1. The Hierarchy of Trust: Administrative Distance

A router might receive the same path information from multiple protocols. How does it decide which one to believe? It uses Administrative Distance ($\\text{AD}$).

Lower is better. If $\\text{OSPF}$ says to go left and a static route says to go right, the router goes right.

2. $\\text{IGP}$: Intelligence Within the Walls

Interior Gateway Protocols are designed for speed and reliability within your own network.

$\\text{OSPF}$: The Link-State Master

$\\text{OSPF}$ (Open Shortest Path First) uses the Dijkstra Algorithm. Every router has a complete "map" of the entire network. If a link goes down, the map is updated, and the router recalculates the shortest path using bandwidth as a cost.

$\\text{EIGRP}$: The Hybrid Legend

Originally a Cisco proprietary protocol, $\\text{EIGRP}$ uses DUAL (Diffusing Update Algorithm). It is faster than $\\text{OSPF}$ in specific scenarios because it keeps "feasible successors" (backup routes) pre-calculated and ready for instant use.

3. $\\text{BGP}$: The Protocol that Runs the World

$\\text{BGP}$ is a Path-Vector protocol. It doesn't care about bandwidth. It cares about Policy. $\\text{BGP}$ is how $\\text{ISPs}$ talk to each other.

The $\\text{BGP}$ Decision Process (Simplified)

1. Weight: Highest is preferred (Local to the router).
2. Local Preference: Highest is preferred (Local to the Autonomous System $\\text{AS}$).
3. $\\text{AS-Path}$: Shortest (fewest number of $\\text{AS}$ hops).
4. Origin: $\\text{IGP}$ is preferred over $\\text{EGP}$.
5. $\\text{MED}$: Lowest is preferred (Multi-Exit Discriminator).

4. Redundancy & First Hop Protocols

Routing isn't just about cross-country paths; it's about the "Last Mile." $\\text{HSRP}$ (Hot Standby Router Protocol) and $\\text{VRRP}$ (Virtual Router Redundancy Protocol) allow two physical routers to share a single "Virtual $\\text{IP}$". If one dies, the other takes over in milliseconds, and the end-users never notice.

5. Routing for Security: $\\text{ACLs}$ and $\\text{VRFs}$

Modern routing includes Virtual Routing and Forwarding ($\\text{VRF}$). This allows a single router to maintain multiple independent routing tables. It's like having separate physical routers for your Guest $\\text{Wi-Fi}$ and your Management network, all sharing the same $\\text{CPU}$.

6. The Shift to $\\text{SD-WAN}$

We are moving away from "Command Line Routing" toward $\\text{SD-WAN}$. $\\text{SD-WAN}$ builds an overlay on top of the physical internet, dynamically switching traffic between $\\text{MPLS}$, $\\text{LTE}$, and Broadband based on real-time latency and packet loss.

7. FIB vs RIB Architecture: The Hardware/Software Divide in Packet Forwarding

The fundamental architectural abstraction in any router is the separation between the Routing Information Base (RIB) and the Forwarding Information Base (FIB). The RIB is the control-plane data structure — it contains all routes learned from all routing protocols, including multiple paths to the same destination, before the best-path selection algorithm has chosen a winner. The FIB is the data-plane data structure — a distilled, optimized subset of the RIB that contains exactly one best next-hop per destination prefix, formatted for a high-speed ASIC lookup.

The RIB, maintained by the Route Processor (RP) CPU, is a rich database. Each entry includes administrative distance, metric, tag, next-hop reachability status, and optional policy attributes. In a router with 1.5 million BGP routes (typical for a Tier-2 ISP), the RIB consumes approximately 1.2 GB of DRAM — roughly 800 bytes per route. The FIB, by contrast, is optimized for the TCAM or DRAM-based lookup engine of the forwarding ASIC. A typical IPv4 FIB entry fits in 72 bits (32-bit prefix + 32-bit next-hop + 8-bit output interface index).

The FIB Population Protocol: From RIB to Hardware

The path from RIB to FIB follows a strictly ordered pipeline:

1. Protocol Update: OSPF floods an LSA, BGP sends an UPDATE, or a static route is configured. The protocol-specific database stores it.
2. RIB Insertion: The protocol daemon passes the route to the RIB manager via an internal inter-process communication (IPC) mechanism — often a proprietary socket-based API or shared memory.
3. Best-Path Resolution: The RIB manager runs the selection algorithm (lowest AD, then lowest metric, then protocol-specific tie-breakers).
4. FIB Update: The RIB manager computes the FIB delta — the set of changes between the old FIB and the new FIB — and pushes it to the forwarding ASIC via a low-level driver API (e.g., $\\text{Cisco FMAN-FP}$, $\\text{JUNOS RTABLE}$).
5. Hardware Programming: The ASIC microcode writes the new entries into the TCAM or algorithmic lookup table. This step is rate-limited to prevent ASIC buffer overflows.

The total RIB-to-FIB convergence time is the sum of these five stages. A single BGP route withdrawal in a router with 500,000 prefixes can take 50–200 ms just for the FIB delta computation and TCAM update. During this window, the data plane may still be forwarding using the stale FIB — a deliberate design choice that trades momentary black-holing for CPU stability.

8. Route Summarization Mechanics: Aggregation, Black-Holing, and Prefix Suppression

In a network with 1,000 subnets, every routing protocol must carry 1,000 prefixes in its LSDB or BGP table. Route Summarization — also called Aggregation or Supernetting — collapses multiple contiguous prefixes into a single, shorter-prefix advertisement. This reduces RIB size, FIB TCAM consumption, and, critically for OSPF and IS-IS, the number of SPF recalculations triggered by a single endpoint failure.

Bitwise Aggregation: The Prefix Mathematics

Summarization works by finding the common prefix bits across a set of networks. For four consecutive subnets $10.1.0.0/24$, $10.1.1.0/24$, $10.1.2.0/24$, and $10.1.3.0/24$, the third octet in binary is:

The summary route $10.1.0.0/22$ replaces four /24 entries in the FIB, saving three TCAM slots. On an ASIC with 1 million TCAM entries, this is negligible, but aggregated across an entire data center with 64,000 /32 host routes, the savings translate to a 75% reduction in FIB memory — the difference between a $50,000 line card and a $150,000 one.

The Black-Hole Risk and the Null0 Route

Summarization introduces a risk: when you advertise a summary, downstream routers assume every subnet within that summary range is reachable through you. If a specific subnet within the summary exists on one side of the network and you summarize on the ABR or ASBR, traffic from the far side of the network to a non-existent subnet within the summary range is forwarded to you, then dropped because you have no specific route. This is called a Black-Hole.

To prevent this, engineers configure a Null0 discard route for the summary prefix on the aggregating router. The Null0 route absorbs traffic that matches the summary but not any more-specific route, generating an ICMP unreachable message instead of silently dropping the packet:

Conclusion: The Map is the Territory

Understanding routing is understanding the flow of information in our society. From the simple logic of a home router to the massive $\\text{BGP}$ tables of Cloudflare and Google, the principles of metrics, trust, and path selection are what keep the digital world turning.

Frequently Asked Questions