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

[Markku Kojo <[e-mail]>]

On Wed, 13 Apr 2011, Yuchung Cheng wrote:

> On Wed, Apr 13, 2011 at 5:17 PM, Ethan Blanton <[e-mail]> wrote:
> > Yuchung Cheng spake unto us the following wisdom:
> >> 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.
> >
> > And this is correct behavior, and what 3517/3517bis will both do.
> >
> Ah that's right. The subtle difference is that Linux IsLost(sn)
> returns true if any 3 segments beyond sn are sacked, i.e., the seq# of
> these segments need not be discontinuous. Therefore the loss marking
> is more aggressive after F-R starts (even with FACK disabled).

Right. The difference is that the Linux TCP sender tracks segment 
boundaries and is therefore able to determine that 3 small segments 
have made it to the receiver even though these segments are continuous.

As Ethan noted the draft is written such that it does not require
a TCP sender to keep track of segment boundaries after a segment
has been transmitted (Note in Section 3).

/Markku