Re: [tcpm] WGLC comments for draft-ietf-tcpm-dctcp-04

[Praveen Balasubramanian <[e-mail]>]

Changed to : " If SYN , SYN-ACK and RST packets for DCTCP connections have ECT set in the IP header, they will receive the same treatment as other DCTCP packets when forwarded by a switching fabric under load. Lack of ECT in these packets may result in a higher drop rate depending on the switching fabric configuration. Hence for DCTCP connections, the sender SHOULD set ECT for SYN, SYN-ACK and RST packets."

-----Original Message-----
From: Michael Welzl [mailto:[e-mail]] 
Sent: Thursday, March 16, 2017 1:58 AM
To: <[e-mail]> Fairhurst <[e-mail]>
Cc: Praveen Balasubramanian <[e-mail]>; [e-mail]; [e-mail]
Subject: Re: [tcpm] WGLC comments for draft-ietf-tcpm-dctcp-04


> On 16 Mar 2017, at 09:52, Gorry Fairhurst <[e-mail]> wrote:
> 
> On 16/03/2017, 08:26, Michael Welzl wrote:
>> Hi,
>> 
>> I haven't read the draft since long ago (one of the first versions), but I just read this dialogue and I have only one comment, so I'm removing everything else:
>> 
>> 
>>> ---
>>> 
>>> 3.4.  Handling of SYN, SYN-ACK, RST Packets
>>> 
>>>   " The switching fabric can drop TCP packets that do not have the ECT
>>>    set in the IP header.  If SYN and SYN-ACK packets for DCTCP
>>>    connections do not have ECT set, they will be dropped with high
>>>    probability.  For DCTCP connections, the sender SHOULD set ECT for
>>>    SYN, SYN-ACK and RST packets."
>>> 
>>> - I'd take the position that the fabric can and will drop any packets under overload, as per RFC7567. I'd prefer to explicitly state that to avoid a misconception that ECT eliminates all drop (rather than nearly all drops).
>>> 
>>> 
>>>>> Change this to : " If SYN and SYN-ACK packets for DCTCP connections do not have ECT set in the IP header, they will likely be dropped by the switching fabric under load. For DCTCP connections, the sender SHOULD set ECT for SYN, SYN-ACK and RST packets."
>> To me, "will likely be dropped by the switching fabric under load" doesn't sound quite right - as if they would most probably be dropped whenever there is any form of load (no matter how much). I would suggest to change this to "...they are more likely to be dropped by the switching fabric under load."
>> 
>> Cheers,
>> Michael
> I agree with Michael, but also I don't see it is correct for transport protocolsto need a very specific network behaviour. The choice of which packets are dropped is a design choice of the AQM: If we were to see FQ-Codel or FQ-PIE (for exmaple) deployed, then a common behaviour would be to separately queue new flows (even if non-ECT) - Switch design can change - I therefore think we should not be assuming a specific switch behaviour, but also I do think the ID does not need to go there ....  because I suspect the clause could perhaps be written differently...
> 
> - I'd suggest to change to:
> "SYN and SYN-ACK packets for DCTCP connections set the ECT(0) codepoint in the IP header, this ensures they receive the same treatment as other DCTP packets when forwarded by a switching fabric under load. For DCTCP connections, the sender SHOULD set ECT for SYN, SYN-ACK and RST packets."

This is just an ACK to say that I agree.

Cheers,
Michael