Return-Path: <majordom@ISI.EDU>
Received: from EDRC.CMU.EDU by n1.sp.cs.cmu.edu id aa07792; 19 Aug 97 13:57 EDT
Received: from zephyr.isi.edu by EDRC.CMU.EDU id aa24627; 19 Aug 97 13:57:09 EDT
Received: by zephyr.isi.edu (5.65c/5.61+local-26)
	id <AA02088>; Tue, 19 Aug 1997 10:29:28 -0700
Received: from tnt.isi.edu by zephyr.isi.edu (5.65c/5.61+local-26)
	id <AA02081>; Tue, 19 Aug 1997 10:29:25 -0700
Received: from joyride.CS.Berkeley.EDU (joyride.CS.Berkeley.EDU [128.32.33.57])
	by tnt.isi.edu (8.8.7/8.8.6) with SMTP id KAA19710;
	Tue, 19 Aug 1997 10:29:25 -0700 (PDT)
Received: from joyride.CS.Berkeley.EDU (localhost.Berkeley.EDU [127.0.0.1]) by joyride.CS.Berkeley.EDU (8.6.11/8.6.9) with ESMTP id KAA12118; Tue, 19 Aug 1997 10:29:15 -0700
Message-Id: <199708191729.KAA12118@joyride.CS.Berkeley.EDU>
X-Face: 8oz'i+bl`|5PbRnbf:lhb^%e[KkX6s2O+~WXUjjyZy3<eONU1x6ko7NU'ZWaahSA9@z67tI
 imt6ht.__nYj:ufI1]z(DMC4k*hEJO=Y3iihd[[ZDHRV<%<Gl,tJqvQ`2h*[FyU&7F=>Ew*s3R1@]D
 {~a]r4V]),Mlwru>UYa+!f7aeLD3),v{_U3S*(e/Os}3N7*+U+#;5\W0!-U+zs&>c/Gb2FH/|KZ*Li
 eMcCH0X~${-18~JhYDf3Dc}H1,F<V
X-Url: http://www.cs.berkeley.edu/~padmanab
Reply-To: Venkat Padmanabhan <padmanab@cs.berkeley.edu>
From: Venkat Padmanabhan <padmanab@cs.berkeley.edu>
To: Biaz Saad <saadb@cs.tamu.edu>
Cc: Hari Balakrishnan <hari@cs.berkeley.edu>, touch@ISI.EDU, karn@qualcomm.com,
        end2end-interest@ISI.EDU
Subject: Re: HTTP 1.1 and slow-start restart (or lack thereof) 
In-Reply-To: Your message of "Tue, 19 Aug 1997 09:37:03 CDT."
             <Pine.GSO.3.96.970819092053.11035C-100000@mozart> 
Date: Tue, 19 Aug 1997 10:29:08 -0700
Sender: owner-end2end-interest@ISI.EDU
Precedence: bulk


> On Mon, 18 Aug 1997, Venkat Padmanabhan wrote:
> 
> > 
> > 1. Use a short RTO for the fast start packets, basically by using the
> >    srtt and rttvar measured accurately using the timestamp option. 
> ***********
> >    The justification is that the fast start packets are more likely to get
> >    dropped than other packets, so it is okay to pay the cost of a
> >    fine-grained timer for these packets.
> **************
> > 
> 
> 	It is not the first time that I see that *fast start* packets
> would be more likely to get dropped. Are there measurements/analysis 
> showing that ? We would like to know about these studies because our own
> measurements seem to contradict this generally accepted idea.
> 	Measurements can be found at 
> http://www.cs.tamu.edu/people/saadb/report.ps
> 


The reason for this is simply that in my scheme fast start packets
are explicitly assigned a low priority and routers drop low priority
packets first. So fast start packets are more likely to get dropped.

-Venkat