Re: [tcpm] TCP Connection ID

[Alexander Azimov <[e-mail]>]

Joseph,

Then you wouldn’t be talking about TCP option anymore (options cannot be
> added en-route). You also wouldn’t be talking about MPTCP either - in
> general.
>
I believe that regular TCP options (not experimental) in general have
higher chances to be supported globally.

The server TCP sequence number and timestamps are the only things that most
> TCP clients already keep track of that the server has control ove).
>
> Both of those have other meanings and change during a connection, which
> means that the middlebox would have to keep state to track them. But you
> also said the middlebox doesn’t keep state.
>
That's the purpose of my proposal. To have a reverse function for
connection id that returns the real IP, so the load balancer doesn't need
to perform the connection tracking. Though you need to know the secret to
restore value(s), so it will not have negative security effects.


> I see more clearly what you want, but no way to get there if the client
> browser/OS is unmodified.
>
I've meant that OS/browser is out of administrative control. So you can't
rely on a proprietary software/agent on the client-side. But the open
standard may be successful especially taking into account that main OS
vendors are content providers at the same time, thus already trying to
solve the problem of LB in different ways.

The incremental deployment is crucial, but IMO it looks doable.

-- 
Best regards,
Alexander Azimov