The Dell Networking MX9116N is the Fabric Switching Engine for the PowerEdge MX7000 chassis: a non-blocking 25GbE switch in an MX Fabric A/B I/O sled, with sixteen internal 25GbE ports to the compute sleds, four external QSFP28 100GbE uplinks and twelve external QSFP28-DD ports, running SmartFabric OS10 across a 6.4Tbps fabric.
Read the port count carefully
The sixteen 25GbE ports are internal. They sit on the MX7000 midplane and connect to the mezzanine adapters in the compute sleds - there is no cage, no optic and no cable involved. What you can plug into on the faceplate is four QSFP28 ports at 100GbE and twelve QSFP28-DD ports, Dell's double-density QSFP28 that presents two 100GbE lanes through one connector. Listings that describe this module as a "16-port 25GbE SFP28 switch" are describing the internal fabric as if it were external connectivity, and it is worth being precise about because it changes what optics you need to buy.
Every internal port runs at line rate with no oversubscription. Sixteen sleds at 25Gb each is 400Gb of server-facing demand against a 6.4Tbps full-duplex fabric and 2,380Mpps of forwarding capacity, at sub-450-nanosecond latency. The module is non-blocking by design rather than by configuration.
What the QSFP28-DD ports are actually for
The twelve double-density ports are the reason this module exists rather than a plain in-chassis switch. They do four different jobs depending on what you plug into them:
- Additional uplinks beyond the four dedicated QSFP28 ports, when the chassis needs more than 400Gb northbound.
- Inter-chassis links between MX9116N modules, which is how a multi-chassis fabric is stitched together.
- Breakout to rack servers at 10GbE or 25GbE - a QSFP28-DD to 8xSFP28 cable turns one port into eight server connections, so an MX7000 can front a few rack-mount servers without a separate top-of-rack switch.
- Fabric expansion to as many as nine further MX7000 chassis through MX7116n Fabric Expander Modules.
That last one is the MX Scalable Fabric Architecture. A pair of MX9116N modules in one chassis can carry up to eighty compute sleds across up to ten chassis, with the other nine chassis holding cheap, passive MX7116n expanders instead of full switches. You buy switching once and extend it, rather than buying a switch pair per chassis. SmartFabric Services manages up to a twenty-chassis Multi-Chassis Management group as a single object.
Fibre Channel without a separate SAN switch
Two of the four QSFP28 ports can each be broken out into four 32Gb Fibre Channel connections, giving up to eight FC ports on the module. Those run either as a direct attachment to an FC storage array or as an NPIV Proxy Gateway into an existing FC SAN. For a chassis that needs both Ethernet and a modest amount of Fibre Channel, that removes the need for a separate FC I/O module entirely. Where the FC requirement is larger or the environment is a Connectrix or Brocade fabric with its own management expectations, the Dell Networking MXG610s is the dedicated 32Gb Fibre Channel module for the same chassis.
The module is fully IEEE data center bridging compliant - PFC, ETS, DCBx with iSCSI and FCoE application TLVs - and supports iSCSI, NAS and FCoE transit, RoCEv2 and routable RoCE, and is NVMe-over-Fabrics ready. That is the full converged feature set, not a subset.
SmartFabric OS10 and SmartFabric Services
OS10 is Dell's disaggregated network operating system: an unmodified Linux kernel and Debian userland underneath a hardened networking stack, with the switching hardware abstracted through OCP-SAI. Practically, that means the module supports the full routing set - OSPFv2 and v3, BGP-4 with four-byte ASN and BGP EVPN including VxLAN Layer 2 and Layer 3 gateways with symmetric and asymmetric IRB, VRRPv3, policy-based routing, BFD - alongside VLT and multiple-VLT for active-active multi-chassis link aggregation without spanning-tree blocking.
SmartFabric Services is the automation layer on top. It gives one management pane across every fabric in an MX deployment, integrates VLAN and QoS settings with server deployment templates so a new sled arrives on the right network, pushes fabric-wide firmware upgrades with consistency checks, validates the physical topology and tells you what is miscabled, and heals the fabric automatically once a failure condition clears. Management and telemetry cover SNMP v1/v2c/v3, RestConf APIs, gRPC streaming telemetry with gPB encoding, sFlow, SupportAssist phone-home, and Ansible, Puppet, Chef and SaltStack for configuration management.
Power, cooling and physical fit
The MX9116N has no power supply and no fans of its own - the MX7000 chassis supplies redundant power and cooling to the fabric slots. There is nothing to configure and nothing to spare-stock on that front, and there are no rails. Draw is 260W maximum with 5W QSFP28-DD optics installed, 237W typical. The sled measures 1.18 by 17.11 by 10.94 inches and weighs 8.49 lbs. Standard operating range is 10 to 35 degrees C; expanded-range continuous operation is explicitly not supported on this module, which is worth knowing if the chassis lives in a warm room.
A USB 2.0 Type A port takes configuration from a flash drive and a micro-USB Type B port provides console access.
Choosing between the MX I/O modules
The MX7000 fabric slots take several different modules and they are not interchangeable in role:
- MX9116N Fabric Switching Engine - the switch. 25GbE to the sleds, 100GbE uplinks, scalable fabric host (this page).
- MX7116n Fabric Expander - passive extension of an MX9116N's fabric into another chassis; it is not a switch and cannot stand alone.
- MXG610s - 32Gb Fibre Channel I/O module for chassis that need a real FC fabric rather than eight breakout ports.
If the platform is the older PowerEdge M1000e rather than the MX7000, the equivalent Ethernet modules are the PowerConnect M8024-K at 10GbE and the PowerConnect M6220 at 1GbE. M-series and MX-series modules share no compatibility.
Condition, testing and support
Every unit is professionally refurbished: inspected, cleaned, configuration-wiped, firmware-checked and functionally tested before it ships, and it ships with warranty. MX7000 fabric modules come to the secondary market in small numbers because the platform is still in production, so availability moves - if you need a matched pair for a redundant fabric, or a specific OS10 version, ask before ordering and we will confirm what is on the shelf.
Dell Networking MX9116N Specifications
Ports and Fabric
| Form factor | Full-featured 25/100GbE switch in PowerEdge MX Fabric A/B I/O sled form factor |
|---|---|
| Port configuration | 16x internal 25GbE server-facing ports, 4x external QSFP28 100GbE, 12x external QSFP28-DD, 1x USB 2.0 Type A, 1x micro-USB Type B console |
| Uplinks | 4x QSFP28 100GbE uplinks plus 12x QSFP28-DD ports usable as uplinks, inter-chassis links, 10/25GbE breakout to rack servers, or fabric expansion to up to 9 additional MX7000 chassis |
| Fibre Channel | Up to eight 32Gb FC connections - two of the QSFP28 ports carry four FC each - supporting both NPIV Proxy Gateway and Direct Attach FC configurations; FIP snooping |
| Scalable fabric | Up to 80 MX compute sleds across up to 10 MX7000 chassis via MX7116n Fabric Expander Modules; SmartFabric Services manages up to a 20-chassis Multi-Chassis Management group |
| Stacking | No traditional stacking - scale is achieved through the MX Scalable Fabric Architecture and VLT/mVLT rather than a stack ring |
Performance and Scale
| Switching I/O bandwidth | 6.4Tbps (full duplex), non-blocking |
|---|---|
| Forwarding capacity | 2380Mpps |
| Latency | Sub 450 nanoseconds |
| MAC addresses | 137K |
| IPv4 routes | 130K unicast; 48K ARP entries |
| IPv6 routes | 130K unicast |
| VLANs | 60K P*V Layer 2, 30K P*V Layer 3 in Full Switch mode |
| MST instances | 32 |
| PVST+ instances | 128 |
| Link aggregation | 128 LAG groups, 16 members per group, enhanced hashing |
| ACL entries | Layer 2 egress 1020, Layer 2 ingress 767, IPv4 egress 767, IPv4 ingress 767, IPv6 egress 512, IPv6 ingress 767 |
| iSCSI sessions | 256 |
| Jumbo frames | Yes - 9K |
Software, Routing and Convergence
| Operating system | Dell SmartFabric OS10 (Debian Linux 8, kernel 3.16); ONIE-capable |
|---|---|
| Routing | Line-rate Layer 2 and Layer 3, IPv4 and IPv6. OSPFv2/v3, BGP-4 with 4-byte ASN, BGP EVPN with VxLAN L2/L3 gateway, VLT and mVLT multi-chassis link aggregation, VRRPv3, PBR, BFD |
| Data center bridging | PFC (802.1Qbb), ETS (802.1Qaz), DCBx, DCBx application TLV for iSCSI and FCoE, explicit congestion notification, RoCEv2 and routable RoCE |
| Storage readiness | NVMe-oF ready |
| SmartFabric Services | Single-pane management and automation across every fabric in an MX deployment - I/O aggregation, VLAN and QoS integration with server deployment templates, fabric-wide firmware upgrades, automatic topology validation, automatic healing on failure clearance |
| Quality of service | Rate shaping (egress), rate policing (ingress), prefix list, route-map, round robin / weighted round robin / deficit round robin / strict priority scheduling, weighted random early detect |
| Mirroring | Local mirroring, Remote Port Mirroring (RPM), Encapsulated Remote Port Mirroring (ERPM), port mirroring over VLT |
| Rogue NIC control | Hardware-based protection against NICs sending excessive pause frames |
| Management protocols | SNMP v1/v2c/v3, Telnet, FTP, TACACS+, RADIUS, SSHv2, NTP, RestConf APIs, gRPC streaming telemetry, sFlow (RFC 3176), SupportAssist |
| Automation | Control Plane Services APIs, Ansible, Puppet, Chef, SaltStack, CLI multiline alias automation, Zero Touch Deployment, third-party packages in Docker containers |
Physical, Power and Environmental
| Dimensions (H x W x D) | 1.18 x 17.11 x 10.94 in |
|---|---|
| Weight | 8.49 lbs |
| Maximum power consumption | 260W with 5W QSFP28-DD optics |
| Typical power consumption | 237W with 5W QSFP28-DD optics |
| Power and cooling | Redundant power and cooling supplied by the PowerEdge MX7000 chassis - the module has no power supply of its own |
| Indicators | Power/health LED, ID LED, link/activity LEDs |
| Operating temperature | 50 to 95 F (10 to 35 C) |
| Operating relative humidity | 5% to 85% non-condensing |
| Storage temperature | -40 to 149 F (-40 to 65 C) |
| Storage relative humidity | 5% to 95% non-condensing |
| Expanded operating temperature | Continuous operation not supported |
| Regulatory compliance | UL/CSA 60950-1 2nd ed., EN 60950-1 2nd ed., IEC 60950-1 2nd ed., EN 60825-1 and 60825-2 laser safety, FDA 21 CFR 1040.10 and 1040.11; FCC Part 15 Class A, ICES-3 Class A, EN 55024 Class A, AS/NZS CISPR 32 Class A, VCCI Class A, EN 300 386; EN 50581 RoHS compliant |

