IPv6 deployment for small residential providers
Just to follow up on the question I just asked following the BCOP draft on troubleshooting IPv6 for residential providers, should we write a companion document focused on small providers that explains how to deploy IPv6? I am happy to start writing, would anyone like to help? Cheers, -w
Hi William, I have a good experience and background on this subject, since I had a regional ISP in Brazil for the last 17 years before selling the company (sept/2013) and coming to Europe (july/2014) to work for a SaaS provider. I'd be mostly happy to join you on this effort since I understand that this document can be really helpful for lots of small companies that are responsible for providing internet where big companies can't reach, and mostly aren't prepared for v6 yet. Cheers and keep in touch. see you tomorrow. vdeluca On Mon, Nov 3, 2014 at 6:37 PM, William Waites <wwaites@tardis.ed.ac.uk> wrote:
Just to follow up on the question I just asked following the BCOP draft on troubleshooting IPv6 for residential providers, should we write a companion document focused on small providers that explains how to deploy IPv6?
I am happy to start writing, would anyone like to help?
Cheers, -w
I've started writing here: https://pad.okfn.org/p/bcop-small-ipv6 Today, a couple of paragraphs about the intended audience before getting into the meat of it. -w
On Sun, Nov 9, 2014 at 12:47 PM, William Waites <wwaites@tardis.ed.ac.uk> wrote:
I've started writing here:
https://pad.okfn.org/p/bcop-small-ipv6
Today, a couple of paragraphs about the intended audience before getting into the meat of it.
It's a good start, but could you rewrite the part on "Address Allocation" ".... ipv6 not so different (only forget scarcity and use /64 by default and /56 or /48 if requested" I guess the allocation should be replaced with assignment toward end-users as a starter, then the next thing is the size you mention. Giving a /64 toward end-users will break many things, it will break homenet design (IETF homenet) and not to forget it's against the original intention when we relaxed the /48-for-everyone. Probably biggest clue for this is the HD-Ratio in http://www.ripe.net/ripe/docs/ripe-589 "In IPv6, "utilisation" is only measured in terms of the bits to the left of the efficiency measurement unit (/56)." Replace it with something along this "by default give everyone a /56, on request a /48". It is really that simple. some background on the /56 size. Sometime before 2005 a discussion started if /48 for everyone was too strict, not so much about wasteful but more than anyone ever would need. After some back and forth RIPE changed it in 2005, the earliest document I found was 2005-08. I've tried to fill a /48 on just my own stuff in many ways but it's almost impossible, a /56 on the other hand is possible to fill but it's hard. I did tunnels between several machines I own/control, vpn so I could inside my own network, each service and LAN that got a /64 etc. It is documented quite a few more places than just in RIPE documents. The original intention was that we thought /56 was the right and recommended (lower cases) sizes for regular end-users. /48 was the right size for bigger end-users like enterprises. Over the years through rewrites it seems to have been relaxes so it's not that easy to see that it was ment as a recommanded (lower cases) on assigments sizes. I guess the reason for it being lower case is that none can dictate how an ISP/or anyone should do their assigment of the address space, only thing was to make an recommendation. Here is a few documents that mention the /56 http://www.ripe.net/ripe/policies/proposals/2005-08 http://datatracker.ietf.org/doc/draft-ietf-6man-why64/?include_text=1 http://www.ripe.net/ripe/policies/proposals/2005-08 http://www.ripe.net/ripe/policies/proposals/2006-02 http://www.ripe.net/ripe/docs/ripe-589 http://tools.ietf.org/html/draft-narten-iana-rir-ipv6-considerations-00 -- Roger Jorgensen | ROJO9-RIPE rogerj@gmail.com | - IPv6 is The Key! http://www.jorgensen.no | roger@jorgensen.no
On Sun, 9 Nov 2014 13:54:07 +0100, Roger Jørgensen <rogerj@gmail.com> said: > It's a good start, but could you rewrite the part on "Address > Allocation" Well, yes, that was just a placeholder sentence! But I've made the change as you asked. I'm not sure I agree though, and the reason is not to do with efficiency of address space use but operational ease of provisioning. Operationally, what does this mean? The most common case is going to be a single subnet, so how is the gateway going to know which one out of the /56 to use? Somebody has to pick a /64 to put on the inside ethernet interface. How is this done? No problem *assigning* a /56 but using it is another matter. -w
On Sun, Nov 9, 2014 at 2:11 PM, William Waites <wwaites@tardis.ed.ac.uk> wrote:
On Sun, 9 Nov 2014 13:54:07 +0100, Roger Jørgensen <rogerj@gmail.com> said:
> It's a good start, but could you rewrite the part on "Address > Allocation"
Well, yes, that was just a placeholder sentence! But I've made the change as you asked. I'm not sure I agree though, and the reason is not to do with efficiency of address space use but operational ease of provisioning.
Operationally, what does this mean? The most common case is going to be a single subnet, so how is the gateway going to know which one out of the /56 to use? Somebody has to pick a /64 to put on the inside ethernet interface. How is this done? No problem *assigning* a /56 but using it is another matter.
ah I see, bad wording from my side. Any _end-user_ should get minimum a /56 for their use, an assignment. How they use that assignment are another technical matter - that's the operational side. On the actual use of IPv6 addresses I guess https://tools.ietf.org/html/draft-carpenter-6man-why64-01 is a better source for information. It mention cases where a /64 is the best choice, and where other sizes can, and can not be used. -- Roger Jorgensen | ROJO9-RIPE rogerj@gmail.com | - IPv6 is The Key! http://www.jorgensen.no | roger@jorgensen.no
On 09/11/14 13:11, William Waites wrote:
Operationally, what does this mean? The most common case is going to be a single subnet, so how is the gateway going to know which one out of the /56 to use? Somebody has to pick a /64 to put on the inside ethernet interface. How is this done? No problem *assigning* a /56 but using it is another matter.
Hi, I've seen all sorts of tricks being made on this topic... Some of them assign /56 to residential customer and: - use wan interfaces unnunmbered (just link-local addreses and installed route on BRAS towards wan ll interface) -or- - use the first /64 for ppp/wan link and the rest for LAN -or- - assign separate /64 for wan link and separate /56 for LAN side. /56 on the LAN side gets auto-configured as on well behaved CPEs there is a script for that. Usually, one address (from first /64) is assigned to local loopback and next /64 is put on first L3 port and next one on next L3 port and so on... you can of course change the ID of where the /64 assignments starts. http://wiki.openwrt.org/doc/uci/network6#downstream.configuration.for.lan-in... (an example). Cheers, Jan
Hi William, On 09 Nov 2014, at 12:47, William Waites <wwaites@tardis.ed.ac.uk> wrote:
I've started writing here:
https://pad.okfn.org/p/bcop-small-ipv6
Today, a couple of paragraphs about the intended audience before getting into the meat of it.
-w
Thanks, a good start. Just one comment: "they also need ipv4 addresses to use as router-id (need not be global, but need to be unique).” is not true. It needs to be a 32-bit number, but it doesn’t have to be an IPv4 address. This should be made clear, as I notice in our training courses, a lot of engineers seem to think that the router ID MUST be an IPv4 address, while it normally is, it is not mandatory. 0.0.0.1 is a valid router-ID, I believe. In an IPv6-only network, for example, when you have no IPv4 addresses, you can just make something 32-bitty up and use that as the router ID. On the address assignment: What we see and hear in practice in our courses, is assign something on 4-bit boundary, big enough to cater for the next 10 years. So: a /64 only if you are absolutely sure that the customer will never come back for one more subnet (not likely). a /60 (if you are conservative) a /56 (most common for residential users) a /52 (we see this in some cases for both residential customers and business customers) a /48 (for business customers, or for residential customers if you are generous, and have a one-size-fits-all-approach) I hope this helps, Cheers, Nathalie
Hi Nathalie,
This should be made clear, as I notice in our training courses, a lot of engineers seem to think that the router ID MUST be an IPv4 address, while it normally is, it is not mandatory. 0.0.0.1 is a valid router-ID, I believe. In an IPv6-only network, for example, when you have no IPv4 addresses, you can just make something 32-bitty up and use that as the router ID.
True, but it has to be unique a least between a router and all its neighbours. So if everybody on an internet exchange starts using 0.0.0.x then there will be trouble :)
On the address assignment: What we see and hear in practice in our courses, is assign something on 4-bit boundary, big enough to cater for the next 10 years. So: a /64 only if you are absolutely sure that the customer will never come back for one more subnet (not likely). a /60 (if you are conservative)
Don't be :)
a /56 (most common for residential users) a /52 (we see this in some cases for both residential customers and business customers) a /48 (for business customers, or for residential customers if you are generous, and have a one-size-fits-all-approach)
I would phrase this as "a /48 (for business customers, and for residential customers if you are not stingy, and/or have a one-size-fits-all-approach)" But those are minor details. The most common advice is - give a /48 or a /56 to residential customers - give a /48 to business customers - in case of doubt err in the direction of /48 There are not much advantages in giving other sizes to users. Cheers! Sander
On 10/11/14 19:19, Sander Steffann wrote:
- give a /48 or a /56 to residential customers - give a /48 to business customers - in case of doubt err in the direction of /48
Ditto... ;) Cheers, Jan
participants (6)
-
Jan Zorz ＠ go6.si -
Nathalie Trenaman -
Roger Jørgensen -
Sander Steffann -
Vicente De Luca -
William Waites