Re: [tcpm] Comments on draft-blanton-tcpm-3517bis-01

[Yuchung Cheng <[e-mail]>]

Thanks for the clarification. My example assumes SMSS=1460 (sorry for
not being clear).

AFAIK, Linux (even with fack disabled) triggers the fast-recovery when
any 3 packets (of any size) are sacked beyond SND.UNA.

On Wed, Apr 13, 2011 at 1:13 PM, Ethan Blanton <[e-mail]> wrote:
> Yuchung Cheng spake unto us the following wisdom:
>> I have two questions:
>>
>> 1) IsLost (SeqNum):
>>
>>       This routine returns whether the given sequence number is
>>       considered to be lost.  The routine returns true when either
>>       DupThresh discontiguous SACKed sequences have arrived above
>>       'SeqNum' or more than (DupThresh - 1) * SMSS bytes with sequence
>>       numbers greater than 'SeqNum' have been SACKed.  Otherwise, the
>>       routine returns false.
>>
>> Does RFC require DupThresh discontinuous sack blocks? e.g.,
>> For example, if a sender sends 4 data packets
>> send pkt1: 1-100
>> send pkt2: 101-200
>> send pkt3: 201-300
>> send pkt4: 301-400
>> receive ack: ack1 with one sack block 101-400.
>> does IsLost(1) returns true?
>
> Let me clerify this a bit.  Are you positing this when SMSS >> 100
> bytes?  If so, then the answer is "maybe".  The draft is written in
> the context of a TCP which does not keep track of segments.   If, in
> this case, SMSS >> 100 bytes, then IsLost(1) will return false for a
> pure byte-counting TCP.  Note from Section 3:
>
>   Finally, note that the algorithm in this document assumes a
>   sender that is not keeping track of segment boundaries after
>   transmitting a segment.  It is possible that a sender that did
>   keep this extra state may be able to use a more refined and
>   precise algorithm than the one presented herein, however, we
>   leave this as future work.
>
> If SMSS == 100 bytes, then yes, IsLost(1) will return true.
>
>> 2)
>> If the incoming ACK is a duplicate acknowledgment and the TCP is
>>
>>    not currently in loss recovery, the TCP MUST increase DupAcks by
>>    one and take the following steps:
>>
>>    (1) If DupAcks >= DupThresh, go to step (4).
>>
>>    (2) If DupAcks < DupThresh but IsLost (SND.UNA) returns
>>        true---indicating at least three segments have arrived above
>>        the current cumulative acknowledgment point, which is taken
>>        to indicate loss---go to step (4).
>>
>> can we combine (1) and (2)? go to steps (4) if IsLost(SND.UNA) is
>> true. I guess it depends on your answer of my first question.
>
> No, not unless you assume a TCP which tracks segments, you assume that
> all segments are SMSS-sized, or you *always* use Nagle's algorithm.
> If a TCP is sending segments of size < SMSS, then counting incoming
> DupAcks allows you to trigger loss recovery when 3 incoming ACKs with
> new SACK information arrive, regardless of the size of the segments
> triggering those ACKs.
>
> Hope that helps.
>
> Ethan
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.10 (GNU/Linux)
>
> iQEVAwUBTaYD2/8fixZ3H8crAQgazQf/W0CjlbO5bnIEFJq9syULb/R+bQrF7F+1
> 8g0VtV23PGoifylU+o6k6hxfyL1k4Q4uCpm/DgZEW6aYwo1CBOhTKYhhdAEqDwMK
> m5MRDpt2ZBmULZ8A8NaflrO0mSfLmsRuFcEpXrJm0mQNbIvK65ls2koNcxJ3Zyjl
> 2LsjT5iEzyi7/fBQNPe/AhwInR54ft4kFqZ+rp8AeAffts95xSRm6w4mXQbM18dQ
> urughzVo+dDM+NLO2TxsoeiUomEjRTfRmCfVQ4gx07iMYrLR3aThfniSIMHb0RlB
> 62E/4wLV+mQ/t97D2CgIFkmzp5VAeRCTFnEMg264xXl3hsOGtA+asQ==
> =g9zx
> -----END PGP SIGNATURE-----
>
>