IPv6 Network - Interdomain Routing Guidelines

Around the beginning of 2005, the IPv4 global routing table reached a size of 150,000 prefixes, even though there were fewer than 20,000 active ASes. (For IPv6, the numbers were 700 and 500, respectively, at that time.) During the past 15 years, there were several times that router hardware could only barely keep up with the growth of the global routing table. The reason for this state of affairs is manifold, with factors being:

The main approach to avoid these problems in IPv6 was to give out very large address blocks to ISPs in order to avoid fragmenting their IPv6 BGP announcements. So while a small ISP may receive between a /21 and a /19 initially in IPv4, all ISPs get a /32 in IPv6. Even though an end-user gets a /48, no questions asked, in IPv6, a /32 serves some 65,000 customers. A /19 in IPv4 on the other hand, is enough to supply no more than some 8000 customers with just a single IPv4 address. Really large ISPs can even get initial IPv6 blocks that are bigger than a /32. Unfortunately, there is no easy way to limit the number of prefixes in the global routing table due to multihoming. The IETF is currently working on ways to achieve multihoming benefits without the need to run BGP and inject a prefix into the global routing table, but it’s slow going. In the mean time, there are no provisions for multihoming in IPv6.

Only ISPs that plan on giving out IPv6 address space to at least 200 customers in two years and operators of critical Internet resources (think root nameservers) can obtain independent IPv6 address blocks. (See Chapters 2 and 11 for more details on obtaining address space.) Everyone else has to obtain addresses from their ISP and renumber when changing ISPs. However, there is considerable pressure to allow provider-independent (PI) address space in IPv6 so that end users can multihome and/or move from ISP to ISP while taking their address space with them. A practical issue is how to filter.

In IPv4, address blocks that are smaller than a /24 are filtered out routinely. So someone who wants to multihome or announce a prefix for some other reason, must at least obtain a /24. This requires fairly extensive documentation of a real-world need for so many addresses, which is a significant hurdle. In IPv6, on the other hand, anyone can get a /48, so if those aren’t filtered out, the IPv6 global routing table will most likely become very large, much larger than the IPv4 table. A solution to this would be to give out address blocks that are larger than a /48 to people who want to multihome and/or want PI address space.

However, then the question becomes: who deserves such a bigger block, and who doesn’t? If you would like to filter out /48s, be careful that several Regional Internet Registries (RIRs) give out /48 blocks to Internet exchanges so the routers of the IX members can interconnect by using independent address space. These prefixes can be filtered out without problems because they’re only relevant to networks that connect to those Internet exchanges. However, ARIN, the American Registry for Internet Numbers, also gives out /48 for “critical Internet infrastructure,” as explained on http://www.arin.net/reference/micro_allocations.html. These micro allocations currently come from the 2001:500::/30 block. Listing 4-39 implements a prefix length filter that takes micro allocations into consideration.

Tags: Add more tags...,

or Close

Comment an article   Name   E-mail    Title Available characters: 4000  Notify me of follow-up comments Enter what you see:

No comment posted