[AusNOG] Prediction: Google et. al. may artificially penalise IPv4 clients

[Paul Brooks]

On 2/05/2017 9:37 AM, Geoff Huston wrote:
>
> You asked.
>
> IPv6 Packet Fragmentation handling is broken. 
>
> Equally, IPv6 Extension Header handling is a mess. 
>
> SLAAC and RDNSS and DHCP6 is a chaotic mess.
>
> Routing coverage is erratic. 
>
> As long as you treat IPv6 kindly, use very small TCP MSS values, avoid UDP, pay care in delivering all ICMPv6 messages, run host MSS caches af necessary, and be ultra careful about routing both in terms of the completeness of the routes you see and the extent to which your routes get propagated then IPv6 will likely work just fine once you’ve got past the auto-config stage. But push it harder in any of these areas and you'll regret it.
These sound like corner-cases that are unlikely to be encountered much by a small- to
mid-sized ISP during internal toe-in-the-water enablement or testing, but might come
to light during investigation of customer complaints after IPv6 has gone into customer
trials - if these were the main issues, I'd expect to see much more trialing and
testing and almost-production implementations around within service providers.
I suspect most of the industry hasn't made it anywhere close to the "push it harder"
stage where these issues might be experienced and reported by end-users in practice.

Paul.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: paul_brooks.vcf
Type: text/x-vcard
Size: 303 bytes
Desc: not available
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20170502/d61c5208/attachment.vcf>