[Bloat] Fwd: [iccrg] Initial Spreading: a proposal to safely enlarge the TCP's IW
dave.taht at gmail.com
Sat Dec 21 13:03:39 EST 2013
---------- Forwarded message ----------
From: renaud sallantin <renaud.sallantin at gmail.com>
Date: Sat, Dec 21, 2013 at 8:31 AM
Subject: [iccrg] Initial Spreading: a proposal to safely enlarge the TCP's IW
To: iccrg at irtf.org
Cc: cedric baudoin <cedric.baudoin at thalesaleniaspace.com>, Fabrice
Arnal <fabrice.arnal at thalesaleniaspace.com>, Dubois Emmanuel
<emmanuel.dubois at cnes.fr>, Andre-Luc Beylot
<andre-luc.beylot at enseeiht.fr>, Chaput <emmanuel.chaput at enseeiht.fr>
We would like to share and discuss with you a mechanism we called
“Initial Spreading” (IS).
The idea consists in the combination of high IW (e.g. IW=10), jointly
with a spreading of those segments over, at most, the first RTT of the
connection. After the sending of the IW, we switch back to the
scheduling of the TCP algorithms…
Our motivation lays in the fact that raising IW, as largely discussed
previously, is good, unless the network is congested.
In that case, IS prevents the large initial burst to occur, and to
this respect is not too aggressive for the network, but still avoids
the drawbacks of solutions such as Pacing.
We have shown we can achieve a sounding compromise to the enhancement
of the TCP connection startup, in various environments : LAN, wireless
and LFN networks (including satellite access which was our initial
We recently presented this work to the LCN’13 conference in Sydney,
supported by extensive NS2 simulations.
(R. Sallantin, C. Baudoin, E. Chaput, E. Dubois, F. Arnal, and A.
Beylot, “Initial spreading: a fast start-up tcp mechanism,”
proceedings of LCN, October 23, 2013.)
The paper hasn’t been published yet on IEEEXplore but should be
available soon. A draft version of this work is available here :
Finally, we have implemented this scheme in the Linux-kernel.
Preliminary results validate the proposed scheme and our simulation
experiments. We will open the sources and provide related packages by
the end of January.
We are also working on the optimization of the interval between
successive segments during the first RTT, taking into account impact
of bursts and implementation constraints.
We would be more than happy to receive at this stage your feedback.
Do we have to ask for an ID first?
Thales Alenia Space/CNES/TeSA PhD
iccrg mailing list
iccrg at irtf.org
Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html
More information about the Bloat