[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: I-D Action:draft-ietf-v6ops-ipv6-cpe-router-03.txt

>           (b)  In a dual-stack environment with IPCP and IPV6CP running
>                over one PPP logical channel, the NCPs MUST be treated
>                as independent of each other and start and terminate
>                independently.


This by the way is a nice one...I agree with the proposed text, but operationally we encountered some nice race conditions on this which you might want to catch in later revisions.

(For reference we run a modest install base of about 300k DSL lines, mostly on PPPoA.)

Ocassionally it happens that the IPCP gets interrupted because of packetloss causing echos to get dropped or it fails to establish NCP at all (lack of avaiable addresses in the free pool for instance).

Now on normal occasions from a customer perspective this means the line is broken and what usually happens (all major vendors) is that the CPE automatically tries to reopen either by starting IPCP again or even dropping the PPP completely and start from scratch.

While doing tests with IPv6 in a dual-stack environment we found a race condition in which the IPCP fails and the IPv6CP didn't, given IPv6 is not well spread these days it caused the customers to phone support because it felt "their line was broken" as they coudn't reach about 99.99% of the internet. Technically this behavior is correct in respect to the draft, IPCP and IPv6CP is completely seperated it out. However what we observed is that some vendors (including major ones like Cisco) fail to re-establish a working IPv4 connection.

Now there are always multiple sides to every story because we did find a situation where the CPE tried to reopen IPCP but our access box (JUNOSe) kept rejecting this until the whole PPP session was killed and rebuild., I also seen situations where the CPE was perfectly happy with the IPv6CP only connection and didn't even tried to re-establish an IPv4 connection.

I'm not sure how to handle this, the text in the draft is correct in the sense that you want it to be seperate, but in real life deployments, especially during the transitioning phase you, really want IPv4 to work, even it would mean to break down a working IPv6 link.

MarcoH