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

ToR vs MoR: Data Center Switch Placement, Cabling and Network Design

By C-LIGHT Marketing 丨 Jul 7, 2025
Table of Contents

    Data center access switching must balance server connectivity, cable management, port utilization, and network expansion. As rack density increases, the physical location of access switches becomes an important infrastructure decision. Top-of-Rack (ToR) and Middle-of-Row (MoR) are two deployment models that distribute switching equipment differently across server racks.

    ToR places access switches inside individual server racks, keeping server-to-switch connections short and localized. MoR moves access switches into dedicated network cabinets near the middle of a server row, allowing multiple nearby server racks to share switching resources. This arrangement can reduce the longest horizontal cable runs associated with centralized switching while avoiding the need to install a separate access switch in every server cabinet.

    The choice affects more than cable distance. Switch quantity, port utilization, rack power, cooling, maintenance, redundancy, and optical connectivity all influence the suitability of each design.

    1. What Is ToR (Top-of-Rack) Switching?

    Top-of-Rack switching is a data center access model in which one or more network switches are installed inside each server rack, commonly near the top. Servers, storage devices, and other equipment in that rack connect directly to the local switch through short copper or fiber-optic links.

    The ToR switch aggregates traffic from the rack and forwards it to the upstream network. In a spine-leaf architecture, the ToR switch commonly performs the leaf role and connects to multiple spine switches through high-speed uplinks.

    Typical characteristics of ToR deployment include:

    • Local switching: Servers connect to access switches within the same rack.

    • Short server connections: DAC, AOC, copper Ethernet, or short optical links can be selected according to interface and distance requirements.

    • Distributed equipment: Each rack has its own access-switch capacity.

    • Independent rack expansion: New racks can be deployed with their own switches and connected to the network fabric.

    • Rack-level maintenance: Access-switch ports and local server connections can be managed within individual cabinets.

    ToR is especially common in standardized server environments where each rack is designed as a relatively independent deployment unit. It also simplifies the physical relationship between servers and their access switches.

    2. What Is MoR (Middle-of-Row) Switching?

    Middle-of-Row switching places access switches in a dedicated network cabinet near the middle of a server row. Rather than installing switches in every server rack, the MoR cabinet provides network access for several nearby racks.

    Servers connect to the centralized MoR switches through horizontal cabling routes. Depending on the facility design, these connections may use copper Ethernet, fiber-optic cabling, patch panels, or structured cabling systems.

    MoR is intended to balance the cable distances and equipment distribution of a row-based network. Because the switch cabinet is positioned between the server racks it serves, connections from racks on either side can generally be shorter than connections to a switch located at one end of the row.

    Common MoR characteristics include:

    • Centralized access switching: A dedicated network cabinet serves multiple server racks.

    • Moderate cable distances: Horizontal connections are distributed to racks on both sides of the switching location.

    • Shared port capacity: Switch ports can be allocated across several server cabinets.

    • Reduced switch count: Fewer access switches may be required compared with installing switches in every rack.

    • Centralized maintenance: Network equipment is concentrated in designated cabinets.

    MoR is particularly relevant when server racks have different equipment densities or when centralized switching provides better port utilization. It also requires careful planning for cable pathways, network cabinet placement, power, and cooling.

    3. ToR vs MoR: Key Differences

    The main distinction between ToR and MoR is the physical location of access switching. ToR distributes switches across individual server racks, whereas MoR consolidates them near the center of a row.

    FeatureToR (Top-of-Rack)MoR (Middle-of-Row)
    Switch locationInside or near the top of each server rackIn dedicated network cabinets near the middle of a row
    Switch distributionDistributed across individual racksCentralized for groups of nearby racks
    Server cablingShort connections within each rackHorizontal connections between server racks and MoR cabinets
    Cable distanceGenerally shortest for local server connectionsUsually shorter than EoR for racks on both sides of the MoR location
    Number of switchesUsually more switches across the facilityPotentially fewer switches serving multiple racks
    Port utilizationUnused ports may remain in lightly populated racksPorts can be shared across multiple server racks
    Rack powerSwitches add power demand to server racksAccess-switch power is concentrated in network cabinets
    CoolingSwitch heat contributes to server-rack thermal loadDedicated cooling may be required for network cabinets
    Cable managementRelatively simple local server cablingRequires organized horizontal cable distribution
    ExpansionCan add access capacity rack by rackRequires available centralized ports and cable capacity
    MaintenanceMore distributed devices to configure and maintainAccess equipment is concentrated in fewer locations
    Important: MoR and EoR are physical deployment models, not separate Ethernet forwarding protocols. Both can participate in a spine-leaf fabric when their access switches perform the leaf role and connect to the upstream spine layer.

    4. How ToR and MoR Network Layouts Work

    The physical layout shows why the two models have different cable requirements. In ToR, each server rack has a local switch. In MoR, multiple server racks connect to shared switches in a network cabinet near the center of the row.

    ToR deployment layout

    Top-of-Rack Architecture
    Rack 1
    ToR Switch
    Server 1
    Server 2
    Server 3
    Rack 2
    ToR Switch
    Server 1
    Server 2
    Server 3
    Rack 3
    ToR Switch
    Server 1
    Server 2
    Server 3

    Each server rack has its own access switch and local server connections.

    Upstream Network / Spine Layer

    Each ToR switch aggregates traffic from its local servers before forwarding it upstream. This reduces the need for long horizontal server cables, but the data center must manage a larger number of access switches and their uplinks.

    MoR deployment layout

    Middle-of-Row Architecture
    Rack 1
    Server 1
    Server 2
    Server 3
    Rack 2
    Server 1
    Server 2
    Server 3
    MoR Network Rack
    Access Switches
    Patch Panels
    Uplink Ports
    Rack 4
    Server 1
    Server 2
    Server 3
    Rack 5
    Server 1
    Server 2
    Server 3

    Several nearby server racks connect to shared access switches in the middle network cabinet.

    Upstream Network / Spine Layer

    The MoR cabinet provides a shared connection point for nearby server racks. The network team can allocate switch ports across those racks, while horizontal cable routes carry server traffic to the centralized access equipment.

    5. MoR vs EoR: Understanding the Difference

    MoR and EoR share the same general principle: access switches are centralized outside individual server racks. The main difference is the location of the network cabinet within the row.

    EoR places switching equipment at one end of the row. MoR moves the equipment toward the middle, allowing connections to extend in both directions. This can reduce the maximum cable distance for racks near the opposite end of the row.

    FeatureMoREoR
    Switch locationNear the middle of the rowAt one end of the row
    Cable distributionRuns toward racks on both sidesRuns from the end cabinet toward the entire row
    Maximum cable distanceOften lower for a comparable row layoutCan be longer for racks farthest from the cabinet
    Network cabinet placementRequires space near the row centerUses space at the row end
    Port consolidationCentralizes access ports for nearby racksCentralizes access ports for the row
    Design considerationCenter placement, aisle layout, and access to both sidesLonger cable pathways and end-of-row space

    MoR does not automatically halve the cable distance compared with EoR. The actual result depends on row length, network-cabinet position, rack spacing, and cable routing. For an idealized straight row with evenly distributed racks, placing the switch cabinet near the center can reduce the longest horizontal run substantially.

    6. Cabling Requirements and Infrastructure

    Cabling is a major factor in the ToR and MoR decision. ToR keeps most server connections inside individual racks, while MoR requires structured horizontal cabling between server cabinets and centralized network equipment.

    ToR cabling considerations:

    • Short copper or optical connections between servers and local switches.

    • Reduced horizontal server cabling between separate racks.

    • Fiber uplinks from individual ToR switches to upstream network devices.

    • Consistent cable labeling and rack-level patching across the facility.

    MoR cabling considerations:

    • Horizontal cable routes from multiple server racks to the network cabinet.

    • Structured copper or fiber distribution according to interface speed and distance.

    • Patch panels and cable-management systems for organized cross-connects.

    • Enough pathway capacity for future server additions and higher-speed links.

    • Separation of redundant paths where required by the network design.

    Cabling FactorToRMoR
    Server-to-switch distanceShort, within the rackHorizontal, between racks
    Horizontal cable volumeGenerally lower for server accessHigher because multiple racks connect to shared switches
    Patch-panel usageMay be limited for local connectionsOften useful for centralized structured cabling
    Cable pathway planningFocuses on local rack routing and uplinksRequires routes from both sides of the row
    Connection changesUsually handled within the server rackMay involve centralized patching and switch-port allocation
    Future cable capacityExpanded at individual racksMust be reserved in shared pathways and network cabinets

    The cable medium should be selected according to actual interface requirements, reach, installation conditions, and supported standards. Copper may be appropriate for short server connections, while optical fiber provides flexibility for longer horizontal links and high-speed uplinks.

    7. Port Utilization, Switch Count and Cost

    ToR and MoR have different effects on the number of switches required and the utilization of available access ports.

    In a ToR design, each rack receives dedicated switch ports. This makes it easy to deploy and expand individual racks, but some ports may remain unused when a rack has fewer servers than expected. Across a large facility, the number of switches, power supplies, uplinks, and management endpoints can become substantial.

    MoR consolidates ports across multiple racks. If one rack is nearly full and another has spare capacity, shared switching resources may make port allocation more flexible. This can improve utilization, although it also introduces the need to manage horizontal cabling and centralized network equipment.

    Cost ComponentToRMoR
    Switch hardwareMore individual access switches may be neededFewer high-density switches may serve multiple racks
    Port utilizationUnused capacity can remain in lightly populated racksPorts can be allocated across several racks
    Horizontal cablingGenerally less extensive for server accessMore cable routes and structured cabling components
    Power distributionSwitch power distributed among server racksPower concentrated in network cabinets
    Cooling infrastructureAdditional heat inside server racksDedicated cooling capacity for network cabinets
    Management workloadMore individual switches to maintainFewer centralized access devices may simplify operations

    There is no universal cost advantage. ToR can reduce horizontal cabling and support incremental deployment, while MoR can reduce switch count and improve port sharing. The total cost should include equipment, transceivers, cables, patch panels, installation, power, cooling, maintenance, and future upgrades.

    8. Power, Cooling and Rack Density

    Switch placement also affects how power consumption and heat are distributed across a data center.

    In ToR architecture, access switches are installed alongside servers. This increases the power and cooling requirements of individual server racks. In high-density environments, where servers already consume substantial power, the additional switch load must be included in rack-level capacity planning.

    MoR moves access-switch power and heat into dedicated network cabinets. This can free server racks from some networking equipment and make server-rack power planning more straightforward. However, the network cabinets require their own power distribution, cooling capacity, and physical space.

    Engineers should consider:

    • Available power capacity in server racks and network cabinets.

    • Switch power consumption at the expected port configuration.

    • Airflow direction and cooling compatibility.

    • Heat concentration in centralized network cabinets.

    • Redundant power feeds and power-supply arrangements.

    • Future server and switch density increases.

    For AI and high-performance computing environments, this consideration becomes particularly important. GPU servers can create high rack-level power and thermal loads, so moving access switches to a separate network cabinet may help distribute equipment heat. However, the longer or more complex cabling arrangement must still satisfy the required link performance and redundancy.

    9. Network Performance and Redundancy

    ToR and MoR do not inherently determine network latency or forwarding performance. These characteristics depend on switch architecture, port speed, forwarding capacity, oversubscription, traffic patterns, and the logical network topology.

    ToR provides direct local switching for servers connected to the same rack switch. Traffic between different racks normally travels through the upstream network. In a spine-leaf fabric, each leaf connects to the spine layer to provide paths between racks.

    MoR can allow servers connected to the same centralized access switch to communicate through that switch. However, the actual forwarding path depends on port allocation, VLAN or routing design, and how the network is divided into access and upstream layers.

    Redundancy planning should include:

    • Switch redundancy: Deploy independent access switches where availability requirements justify it.

    • Server connectivity: Use suitable dual-homing or multihoming designs for servers with multiple network interfaces.

    • Uplink redundancy: Connect access switches to independent upstream devices and paths.

    • Physical path separation: Avoid placing redundant cables in the same vulnerable pathway wherever practical.

    • Power resilience: Provide independent power sources and appropriate backup arrangements.

    MoR centralizes switching for multiple racks, so a failure in a shared access device can affect a larger group of servers. Redundant switching and carefully planned failure domains are therefore important. ToR distributes access switching, but a failed ToR switch can still affect all servers that depend on it without an alternate connection.

    10. Optical Connectivity in ToR and MoR Networks

    Optical connectivity supports high-speed links between access switches and upstream network equipment. The requirements differ according to switch placement, horizontal cable distance, interface speed, and the physical fiber infrastructure.

    In ToR deployments, optical transceivers are commonly used for uplinks from rack-level switches to the spine or aggregation layer. Short server connections may use DAC or AOC assemblies when supported by the endpoint equipment and link distance.

    In MoR deployments, optical fiber may connect multiple server racks to centralized access switches. The required optical reach depends on the row dimensions, cable routing, patch-panel locations, and the selected transceiver standard.

    Connectivity SolutionTypical RoleKey Selection Factors
    DACShort connections between nearby servers and switchesSupported interface, cable length, power, and endpoint compatibility
    AOCShort-to-medium optical connections between compatible devicesData rate, reach, cable routing, and endpoint support
    100G Optical TransceiverHigh-speed access and fabric uplinksWavelength, fiber type, connector, reach, and link budget
    400G Optical TransceiverHigh-capacity connections between access and upstream switchesForm factor, Ethernet standard, lane configuration, and optical interface
    800G Optical TransceiverHigh-bandwidth data center and AI network fabricsSwitch support, module power, cooling, fiber layout, and reach

    For data center deployments, C-LIGHT offers a range of optical transceiver solutions for high-speed Ethernet connectivity. Module selection should be based on the exact switch model, port speed, optical standard, wavelength, fiber type, connector, transmission distance, and optical link budget.

    MoR designs should also account for the additional optical loss introduced by patch panels, connectors, and intermediate cross-connects. Fiber polarity, connector cleanliness, cable labeling, and link testing are important for reliable centralized cabling systems.

    11. When Should You Choose ToR or MoR?

    The right deployment model depends on server density, row dimensions, available network cabinet space, cable pathways, and operational priorities. Neither architecture is universally better.

    ToR is often suitable when:

    • Server racks are deployed independently or incrementally.

    • Short local server connections are a priority.

    • Rack-level network management and fault isolation are important.

    • There is sufficient power and cooling capacity in server racks.

    • The facility uses a distributed spine-leaf architecture.

    MoR is often suitable when:

    • Several adjacent server racks can share access-switch capacity.

    • Switch port utilization is important across multiple cabinets.

    • Centralized network equipment is preferred for maintenance.

    • The facility has suitable horizontal cable pathways.

    • Dedicated network cabinets can accommodate switching, power, and cooling.

    Deployment RequirementPotentially Suitable ModelReason
    Independent rack-by-rack expansionToRAccess capacity can be added with each new rack.
    Short server-to-switch connectionsToRMost local connections remain inside the rack.
    Shared switch capacity across racksMoRAccess ports can be allocated across multiple cabinets.
    Reduced switching equipment inside server racksMoRAccess switches are moved to dedicated network cabinets.
    Centralized maintenanceMoRNetwork equipment is concentrated in fewer locations.
    Very short local server cablingToRServer and access-switch ports are in the same rack.
    Balanced cable distances across a rowMoRSwitches can serve racks on both sides of the network cabinet.

    12. Conclusion

    ToR and MoR are two approaches to positioning access switches in data center infrastructure. ToR places switches in individual server racks, providing short local connections and straightforward rack-level expansion. MoR centralizes access switching near the middle of a row, allowing several racks to share network equipment and balancing horizontal cable distances.

    ToR is often attractive for modular deployments, dense server racks, and distributed spine-leaf environments. MoR can be a practical option when centralized switching, shared port capacity, and reduced network equipment inside server racks are important.

    The final decision should consider more than physical switch placement. Cable infrastructure, port utilization, power, cooling, redundancy, traffic patterns, and future bandwidth requirements all affect the long-term performance and cost of the network.

    For high-speed Ethernet and data center optical connectivity requirements, visit C-LIGHT Optical Transceivers or 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