Re: [multipathtcp] New Version of MPTCP Architecture Draft

Hi Yoshifumi,

Let me try to answer those inline.

On 5 Feb 2010, at 12:03, Yoshifumi Nishida wrote:

> 
> Hello Alan, 
> 
> I have several questions on the reliability section.
> Could you elaborate a bit?
> 
> In section 4.2, you mentioned that there are currently no connection 
> levels acks. However, after a few lines later, you also mentioned connection
> level MPTCP ACKs are not cumulative. Isn't it contradictionary?
> 
> If connection level MPTCP ACK exists, what does it look like? Why it cannot
> be cumulative? 
This is a misleading phrase. It tries to say that currently, the acks at the connection level are inferred from subflow acks but do not have the same cumulative nature as tcp cumulative acks.

> 
> ".. The receiver can simply drop out-of-order segements if needed (for instnace,
> due to memory pressure)."  Could you elaborate how this situation happens?
Well, in TCP any out of order data can be dropped, even if these were SACKed. This is called SACK reneging, and it will not happen frequently, but it is possible. You can't do this with MPTCP; if something was acked
at subflow level, it cannot be dropped in the rerder buffer at the connection level.

> 
> ".. if the proxy acks a segment and then crashes, the sender will not retransmit
> the lost segment on another subflow." To deal with this case, I think MPTCP 
> needs to keep data for a while even if it is acked by subflow. But, it looks 
> very expensive. Do we need to support this?

You are right. Dealing with this case means that we need to introduce an explicit DATA ACK and that the sender must keep the data until it's acked at connection level. It's unclear what the best solution is.

Cheers
Costin
> 
> Thanks,
> --
> Yoshifumi Nishida
> nishida@sfc.wide.ad.jp
> 
> From: "Ford, Alan" <alan.ford@roke.co.uk>
> Subject: [multipathtcp] New Version of MPTCP Architecture Draft
> Date: Wed, 3 Feb 2010 11:03:24 -0000
> Message-ID: <2181C5F19DD0254692452BFF3EAF1D68095EDC61@rsys005a.comm.ad.roke.co.uk>
> 
>> Hi all,
>> 
>> We have submitted an updated version of the MPTCP architecture draft,
>> which can be found at:
>> http://www.ietf.org/id/draft-ford-mptcp-architecture-01.txt
>> 
>> This is a significant change from the previous version, concentrating
>> now on how core architectural principles will lead to high-level design
>> decisions for a MPTCP protocol.
>> 
>> It is intended to present and discuss this draft at the interim audio
>> conference next week. In the meantime, we'd be very interested in
>> feedback and discussion, so please do send comments to the list.
>> 
>> Cheers,
>> Alan
>> 
>> _______________________________________________
>> multipathtcp mailing list
>> multipathtcp@ietf.org
>> https://www.ietf.org/mailman/listinfo/multipathtcp
> _______________________________________________
> multipathtcp mailing list
> multipathtcp@ietf.org
> https://www.ietf.org/mailman/listinfo/multipathtcp