TCP SACK of FreeBSD stable

[Noritoshi Demizu]

I am observing the TCP SACK behavior of FreeBSD stale.

I found the followings:

  o After the third duplicate ACK, i.e., it enters recovery mode,
    it seems that the outstanding window size may not be halved.
    It sends one data segment for each duplicate ACK received.
    If the number of lost data and ACK segment was small,
    the outstanding window size would not be reduced enough.

  o In recovery mode, it retransmits data when the first SACK block
    above the data is received.  It would be weak in reordering
    of data or ACK segments.

I put the results of my experiments at

  http://www.dd.iij4u.or.jp/~demizu/memo/2005-0316/

Any comments are welcome.

Thank you.

Regards,
Noritoshi Demizu