Abstract: This memo discusses strategies for address assignment of the existing IP address space with a view to conserve the address space and stem the explosive growth of routing tables in default-route-free routers run by transit routing domain providers.
Introduction
In June 1992, a group of network engineers published RFC 1338, a document that would fundamentally alter the architecture of the Internet. This memo outlined the strategy we now know as Classless Inter-Domain Routing (CIDR). By moving away from rigid "class-based" IP addressing and introducing hierarchical allocation, the authors proposed a way to prevent the early Internet's routing tables from collapsing under their own weight.
The Crisis of the Early 1990s
By the early 90s, the Internet was a victim of its own success, facing an existential threat from two directions: "Class B" address exhaustion and a routing table explosion. The system at the time relied on a "classful" addressing system:
- Class A: Huge blocks (16 million addresses), which were mostly already allocated.
- Class B: Mid-sized blocks (65,534 addresses), the "Goldilocks" choice for most organizations, but predicted to run out within two years.
- Class C: Small blocks (254 addresses), which were too small for most businesses, leading organizations to request multiple Class C blocks instead.
This created a massive scalability headache. If an organization used eight Class C blocks, routers had to maintain eight separate entries in their routing tables. These tables were doubling in size approximately every 10 months, threatening to overwhelm the memory and processing power of contemporary hardware.
Historical Growth and Data
The authors performed a "back-of-the-envelope" analysis using real-world data from the era, specifically looking at the MERIT Routing Data from the NSFNET backbone. The data showed a stark trend: the number of advertised routes was growing exponentially.
As seen in the data above, the routing table grew from a mere 173 entries in July 1988 to 4,775 entries by February 1992. Without intervention, the authors projected that routing tables would explode to over 75,000 entries in just a few years.
The Supernetting Strategy
The authors proposed a shift from random, flat address allocation to hierarchical, bitmask-oriented allocation, a method they termed "Supernetting."
- Address Aggregation: Instead of giving an organization a random Class B, a Service Provider (SP) is given a large "supernet" block (e.g., a block of 2048 Class C numbers). The SP then sub-allocates smaller chunks to its customers.
- Route Summarization: Because all those customers "live" within the SP's address space, the SP only needs to advertise one single "summary" route to the rest of the Internet.
- Longest-Match Routing: To make this work, routers moved away from class-based lookups to "longest-match" lookups. If a specific route exists for a customer (perhaps they are multi-homed), that specific route takes precedence over the general "supernet" advertisement.
- Loop Prevention: The paper introduces the "sink" route requirement—if a router summarizes a block, it must discard traffic for addresses within that block that don't have an explicit destination to prevent infinite routing loops.
Key Insights and Long-term Impact
The most profound insight of RFC 1338 is that IP addresses should follow the topology of the network. If an address is a subset of a provider's address, the global Internet doesn't need to know the specific organization; it only needs to know how to find the provider.
The authors also acknowledged the messiness of the real world. Organizations move providers or connect to multiple ones ("multi-homing"). They explained how longest-match routing allows these "anomalous" routes to exist by "punching a hole" in the aggregate advertisement without breaking the system.
Results and Limitations
The quantitative predictions were stark. At the time, routing tables were growing at 150% annually. The authors projected that with Supernetting, the annual growth rate would drop to only 6%, reaching only ~13,000 entries by 1995 instead of the feared 75,000.
History shows they were largely correct. However, the authors were humble; they didn't claim this solved the 32-bit limit (IPv4 exhaustion). They explicitly framed this as a "short to mid-term" strategy to keep the lights on while a long-term solution (which eventually became IPv6) was engineered. While aggregation remains difficult for multi-homed sites and re-addressing remains a "pain point," CIDR is what allowed the IPv4 Internet to scale to the billions of devices we see today.
Summary updated