The Fragmented Internet
Addressing and Routing - BFFs
Internet Protocol addressing and routing are two concepts that go hand-in-hand. Addressing policies and strategies are first drafted and then implemented onto the physical or virtual network topology. Routers exchange advertisements about these address blocks to facilitate end-to-end reachability. It is sometimes difficult to distinguish between the addressing plan and the routing plan because they are “brothers in arms”.
Less Route Summarization Taking Place
Before the introduction of Classless Inter-Domain Routing (CIDR), routes often used their classful mask. Long gone are the Class A/B/C designations of public address space and the use of private addressing and NAT is completely pervasive. Now with Carrier Grade NAT (CGN) on the way, we are becoming concerned about the future condition of the Internet.
Today, we notice that fewer organizations are using summary statements in their router configurations. Some enterprise organizations are not effectively summarizing on their Internet edge. Furthermore, most enterprises do not do a good job of summarizing within their internal networks. Some network administrators may be unaware of the benefits of summarization and how it should be configured. In their defense, routers have faster processors and now they can have more neighbor adjacencies yet still converge the network sufficiently fast. Some organizations may be intentionally advertising more specific routes to alternate upstream ISPs. This may be due to IPv4 subnets being further broken apart and placed in different locations in the network topology.
Back in the mid-1990s top tier ISPs would not accept routes any smaller than a /18. In the mid-1990s, I remember hooking up my first Internet connection using a Cisco Systems AGS+ router with a T1 and configuring BGP to advertise a Class-B /16 prefix. Now those rules have been relaxed and ISPs are willing to accept smaller and smaller blocks or addresses from their customers. Enterprise organizations often ask their upstream ISPs to inject more specific routes from other provider-assigned address allocations. Due to increased IPv4 subnetting, there are more and more routes in the Internet routing tables and also within organization's private networks.
Larger Routing Tables – Coming to an Internet Near You
The size of the IPv4 Internet routing tables are growing at a steady pace. One of the best ways to visualize this “Growth of the BGP Table” is to consult Geoff Huston’s BGP page. Looking at this graph, you can see that the Internet’s routing table size growth has not been affected by the global economic recession of recent years.
Over time the IPv4 BGP tables will continue to grow. Consider the memory and CPU required by all routers in the core of the Internet when the BGP tables grows to 600,000 routes, or even 1,000,000 routes. This creates a drain on Ternary Content-Addressable Memory (TCAM) space in routers. Not just large ISPs, but smaller ISPs must continually acquire larger and more capable routers making the network device manufacturer’s shareholders very happy.
IPv4 Address Crunch Leads to Address Open Market
As IPv4 address space shortages becomes more of an issue, organizations will not be able to acquire IPv4 addresses from their Regional Internet Registry (RIR) or their service providers. Several of the RIRs are now operating in an extreme address conservation mode and have essentially run-out of IPv4 addresses to allocate to the organizations that operate within their geography. Most of the RIRs have policies for transferring ownership between organizations. Here are links to the ARIN, RIPE NCC, APNIC, LACNIC, and AfriNIC transfer policies.
Now that the RIRs permit address transfers, this will establish an open market of IPv4 address trading that will grow. As IPv4 address transfers will become popular there will be smaller and smaller blocks of addresses traded between organizations. Currently, the ARIN Number Resource Policy Manual (NRPM) states that the minimum transfer block size is a /24. Most of the other RIRs policies also have a /24 minimum transfer block size.
Longer Prefix Lengths Observed
The average IPv4 prefix length is increasing as IPv4 address crunch continues. In other words, smaller blocks of IPv4 addresses are showing up on the Internet. Years ago a /20 used to be the smallest subnet mask in general use. However, over time those rules were relaxed and now ISPs will accept a /24 prefix. Market pressures may even force ISPs to accept prefix lengths longer than a /24.
Geoff Huston’s Route-Views data gives us an indication of this increase in prefix length. Look at the graph on his site titled “Average Prefix Size”. You can see a subtle upward trend to longer prefix lengths.
Further down on his page is a section titled “Prefix Length”. His site has graphs for the number of /8s through /32s observed in the BGP tables. There is also a specific graph of the number of /24 routes and we see a definite rise in the number of /24s advertised each year. This could also be attributed to the natural growth of the Internet. However, if we look further down on his page we can see a subtle upward trend of much longer prefix lengths as well. We can also see on this page the number of prefixes that have lengths between /25 and /32. While these prefix lengths are not increasing similar to the pace of the /24s, it is still amazing to think that organizations are advertising those long prefixes into the core of the Internet. It is amazing to see that there are /27s, /28s, and even /29s advertised.
We expect to utilize the IPv4 Internet for decades to come. While IPv6 adoption is increasing and approximately doubling every year, it will take many years before the whole Internet is bi-lingual and all organizations are fully reachable over both IPv4 and IPv6 transport. In the meantime we should be aware of how our addressing issues are affecting our routing issues. We want organizations to behave like good Internet citizens because we all share this valuable resource. This bring to mind the word “ubuntu” which means “we are what we are because of all of us”. None of us wants to live in a swamp, so please consider your actions carefully before you deaggregate address blocks and advertise the more specifics.