Re: [tcpm] Possible error in accurate-ecn

[Bob Briscoe <[e-mail]>]

Yoshi,

On 11/11/2020 09:37, Yoshifumi Nishida wrote:
>
>
> On Tue, Nov 10, 2020 at 2:08 AM Mirja Kuehlewind <[e-mail] 
> <mailto:[e-mail]>> wrote:
>
>     Hi Bob,
>
>     Please see below.
>
>     > On 10. Nov 2020, at 01:31, Bob Briscoe <[e-mail]
>     <mailto:[e-mail]>> wrote:
>     >
>     > Mirja, Richard, Ilpo, tcpm list,
>     >
>     > I've just been reading through the accurate-ecn draft to
>     double-check. I think there's a problem with the following text...
>     >
>     > 3.2.2.5.1.  Data Receiver Safety Procedures
>     >
>     >    An AccECN Data Receiver:
>     >
>     >    o  SHOULD immediately send an ACK whenever a data packet
>     marked CE
>     >       arrives after the previous [data] packet was not CE.
>     >
>     >    o  MUST immediately send an ACK once 'n' CE marks have
>     arrived since
>     >       the previous ACK, where 'n' SHOULD be 2 and MUST be no greater
>     >       than 6.
>     >
>     > ...
>     >    For the avoidance of doubt, the change-triggered ACK mechanism is
>     >    deliberately worded to solely apply to data packets, and to
>     ignore
>     >    the arrival of a control packet with no payload, because it is
>     >    important that TCP does not acknowledge pure ACKs.
>     >
>     >
>     > In the first bullet, I think it doesn't matter whether the
>     previous packet marked CE was a data packet or a pure ACK (i.e we
>     should remove the second occurrence of 'data' that I have put in
>     [square brackets].
>
>     I believe it does matter. If the previous packet was a pure ACK
>     and was CE marked, you didn’t send an ACK and so you should
>     immediately send one now. If the previous packet was a data packet
>     and CE marked, you don’t need to send an immediate ACK because you
>     already did this with the previous packet. However, it text might
>     be a but ambiguous because what is meant is “if the previous
>     packet was a data packet and CE marked”. So this does not apply if
>     we e.g. have a CE-marked data packets, then a pure ACK that is not
>     CE marked, and then another CE-marked data packet.
>
>     >
>     > The second bullet doesn't consider the possibility that the
>     'n'th  CE mark might arrive on a pure ACK. Then, the wording as it
>     stands says the Data Receiver MUST immediately ACK a pure ACK. I
>     know TCP never ACKs a pure ACK, but I'm not actually sure it does
>     any harm to do so in this case (it cannot cause an infinite loop
>     of ACKs). However, given it would be unorthodox, we maybe ought to
>     rule it out by rewording anyway?
>
>
> even though it won't cause an infinite loop, can they be dup acks? If 
> so, it doesn't look good to me especially when early retransmit is 
> activated.

[BB] I'm not expert in TCP loss recovery but, surely if TCP A has 
nothing outstanding, it cannot consider receipt of dup acknowledgements 
from TCP B as a sign of loss? Nonetheless, there would be a chance that 
such a dup acknowledgement might be in flight when TCP A happened to 
have just sent some more data, and they crossed in flight.

To avoid this case, it would be possible for TCP B to send pure ACK(s) 
with SEG.SEQ = SND.NXT-1 (using the same trick as keepalive probes). 
Being out of window, each of these pure ACKs would elicit a further pure 
ACK from TCP A. However, that would acknowledge SND-NXT, so the loop 
would stop there. Nonetheless, if TCP B started sending data again, and 
if TCP B's data crossed the pure ACKs that it had recently elicited from 
TCP A, once these pure ACKs arrived at TCP B, I think it would treat 
them as DupACKs.

It's seems possible for TCP B to recognize that it elicited these 
DupACKs from TCP A, and to therefore work out that it could ignore them.

I've kept on digging into this particular obscure corner of ACK 
congestion control, only 'cos I wanted to see if there was a useful end 
to this hole (so it would be worth the AccECN RFC allowing others to dig 
there later, rather than ruling it out).

Nonethless, the AccECN draft itself doesn't have to go anywhere near 
this subject - it can leave all that for a potential future draft about 
using AccECN for ACK congestion control.

>     I think we also can leave this as it is because if that ’n’ packet
>     is a pure ACK that still means that you have unacked data packets
>     with CE marks and you should trigger an ACK for those packet now
>     (rather than waiting for the delayed ACK timer to expire). However
>     this case is less important. Also we should probably make sure
>     that this doesn’t apply if there are only pure ACKs with CE marks,
>     maybe by add “if unacknowledged data are outstanding” or something.
>
> "can send a pure ack on a pure ack when the ack value advances"?

[BB] Yes. That's a fairly safe thing to require.
I think the best strategy here is to require an immediate ACK in 
response to a data packet if there are at least 'n' CE marks to be 
reported. And to extend that requirement to immediately ACKing a pure 
ACK if the ACK value has advanced since the last ACK. But just to say 
nothing about ACKing pure ACKs in other cases, so we don't preclude a 
case that can be made safe in future (e.g. with something like the 
keepalive approach above).

If you (Yoshi), Mirja and others agree, let's try to work out the 
wording of this second bullet.


Bob


> --
> Yoshi

-- 
________________________________________________________________
Bob Briscoe                               http://bobbriscoe.net/