expanded independently.

    • Short server-to-switch connections are a priority.

    • Rack-level fault isolation and management are important.

    • Horizontal cabling pathways are limited or difficult to expand.

    • Each rack has enough equipment to justify its access-switch capacity.

    MoR is often suitable when:

    • Several server racks share a common network zone.

    • Centralized switch management is preferred.

    • Higher port utilization can reduce unused access capacity.

    • Dedicated network cabinets and structured cabling are available.

    • Server racks need to reserve more space for compute equipment.

    There is no universal choice for every data center. A facility with small, independently managed server racks may benefit from ToR, while a larger deployment with repeatable row layouts and centralized cabling may find MoR more efficient. Some data centers also use a hybrid design, selecting different access-layer placements for different server zones or workload groups.

    11. Design Considerations Before Deployment

    Before selecting an access-switch placement model, infrastructure teams should evaluate the complete physical and logical design rather than comparing switch locations in isolation.

    • Rack population: Estimate current and future server counts, port requirements and expected rack density.

    • Bandwidth demand: Determine server-facing speeds, uplink capacity and acceptable oversubscription.

    • Cable reach: Calculate actual routed cable lengths and verify the supported limits for each selected interface and assembly.

    • Power and thermal capacity: Check rack power budgets, cooling capacity and equipment airflow requirements.

    • Redundancy: Design independent switch, power and cable paths where required by service availability objectives.

    • Operations: Account for labeling, patching, remote management, troubleshooting and equipment replacement.

    • Expansion: Reserve sufficient ports, cable pathways, rack space and upstream network capacity for future upgrades.

    These factors should be considered together. For example, MoR may reduce the number of access switches but increase horizontal cable requirements. ToR may simplify local cabling but introduce more distributed power and cooling loads. The preferred design is the one that meets the facility's performance, availability and lifecycle cost requirements.

    12. Conclusion

    ToR and Middle-of-Row architecture provide two ways to organize data center access switching. ToR prioritizes short local connections, rack-level deployment flexibility and distributed fault domains. MoR centralizes access switches, potentially improving port utilization and freeing server-rack space, while requiring more structured horizontal cabling and dedicated network infrastructure.

    The decision should be based on rack density, server connectivity, switch utilization, cable reach, power and cooling, redundancy and future expansion. For high-speed Ethernet and AI data center networks, the physical placement of access switches must also align with the optical and electrical connectivity strategy. Careful planning across these areas helps create a network infrastructure that is scalable, maintainable and ready for higher bandwidth requirements.

    For optical connectivity requirements, product selection or technical support, Contact C-LIGHT.

    For any questions, please contact us by email or WhatsApp.

    Email: sales@c-light.com

    WhatsApp: +86 132 6656 7067

    Related Articles

    Call
    Top