Loss recovery at tail

[hiren panchasara]

On 11/18/16 at 09:20P, Michael Tuexen wrote:
> Dear all,
> 
> in the last telco we discussed the behaviour when that last N packets are dropped.
> It was mentioned that multiple retransmission timers where used to recover.
> 
> I wrote a packetdrill script which
> * Get the cwnd to 30 segments
> * Sends 30 segments, which all get lost
> * Observe how the 30 dropped segments are retransmitted.
> 
> It uses only a single timeout as one would expect.
> So this script does NOT reproduce the problem, but I'm attaching it such that
> you can see how the stack behaves.
> Tested with FreeBSD head r308802.

Thanks for testing.

https://reviews.freebsd.org/D8556 was the issue detecting a valid
RTO as bad and resetting snd_nxt to snd_max (instead of snd_una as done by
valid RTO) causing us to not trigger retransmit of missed packets but
wait for RTO for all of lost packets.

Cheers,
Hiren
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 603 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-transport/attachments/20161118/75731ecd/attachment.sig>