[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Middleboxes [Was: Flow label versus Extension header - protocol itself]



marcelo bagnulo braun wrote:

But you need to given all the addresses to the host so that there are redundant addresses available for it during initial communication, don't you?
>
yes, unless you are sure that you are communicating with a shim enabled peer (the peer or a middlebox in front of it) in which case you could use shim capabilities even for dealing with outages in the locator used for initial contact

How would the DHCP server, when handing out address, know whether or not the local host will communicate with a shim capable peer?
The local host might communicate both with shim incapable peers and with shim capable peers. In the first case it makes sense for the host to know all its addresses with the hope that default address selection and application retries finds a working address pair. In the second case the middlebox can help so it makes more sense (does it?) to only give the host one address, which is part of a HBA/CGA set maintained by the middlebox.


If the DHCP server only hands back one address to the host, what would happen when a shim6 capable host is used? It would want all the addresses/prefixes so it could do shim6 on its own.


i guess that the DHCP server should know if the requesting host is shim capable or not

This is at least solvable, with a new DHCPv6 option saying "I'm shim6 capable - disable any shim6 middlebox stuff".


imho middleboxes don't really fit nicely in the shim arch. I mean shim arch is e2e by design, not suited for middle boxes i guess.

Well, TCP is e2e yet there are plenty of transparent TCP proxies which work most of the time. (They are a single point of failure for a connection etc, but they don't seem to run into the same type of issues we see for a shim6 middlebox.) Perhaps the added complexity of both overloading the IP addresses to function as ULIDs and locators, combined with handling failures during initial communication in the application/ULP, and failure of existing communication in the shim, is what makes this so hard.


any middlebox support will be some form of hack imho

Agreed. Question is it would be a useful hack, or just asking for trouble.

   Erik