Re: Interim Policy proposal for IPv6 Address Assignment Policy for Internet Exchange Points
One of the points that has been raised a couple of times is... There is a lot of IPv6 address space - why not just give IXP's a sub TLA. The argument against this seems to be - if IXP's are treated as a special case then there is a precedent for every other 'special case' to justify their case. ... but there is still *lots* [1] of address space so why not give everyone who wants one a sub TLA? AIUI the reason for restricting who gets sub TLAs is... In the current IPv4 world (multiprovider) multihomed customers have their own address space and AS number this fills the default-free routing table and won't scale. In IPv6, addressing is hierarchical and multihomed customers have multiple addresses, this keeps the default-free routing table small and scalable. If everyone (multihomed customers) gets a sub TLA then the default-free routing table fills up with multihomed customer sub TLAs and IPv6 routing tables are as cluttered and unscalable as IPv4. Is this the issue? If not, why not give out sub TLAs to all who want them? Chris - LINX [1] There are 2^32 (~ 4,000,000,000) /35s available... +--+-----+-----+---+-----+------+------------------+ |-3|--13-|--13-|-6-|--13-|--16--|------64 bits-----| +--+-----+-----+---+-----+------+------------------+ |FP|-TLA-|-sub-|Res|-NLA-|--SLA-|---Interface ID---| |--|-ID--|-TLA-|---|--ID-|--ID--|------------------| +--+-----+-----+---+-----+------+------------------+
On Tue, Sep 04, 2001 at 11:52:41AM +0100, Chris Fletcher wrote:
AIUI the reason for restricting who gets sub TLAs is... If everyone (multihomed customers) gets a sub TLA then the default-free routing table fills up with multihomed customer sub TLAs and IPv6 routing tables are as cluttered and unscalable as IPv4.
Bingo! Thats the real reason. The only known way of keeping routing working is keeping the tables small. There is no known technical means of doing this at the moment, so we are forced into the administrative realm. Thus you see all sorts of rules & regulations about addreses and how to get them and use them. [See ptomaine, multi6, and some of the IRTF work for more about this.] --asp@partan.com (Andrew Partan)
On Tue, 4 Sep 2001, Andrew Partan wrote:
AIUI the reason for restricting who gets sub TLAs is... If everyone (multihomed customers) gets a sub TLA then the default-free routing table fills up with multihomed customer sub TLAs and IPv6 routing tables are as cluttered and unscalable as IPv4.
Bingo! Thats the real reason. The only known way of keeping routing working is keeping the tables small. There is no known technical means of doing this at the moment, so we are forced into the administrative realm. Thus you see all sorts of rules & regulations about addreses and how to get them and use them.
This comes close to making sense - until you realise that RIPE's rules for allocating IPv6 address space are MORE stringent than those used for allocating IPv4 address space. If the rule was that the fact that RIPE had allocated IPv4 address space automatically qualified an ISP for an allocation of IPv6 addresses, then its obvious that the IPv6 routing tables would grow no faster than the IPv4 tables [1]. But in fact what we have are much more restrictive rules for allocations from a very much larger address space. This is not easily explained as the consequence of rational policies on the allocation of address space. But it is an obvious consequence of one might call the Galbavy Rule: the urge to regulate is independent of any practical need for regulation. [2] NOTE: [1] Of course RIPE assignment policies mean that most ISPs will come to have multiple IPv4 address blocks, whereas they would have only one IPv6 block, so the IPv6 routing table should grow much more slowly than the IPv4 routing table. [2] And one should not forget that once you get your IPv6 prefix, you should never have to go back to RIPE again, depriving the regulators of anything to do. -- Jim Dixon VBCnet GB Ltd http://www.vbc.net tel +44 117 929 1316 fax +44 117 927 2015
Colleagues, Why don't we give every LIR an IPv6 TLA without any further qualification? There's only 2859 LIRs in RIPE! How many in the World? Somewhere between 8000 and 16000? I've seen figures suggesting there's only 15K ISPs in the World. 16K routes is a lot less than 100K+. Of course we have to fix the multihoming problem to keep the number of routes down. But even if we only give TLAs to the few global tier 1 transit providers if we don't fix the multihoming problem we will end up with a huge number of routes. So the problem is to fix multihoming with a technical solution and not to manage the problem through regulation. Peter Willis. BTexact Technologies.
On Wed, 5 Sep 2001, Peter Willis wrote:
Colleagues,
Why don't we give every LIR an IPv6 TLA without any further qualification? There's only 2859 LIRs in RIPE! How many in the World? Somewhere between 8000 and 16000? I've seen figures suggesting there's only 15K ISPs in the World. 16K routes is a lot less than 100K+.
Indeed. In fact, if the policy was simply to give an IPv6 TLA to any LIR who asked for one, it's likely that perhaps 10% would apply. So a negligible 300 or so additional IPv6 routes and a negligible amount of paperwork -- in fact this could just be handled by a robot hanging off www.ripe.net. The only work involved is writing the script.
Of course we have to fix the multihoming problem to keep the number of routes down. But even if we only give TLAs to the few global tier 1 transit providers if we don't fix the multihoming problem we will end up with a huge number of routes. So the problem is to fix multihoming with a technical solution and not to manage the problem through regulation.
Precisely.
Peter Willis. BTexact Technologies.
-- Jim Dixon VBCnet GB Ltd http://www.vbc.net tel +44 117 929 1316 fax +44 117 927 2015
Is this the issue? If not, why not give out sub TLAs to all who want them?
Becuase we don't want a PTOMAINng? - kurtis -
participants (5)
-
Andrew Partan
-
Chris Fletcher
-
Jim Dixon
-
Kurt Erik Lindqvist KPNQwest Sweden
-
Peter Willis