Hello Antonio, and thanks both for your explanation as well as your question 
that made me understand things better.

8 feb 2012 kl. 16:13 skrev Antonio Quartulli:

> Hello Christian and thank you for your explanation,
> 
> On Wed, Feb 08, 2012 at 03:53:22PM +0100, Christian Huldt wrote:
>> 
>> 8 feb 2012 kl. 14:26 skrev Antonio Quartulli:
>> 
>>> Hello Christian,
>>> 
>>> On Wed, Feb 08, 2012 at 12:29:52PM +0100, Christian Huldt wrote:
>>>> As a complete newbie to batman, please excuse if this question is 
>>>> unusually stupid, but I just got home after getting update on the mptcp 
>>>> kernel module http://mptcp.info.ucl.ac.be/ and it seems to me that running 
>>>> mptcp over a batman mesh could potentially be quite nice. It that correct 
>>>> or am I just being ignorant?
>>>> 
>>> 
>>> Before answering your question, I would like to read and understand a
>>> bit more about mptcp, but actually I couldn't find any resource explaining 
>>> the
>>> concept behind it an how it is supposed to work. The only document I've been
>>> able to find is Sébastien's Thesis...but I would rather prefer a lighter 
>>> reading
>>> first :-)
>>> 
>>> 
>>> Anyway, is it somehow similar to ip multihoming with load balancing?
>> 
>> I am unfortunately not able to say whether that is the case (yet…), I saw a 
>> demo and (think I) understood at least parts of the talk.
>> This is my limited understanding:
>> This implementation of mptcp works by creating a virtual interface bound to 
>> two (or more?) interfaces and traffic over that virtual interface is by the 
>> mptcp layer sent packet by packet to one of the bound interfaces.  - I am 
>> assuming that it is using the tcp acks to optimize which interface gets more 
>> packets - 
>> 
> 
> It sounds like "bonding at the TCP layer" (let me pass the term).
> 
> 
>> The receiving host has to be mptcp capable as well in order to bring the two 
>> (or more?) connections together to one.
>> 
>> One of the benefits is that except for the two hosts talking, no one needs 
>> to be aware of the "multi path nature" as it is just two normal tcp 
>> connections.
>> 
>> I think that it could be useful (or mayhem) in a mesh network though I would 
>> believe that where will be complications.
> 
> Well, assuming you know how batman handles clients, the only way to exploit 
> such
> "multi-path TCP" is to connect the two (or more) involved interfaces of the 
> client to two
> (or more) different nodes. If not, all the packets will follow the same path 
> in the mesh
> network anyway.

Maybe one could do something similar with routes rather than interfaces? That 
is, by sending packets to different other nodes?
But I think that could complicate things and I should probably dig deeper into 
batman-adv before coming up with more ideas...

> By the way, I'd say that from the batman-adv point of view using mptcp is the
> same as ethernet bonding. (Please correct me if I made any mistakes)

I think that is correct.

Reply via email to