Kitz Forum
Computers & Hardware => Networking => Topic started by: Weaver on January 23, 2020, 02:33:57 PM
-
I wonder how completely impossible it is (or not) to get ipv6 PI space? I’m just imagining it’s a nightmare and that they will sneer at you unless you’re the next VastCo Ltd.
I read that you have to submit some documentation to RIPE. If all you want is a /48, no more, I think it ought to be made into a no-hassle thing. What do you think?
I don’t see why you should be required to be a huge organisation. A small outfit might want dual ISPs.
-
It's relatively easy to get IPv6 PI space.
You have an AS number and kit running BGP, right?
You need these to have PI space. You need to advertise it to upstream providers - who you'll be paying for IP transit at market rates.
https://www.ripe.net/manage-ips-and-asns/as-numbers/request-an-as-number
https://www.ripe.net/manage-ips-and-asns/ipv6
I can't say I rate the chances of a home user being able to justify a /48, though.
3.5. Conservation
Although IPv6 provides an extremely large pool of address space, address policies should avoid unnecessarily wasteful practices. Requests for address space should be supported by appropriate documentation and stockpiling of unused addresses should be avoided.
This information any use? https://www.ripe.net/manage-ips-and-asns/resource-management/number-resources/independent-resources
-
@CarlT I already have a /48, from AA. That’s their default, which seems perhaps a bit extravagant to me. I’m both a home user and a very small business user; my wife has her own B&B business for which I provide support.
Perhaps AA wants to avoid the nightmare of renumbering if a customer is a business it starts one and then they take off like a rocket becoming the next Facebook. They’re perhaps more concerned about the hassle than conservation.
I would have thought that a lot of AA customers would be fine with a /60 or /56. If you gave out /64s to all home users then it would not be the end of the world, seeing as if you realised later that you needed more then there is an AA button that you can hit iirc that just gives you an additional /64 with no hassle. Perhaps /60s for home users and very small businesses who ask for something more, and /56s for those where there is any chance of future expansion, but basically give the users what they want but then yet more still, over the top, just in case.
The reason I would want a block larger than a /64 is for ACLs, would want adjacent address spaces of subnets to have a contiguous range with no holes, so being able to aggregate sub ranges into one. It just keeps the ACLs smaller. But some home users even with multiple /64s that are non-contiguous could just survive like that.
-
Everything I've read suggests that issues a /48 per customer is "best practice". Any ISP issuing less is doing it wrong.
However, an ISP giving out /48 per customer is a very different scenario to requesting it yourself. The ISP deals with setting up the routing and transit, as that's fundamentally what an ISP exists for.
-
Residential customers /64 is best practice, /48 for businesses. Both supply an absurd amount of addresses well in excess of anything a customer might require.
The big thing with PI is a simple one - every PI range increases the size of the global routing table. PA space does not as it falls within a larger operator's supernet.
Going back to IPv4, Weaver, you have a /26 from A&A. Here are all the prefixes they advertise - https://bgp.he.net/AS20712#_prefixes
Can you imagine how much larger that would be if they advertised each and every /29, /28, etc, etc, they provide to customers separately?
That's why PA space is so much easier to get your hands on than PI. Your /26 of PA space was easy enough, you couldn't even get a /26 of PI space, too small, /24 minimum.
-
Everything I've read suggests that issues a /48 per customer is "best practice". Any ISP issuing less is doing it wrong.
However, an ISP giving out /48 per customer is a very different scenario to requesting it yourself. The ISP deals with setting up the routing and transit, as that's fundamentally what an ISP exists for.
You may wish to discuss this with the country's largest ISP - ie BT, who use /56 prefixes.
:)
-
You sound surprised :D
-
Residential customers /64 is best practice, /48 for businesses. Both supply an absurd amount of addresses well in excess of anything a customer might require.
Its my understand that giving out only a /64 is NOT recommended, as you aren't supposed to subnet any smaller than that, which leaves residential customers with only one subnet when its advisable to have IOT on its own subnet.
I agree that giving out a /48 is complete overkill for residential though.
You may wish to discuss this with the country's largest ISP - ie BT, who use /56 prefixes.
:)
That actually seems to be recommended size for residential by RIPE. https://www.ripe.net/publications/ipv6-info-centre/deployment-planning/create-an-addressing-plan
-
@CarlT I understand the implications for the global routing table. RAM is incredibly cheap though and keeps getting more absurdly huge. But I definitely think that more research is needed.
The start of a very half-baked idea: Couldn’t RIPE provide an ‘aggregation object’ that your PI space goes into, then that object alone is visible to the global routing table, and then within the aggregation object we need an entry that says "see AS20712/xxx” ?
Clearly though you wouldn’t just list everyone as PI. Your user would have to demonstrate a need, by either a realistic need to be insulated from the effects of a change in ISP, or else have to actually have multiple ISPs. That would rule out most domestic customers.
Off topic: I see that in that list of prefixes there are some entries that could be taken as subsumed by encompassing shorter prefixes. Is that inefficiency ? or is the longer prefix an incompatible target, an exception, so it has to be listed individually and will be recognised and will be a hit/match by longest-prefix-wins ?
-
@CarlT I understand the implications for the global routing table. RAM is incredibly cheap though and keeps getting more absurdly huge. But I definitely think that more research is needed.
RAM isn't the problem. Parsing through hundreds of thousands of entries to decide where to route every packet is.
The start of a very half-baked idea: Couldn’t RIPE provide an ‘aggregation object’ that your PI space goes into, then that object alone is visible to the global routing table, and then within the aggregation object we need an entry that says "see AS20712/xxx” ?
No. Your space is now Provider Aggregatable not PI. BGP functions on AS Paths.
Clearly though you wouldn’t just list everyone as PI. Your user would have to demonstrate a need, by either a realistic need to be insulated from the effects of a change in ISP, or else have to actually have multiple ISPs. That would rule out most domestic customers.
You do that by getting your own PI space and advertising it to multiple upstreams via BGP.
Off topic: I see that in that list of prefixes there are some entries that could be taken as subsumed by encompassing shorter prefixes. Is that inefficiency ? or is the longer prefix an incompatible target, an exception, so it has to be listed individually and will be recognised and will be a hit/match by longest-prefix-wins ?
There are, and have to be, multiple entries in the table for the same subnet with a preferred path. It's also possible to use Equal-Cost MultiPath where every other criteria matches - I was using this just this week to spread load within a massive network.
BGP path selection - https://www.cisco.com/c/en/us/support/docs/ip/border-gateway-protocol-bgp/13753-25.html
BGP Multipath - the stuff I was doing bright and early yesterday morning - https://www.cisco.com/c/en/us/support/docs/ip/border-gateway-protocol-bgp/13753-25.html#anc5
-
That actually seems to be recommended size for residential by RIPE. https://www.ripe.net/publications/ipv6-info-centre/deployment-planning/create-an-addressing-plan
I stand corrected - thank you.
-
I forgot about equal cost multipath - which is daft in my case.
-
I think AAISP will help you get some IPv6 PI addresses if you ask them (IIRC, they will charge about £100 admin fee), then you'll need to pay the yearly RIPE charge (about 50 GBP/EUR ?).
They will then announce them for you and route them to your home broadband connection.
But I'm not sure why you would want to.
-
Hi
Just to add Cerberus gave me a /56 which from what I have read seems to be considered best practice now.
So that is 18,446,744,073,709,551,616 unique IP addresses * 256 subnets, certainly a few more than my 1 IPv4 address ;D
Regards
Phil