Re: New Version Notification for draft-bonaventure-quic-atsss-overview-00.txt

[Ted Hardie <[e-mail]>]

Just to amplify this a bit:

On Tue, Jun 2, 2020 at 11:49 PM Lars Eggert <[e-mail]> wrote:

>
> Apart from the privacy aspect though, tunneling all traffic through an
> operator proxy that would otherwise not be in the path certainly can have
> performance and management aspects. It's probably going to add delay due to
> route stretch and possibly queueing, and can limit throughput due to
> processing overheads or overload.
>
>
We've been through a lot of this before with various flavors of IP-layer
mobility, and the Home Agent approaches definitely showed all the issues
Lars cites above. There have been attempts to manage that with hierarchies,
fast hand-over rather than tunneling, and even new architectures like LISP
and RINA.   I don't think we want to re-learn all of those lessons from
scratch.  At the very least, we want to start with the understanding that
if ATSSS goes with this design that other tunneling protocols running on
top of it will be very difficult to debug and very hard to keep within
usable latencies.  Another comment mentioned VPNs, which are common and may
become more so.  If you have a user that has to deal with interaction
between the triangle routing to an operator proxy and the triangle routing
to a VPN end point, it will not be quick.

Traffic steering without full knowledge can manage some local
optimization,  and I don't want to denigrate that.  But those local
optimums do not necessarily result in better performance even for single
flows when the endpoints are not clear to steering mechanism.  If that is
priority, sticking with endpoint-based approaches rather than network
steering may be both cheaper and better.

regards,

Ted Hardie