Re: [v6ops] draft-xie-v6ops-network-happyeyeballs

[Mikael Abrahamsson <[e-mail]>]

On Mon, 26 Nov 2018, Ron Bonica wrote:

> This week, please review and comment on 
> draft-xie-v6ops-network-happyeyeballs.

I think it's a great idea to tell the operator about problems, that kind 
of telemetry can really help in operating the network.

However, the proposed method of doing it (syslog) and the information it 
reports (IP addresses and domains the user is communicating with) has huge 
privacy implications. Also you have no idea who might be grabbing this 
information before it even makes it to the ISP).

I think at a minimum there needs to be an encrypted and verified mechanism 
(https perhaps?) for doing the reporting, so you know where the report 
goes.

I see three entities that have interest in this information, it's the ISP, 
the operating system manufacturer, and the web site which is having 
problems. If HE fails, I see very little harm (privacy) in telling the 
website about it using the address family that actually did work. So that 
could be included. For the device manufacturer, they have "send 
diagnostics information to vendor", so here customers can opt in. For the 
ISP, I have no idea. Here it might be better to define an API between the 
device vendor and the ISP, so that the device vendor can report problems 
seen, at an aggregate? I have personally been involved in getting 
operational people from my employer talking to device vendors to figure 
out what might be wrong, and automating this would be great.

Also, the DNSSEC implications are glossed over grossly. We should take for 
granted that end devices are running validating resolvers and handle that 
use-case.

So I'm supportive of the problem statement and it would help ISPs greatly 
if they could get telemetry from end devices, I just have great concerns 
over the method proposed in this draft.

-- 
Mikael Abrahamsson    email: [e-mail]