Core Router
How Core Routers Move Traffic Across the Backbone
In a hierarchical service-provider network, routers are divided by role. The access and edge layers face customers and the radio access network, where provider-edge (PE) routers classify packets, apply access control, run BGP peering, and impose the MPLS labels that define each traffic flow. The core layer sits beneath this, interconnecting PE routers across long-haul fiber. A core router's job is deliberately narrow: take a labeled packet on one high-speed interface and forward it toward the correct PE on another, as fast and predictably as physically possible. Stripping policy out of the core lets the forwarding silicon be tuned for pure throughput, which is why core platforms reach densities that edge platforms cannot.
That throughput comes from distributed forwarding. Each line card carries one or more network-processor or merchant-silicon ASICs that hold a copy of the label-forwarding table and switch packets locally, connected to other cards through a multi-terabit switch fabric. A control plane built on OSPF or IS-IS computes the interior topology while LDP or segment routing distributes labels; the data plane then runs autonomously at line rate. Because every core hop is a single exact-match label lookup rather than a longest-prefix IP search against roughly a million internet routes, latency is low and deterministic, and the same mechanism enables traffic engineering and sub-50 ms protection switching.
For an RF and microwave audience, the core router is the destination that mobile traffic reaches after it leaves the antenna. Millimeter-wave and microwave point-to-point radios, fiber fronthaul, and 5G New Radio sites all feed cell-site routers, which aggregate into metro rings and finally into the core. The capacity planning of that core, the optics it terminates, and its redundancy design directly shape the end-to-end latency and availability budget of any wireless service.
Forwarding Throughput and Packet Rate
PPS = Link Rate / ((Frame + IFG + Preamble) × 8)
For 400G, 64-byte frames: PPS ≈ 400×109 / (84 × 8) ≈ 595 Mpps
Aggregate Chassis Capacity:
Ctotal = Ncards × Nports × Rport
Example: 16 cards × 36 ports × 400G ≈ 230 Tbps (half-duplex sum)
DWDM Fiber Capacity:
Cfiber = λcount × Rλ (e.g. 80 × 800G ≈ 64 Tbps per pair)
Where IFG = 12-byte inter-frame gap, Preamble = 8 bytes, Rport = per-port line rate, λcount = wavelengths per fiber, Rλ = per-wavelength rate. The 84-byte figure is 64-byte frame + 12 IFG + 8 preamble.
Core Router vs. Edge Router vs. Cell-Site Router
| Attribute | Core (P) Router | Edge (PE) Router | Cell-Site Router |
|---|---|---|---|
| Primary role | Backbone label switching | Customer/RAN aggregation, policy | Cell-site traffic grooming |
| Per-chassis capacity | 10 to 100+ Tbps | 1 to 20 Tbps | 100 Gbps to 1 Tbps |
| Typical port speeds | 100G / 400G / 800G | 10G / 100G / 400G | 1G / 10G / 25G / 100G |
| Routing tables held | IGP + label FIB only | Full BGP (~1M routes) + VRFs | IGP + limited static |
| Latency contribution | ~1 to 5 µs per hop | ~5 to 20 µs (with policy) | ~2 to 10 µs |
| Redundancy model | Dual RE/RP + N+1 fabric | Dual homing, VRRP | Ring/dual-uplink |
Frequently Asked Questions
What is the difference between a core router and an edge router?
An edge (PE) router faces customers and the radio access network: it classifies traffic, imposes MPLS labels, enforces ACLs and QoS, and runs full BGP. A core (P) router sits one layer deeper and mostly just swaps MPLS labels between PE routers, so its forwarding ASICs are tuned purely for throughput (10 to 100+ Tbps per chassis). Core routers carry fewer but much faster interfaces (100G, 400G, 800G), while edge routers expose many lower-speed customer ports.
How much throughput does a modern core router handle?
Each line card hosts forwarding ASICs; a 36-port 400G card moves 14.4 Tbps, and a fully loaded multi-slot chassis reaches 100 Tbps or more full-duplex. Forwarding runs at line rate, about 595 Mpps per 400G port for 64-byte frames, with no drops. With 800G coherent optics over DWDM, a single fiber pair carrying 80 wavelengths delivers more than 60 Tbps into one router.
Why do core routers use MPLS label switching instead of plain IP lookup?
A longest-prefix-match against roughly a million internet routes is costly to do at terabit rates on every hop. MPLS has the ingress edge classify the packet once and push a 20-bit label; each core router then does one exact-match lookup, swaps it, and forwards. This is deterministic and enables traffic engineering, sub-50 ms fast reroute, and VPN separation, so the core holds only the label FIB plus the IGP (OSPF or IS-IS) topology, not the full BGP table.