On Mon, Aug 11, 2003 at 02:05:59PM +0200, Gert Doering wrote:
That's what I thought people would do. But they don't. They want to be "independent", and so they go for PI, instead for a suballocation from one of their upstreams.
(And they don't renumber, of course, but just keep the PI)
Correct.
Marcus Ruchti says that the majority of the ISPs will not announce PA space from another AS, does this include more specific announcements?
I just disagree with that statement. PA-based multihoming is a valid approach to solve certain classes of problems. If some ISPs refuse to support this (because it creates more work for their NOC or whatever reason...), the market place has alternatives...
It's the end customer who refuses to use PA address space of one of his upstreams as when he terminates contract with him, he'll need to renumber. This is a commercial factor. Renumbering costs customers real money. As such, PA space ties the end customer somewhat into the PA-providing uplink. The second reason why I find this approach not too viable is that it would create a lot of inconsistant origin BGP announcements. I do understand that this is needed for anycast applications, but this should be limited to a number of "well-known" anycast services. On the other hand, this (and ONLY THIS) point is moot as soon as the end customer gets an ASN of his own and announces the PA more-specific himself. But this leaves still the commercial consideration discussed above. According to my experience, end customers multihome for two reasons: - technical resilience - commercial independence Multi-homing with PA subnets solves only the first problem space. Nowadays, customers are more and more (and more) looking to solve the second problem space as well (and for good reasons). We ISPs are here to solve customer problems - even if it means that we give up some means of customer retention, so we should aim to tackle the second problem space as well. I will respond to the four proposed policy changes in a seperate mail. Best regards, Daniel