[Tmrg] Proposal to increase TCP initial CWND

milenkoski at uacs.edu.mk (Aleksandar Milenkoski) Sun, 18 July 2010 17:55 UTC

From: milenkoski at uacs.edu.mk (Aleksandar Milenkoski)
Date: Sun, 18 Jul 2010 19:55:31 +0200
Subject: [Tmrg] Proposal to increase TCP initial CWND
References: <AANLkTil937lyUzRvUtdqd2qdl9RN7AZ-Mo_cT-dtmqXz@mail.gmail.com>
Message-ID: <8CC704F4BC13FE40BC2209B42A65C8729990B0@Mail.uacs.edu.mk>

Well, the factor of increasing loss rate of "high-cwnd" single flow and in the same time suppressing the competing traffic for sqrt(increase_to_percent(in this case 5.79%)/increase_from_percent(in this case 3.94%)) will always be linearly dependent on the variable increase_to_percent. Anyway, if we introduce multiple flows with the increase start cwnd size, then they will become competing traffic for themselves, aside the flows which have low cwnd. Therefore, the mathematical model could become pretty complex because of the complex situation of multiple dependencies between the influences of each "high-cwnd" flow to all other flows. Maybe the "cheaper" solution is to get link characteristics from the actual data center, and simulate them to obtain some results.

Milenkoski A.

-----Original Message-----
From: tmrg-interest-bounces at ICSI.Berkeley.EDU on behalf of Lachlan Andrew
Sent: Sun 7/18/2010 8:27 AM
To: tmrg; Nandita Dukkipati; Matt Mathis; Jerry Chu; Yuchung Cheng
Subject: [Tmrg] Proposal to increase TCP initial CWND
Greetings TMRG folk,

At the last IETF meeting, some folk from Google proposed increasing
the initial TCP CWND from the current value (2-4 depending on MSS) to
10.  The current draft is at

The discussion at the ICCRG meeting raised some modelling/evaluation
issues which fit well with the expertise at TMRG.  Since TMRG is
chartered as a mailing-list-only RG, let's discuss this now in lieu of
a meeting at IETF 78 in Maastricht.

The basic TMRG-related issue was:  If a data centre changes a TCP
parameter for *its* flows, how can it estimate the degradation caused
to *other* flows sharing resources with these flows?  Google is awash
with data, but it all shares the bias of coming from their data
centres.  What models are appropriate to infer the impact on the
unmeasured flows?

Sections 10.2 and 11 are the most relevant to TMRG.

To start things off, here is one observation:
- An increase in retransmissions from 3.94% to 5.79% seems very
significant, if that rate reflects the loss rate.  Although described
as a 1.85% increase, it seems likely to cause competing traffic to
slow down by a factor of  sqrt(5.79/3.94).  Moreover, that is the
increase in loss when (presumably) a *single* flow on that bottleneck
increases its initial window.  How can we estimate the impact of *all*
flows increasing their initial windows?

I hope this leads to a lively discussion.


Lachlan Andrew? Centre for Advanced Internet Architectures (CAIA)
Swinburne University of Technology, Melbourne, Australia
Ph +61 3 9214 4837

Tmrg-interest mailing list
Tmrg-interest at ICSI.Berkeley.EDU

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ICSI.Berkeley.EDU/pipermail/tmrg-interest/attachments/20100718/4bd2e3f0/attachment.html