Re: [quicwg/base-drafts] Connection migration must be negotiated (#1271)

[Mike Bishop <[e-mail]>]

@tfpauly also hit on my biggest concern here:  What exactly gets turned off?  The point of the connection ID is the ability to survive NAT rebinds; ideally, the load balancer will be looking at the CID and not the 4-tuple at all.  After all, NAT rebinds can cross IP addresses (particularly in the Enterprise or mobile space), so 2-tuple binding isn't going to be sufficient.  But perhaps it's an acceptable Band-Aid for initial deployments.

So what are the protocol mechanics that get turned off?  The server is still going to need the ability to do path validation during the handshake (three-packet limit?), so that machinery still needs to be there.  If you're saying that you'll disable support for NAT rebinding, you're potentially going to be in a painful situation.

Regardless, the text added in #1254 states the principle that clients are the controlling party for all migrations; this remains true even if we merge #1251, because the server requests the client to perform a migration.  I think the most this restriction could get us is that clients should never intentionally change their local IP address -- though I suspect many clients will just be connecting without specifying a local IP address, so even that might have some interesting quirks.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1271#issuecomment-379362183