On Mon, Mar 25, 2013 at 8:09 AM, Girish Venkatachalam
<[email protected]> wrote:
>
> But the best way seems to be aggregating multiple VPN endpoints like a
>  German company pioneered many years ago(10 years).

VPN is a very specific scenario. Cannot be used to access servers that
belong to others not giving VPN.

> TCP is a higher layer protocol. This project may have applications but
> TCP is already
>  a very complicated higher level layer. Adding this on top does not
> make much sense
>  according to me.

SCTP is an alternative transport protocol and has a lot of advantages
over TCP. Suffers from lack of support from boxes in the middle. TCP
though a bit flawed has a very wide adoption.

> Instead the IP and data link layers are best to do this.
Possibly and also agree to an extent. Adoption.....

> In the UNIX world, you can achieve this with BGP AS policy. You can do
> this with
>  ML PPP.
>
> You can do this with ECMP which is what I normally use.ECMP is equal
> cost multipath routing.
>
> Then for simple link priority and failover I can use the trunk(4) idea
> in OpenBSD(not VLAN trunk).
>
> I feel that VPN approach is best.
>
The key is the application is impervious  of TCP or MPTCP. It thinks
it is calling TCP as the call interface is the same. I wonder if the
TCP subflows are just like TCP such that session information will get
created at the CPE that is allowing the outgoing connection to allow
reply data through. This was my doubt.

-- Mohan Sundaram
_______________________________________________
ILUGC Mailing List:
http://www.ae.iitm.ac.in/mailman/listinfo/ilugc

Reply via email to