IPv6 Addressing Plans
From ARIN IPv6 Wiki
This page is for developing a best practice guide to IPv6 addressing plans for an ISP allocation. These are the allocations that are a /32 prefix or shorter. To some extent, a company that receives a /20 or /22 will be doing some things a bit differently due to scale, nevertheless, the same principles apply to developing the plan.
Where possible, a guideline should give its rationale and reference any supporting documentation.
Please do not add critiques or extensive discussions to this page. There is a discussion tab at the top of the page for these things to get worked out on a separate page. In some cases, where there is no clear consensus on some aspect of the guidelines, we will post both positions on this page and copy the discussion material to a separate background page that focuses on the one issue.
Although this document contains input from several people including discussions on the NANOG list, it is not necessarily the last word in IPv6 addressing plans. If you have suggestions for improvement, please post it to the discussion page (see the tab above).
- Separate address block for infrastructure from other uses (enterprise, loopbacks)
- May mean two /48s per PoP
- Document so that you can justify it in your Host Density ratio
- Each individual site should receive plenty of subnets, such as between a /48 and /56. See RFC 6177.
- Summary aggregates for groups of sites where it makes sense, but watch your HD ratio
- Any prefixes shorter than /48 will only be assigned when there is written justification to show that this prefix will meet the RIR HD ratio guidelines within 5 years.
- Each PoP is a site therefore assign a /48 for infrastructure
- No subnets will use prefixes longer than /64.
- Separate address block for router loop-back interfaces
- Generally number all loopbacks out of one /64
- /128 per loopback
- Note that this recommendation violates RFC 4291 - "IP Version 6 Addressing Architecture", which states that for addresses with other than binary 000 as their first 3 bits, the Interface Identifier must be 64 bits long. It isn't really necessary to save IPv6 address space by using /128s on loopbacks. For example, out of a /48, if you allowed for 16384 /64s for loopbacks, you'd still have 49 152 subnets left for links. If your network is big enough that that sort of addressing plan is not going to be large enough, then you probably won't have any issues with getting multiple /48s e.g. a /47 or /46.
- Assign a /64 per LAN / VLAN / subnet
- Organizations with multiple /48 allocations should consider enterprise-wise aggregation levels of /60 or larger blocks for the administration of enterprise policies for common functions such as:
- Realtime traffic, such as voice & video
- Network loopback addresses and Link space
- IETF expects that you will assign a /64 for point-to-point links
- Fewer typos because all subnets are the same size
- You can use longer prefixes but what's the point?
- /126 will break Mobile IPv6 Home Agent discovery
- /112 leaves final 16 bits free for Node IDs
- Use /64 unless you have read and understand RFC 3627
- Note: on pure point-to-point links (e.g., SONET) anything shorter than /127 is vulnerable to ping-pong packet amplification as described in Maz's APNIC 26 presentation. (On Ethernet, this is at most a neighbor cache DoS)
- The enterprise network should receive a prefix sufficient to provide a /48 allocation for each site (office/campus/PoP) at which the company has employees or systems.
- All customers get one /48 unless they can show that they need more than 65k subnets.
- Host count is irrelevant.
- Do not assign to customers from PoP aggregates
- Define aggregate areas which contain several PoPs
- Carry customer networks in iBGP
- Aggregate only in eBGP
- If you have lots of consumer customers you may want to assign /56s to private residence sites.
- Expect the registry to allocate a /32 and reserve one /32
- Plan for the time when you get a second allocation giving you a /31 aggregate.
- If you get more than /32 first time round, ask the RIR how much is reserved so you can plan appropriately.
- If you need private addresses, generate a ULA prefix as defined in RFC 4193
- Use this handy web tool to generate one
- Add it to the registry at the above site, if you want people to know that this is your private space
- Make sure your internal registry people are aware of your ULA prefix(es) so that everybody uses it.
- Point-to-point links may be assigned a /126 prefix if there is written assurance that the drawbacks documented in RFC 3627 will not occur.
The following is an excerpt from RFC 5375, section 2.4:
IPv6 provides network administrators with a significantly larger address space, enabling them to be very creative in how they can define logical and practical address plans. The subnetting of assigned prefixes can be done based on various logical schemes that involve factors such as: o Using existing systems * translate the existing subnet number into IPv6 subnet id * translate the VLAN id into IPv6 subnet id o Rethink * allocate according to your need o Aggregation * Geographical Boundaries - by assigning a common prefix to all subnets within a geographical area * Organizational Boundaries - by assigning a common prefix to an entire organization or group within a corporate infrastructure * Service Type - by reserving certain prefixes for predefined services such as: VoIP, Content Distribution, wireless services, Internet Access, Security areas etc. This type of addressing may create dependencies on IP addresses that can make renumbering harder if the nodes or interfaces supporting those services on the network are sparse within the topology.
The above draft has a lot more on designing addressing plans including some detailed case studies in an appendix.
Bill Herrin posted the following explanation of aggregation on the NANOG mailing list.
If I remember right, this came from a discussion on the ARIN PPML list. I don't clearly remember the discussion, so my apologies in advance if I get some of it wrong.
During discussion and analysis, allocation of addresses by POP was found to be incompatible with a couple goals deemed more important. The general consensus was that you should establish areas consisting of multiple POPs and aggregate by area instead. However, ARIN is not in the business of recommending routing best practices so the recommendation was narrowed to just "don't aggregate by POP" meaning "don't fine-tune your aggregation all the way down to the POP level; stop somewhere above it."
The first problem with aggregating by POP was customer mobility. When a customer moves from the suburbs 10 miles east of the city to the suburbs 10 miles west of the city, he should keep his addresses and that shouldn't cause you any hardship. If your aggregation area includes the city and its suburbs, that's no problem for you. If you've aggregated by the CO where their DSL connects to, then over time as customers migrate around the city you'll end up with an awful mess.
Another problem was that there is a temptation to implement security features at the aggregation points. For example, you might put your egress source address filtering and spoofing protection there to catch anything where you couldn't for whatever reason filter directly on the customer link. If the aggregation point covers a large area, there are a very small number of customer movements for which that will be a problem. If its a single POP you end up screwing a lot of customers. A canonical example of this sort of error is Verizon Business's Ashburn data center. Customers of the data center can't leave the building and keep their IP addresses. Verizon Business (UUnet/MCI) can't attach them anywhere else on the network.
Another problem was DOS attacks. If someone DOSes a network that has moved since its inception then it'll first take out the specific route, then take out the covering route. If you've aggregated by area, there's a good chance the covering route is far enough upstream to handle the DOS. If the covering route is an alternate POP, it probably doesn't so you've allowed the attacker to crush two POPs with one DOS.
Another problem was sparse allocation. Sparse allocation for IPv6 addresses is strongly recommended. If allocated address blocks are not adjacent to each other then when a customer says, "I need more addresses," there is a strong probability that you can grant the request by simply changing the prefix length. This keeps your routing table small and tidy. You get lots and lots of IPv6 addresses, so if you only break them up into a dozen pools you still have plenty with which to do sparse allocation. If you break them up into pools for each of the hundreds or even thousands of POPs that you have and/or create two levels of aggregation (first by POP, second by area), you won't have enough to do effective sparse allocation.
There were other points disfavoring aggregation by POP but I can't remember what they were. I think there was also an assumption that traditional dynamic-IP customers would not each get a static block of IPv6 addresses. If that fails to hold true then it changes the character of the aggregation problem.
Ryan Harden posted this to the NANOG mailing list:
Our numbering plan is this:
- Autoconfigured hosts possible? /64
- Autoconfigured hosts not-possible, we control both sides? /126
- Autoconfigured hosts not-possible, we DON'T control both sides? /64
- Loopback? /128
Within our /48 we've carved it into (4) /50s.
- First, Infrastructure. This makes ACLs cake.
- Within this /50 are smaller allocations for /126s and /128s and /64s.
- Second, User Subnets (16k /64s available)
- All non-infrastructure subnets are assigned from this pool.
- Third, Reserved.
- Fourth, Reserved.
We believe this plan gives us the most flexibility in the future. We made these choices based upon what works the best for us and our tools and not to conserve addresses. Using a single /64 ACL to permit/deny traffic to all ptp at the border was extremely attractive, etc.
Matthew Petach said on the NANOG list:
As I mentioned in my lightning talk at the last NANOG, we reserved a /64 for each PtP link, but configured it as the first /126 out of the /64. That gives us the most flexibility for expanding to the full /64 later if necessary, but prevents us from being victim of the classic v6 neighbor discovery attack that you're prone to if you configure the entire /64 on the link. All someone out on the 'net needs to do is scan up through your address space on the link as quickly as possible, sending single packets at all the non-existent addresses on the link, and watch as your router CPU starts to churn keeping track of all the neighbor discovery messages, state table updates, and incomplete age-outs. With the link configured as a /126, there's a very small limit to the number of neighbor discovery messages, and the amount of state table that needs to be maintained and updated for each PtP link.
It seemed like a reasonable approach for us--but there's more than one way to skin this particular cat.