It's a small step for TLAs but has huge impact on Stub-ISPs
Hello, i am concerned about the current draft for IPv6 adress assignment. Specificaly I am worried about the consequences of the aggregateble global unicast address scheme for smaller ISPs. I see the following problems for non-TLA "Users" which run a Multihomed environment: (Please correct me or point me to the corresponding documents) 1) Multihoming does not reward you with more stability anymore Since each Interface gets two or more IPv6 addresses, a existing TCP connection breaks if the Line on which the packets come in goes down. Of course this could be solved with a new version of TCP, but if you look at the time it took to develop IPv6 this is in the very distant future (not including the time it takes for deployment) (I guess it's the same with UDP - once a programm has choosen a destination IP address, it uses it for the whole transaction - if the link fails, the transaction stops ) AND: If the DNS still announces the TLA-1::SLA: Addresses when the Line to TLA-1 is down, then some connections fail even if the Path via TLA-2 is still up and running. -> so: there must be a probing of the connection !within! the application I fear that the loss of topology information with the TLA scheme will introduce more problems than those which are obvious at first sight. 2) Load balancing gets harder If a station outside your domain opens a connection to the address TLA-1::Interface you cant decide which line the incoming traffic should go. It should be possible to do a weighted DNS round robin to decide which prefix is prepended to the SLA:IntID. But with DNS caching this is only a very unprecise tool. I did not find any documents addressing these problems. Are there any ? Or did only those ISPs that do qualify for TLA assignments write those drafts ;-) ? with kind regards Hannes -- -------------------------------------------------------------------- | Hannes R. Boehm debis Systemhaus EDVg | | Network Engineer Hofmuehlgasse 3-5 | | hannes.boehm@debis.at 1060 Wien | | AUSTRIA | --------------------------------------------------------------------
In your previous mail you wrote: 1) Multihoming does not reward you with more stability anymore => I share your concern! 2) Load balancing gets harder => I am afraid that it is already the case for IPv4 but it is point where something more clever could/should be done... I wrote an I-D draft-ietf-ngtrans-6bone-multi-00.txt about multi-homed routing done but it is a bit old and I should complete it (some details from implementation experience, a lot of new things about address selection). Regards Francis.Dupont@inria.fr
participants (2)
-
Francis Dupont -
Hannes R. Boehm