Re: [tcpm] Increasing the Initial Window

Hi Marco,

I'm not so sure I follow the argument with the mobile end devices -
especially with that data from browserscope which has been brought up.

Opera Mini 5.1.21566	  	 11 connections/hostname
Nokia 75	            	 12 connections/hostname
Blackberry 8900	       	 12 connections/hostname
Opera Mini 4.2.15227	  	 21 connections/hostname
Sony Ericsson K800i	 	 24 connections/hostname

More of the outliers seem to be browsers on mobile devices... Even
though the majority of desktop browsers runs at 6 connections/host. (My
inititive guess is, that these browsers don't sort objects by size when
downloading, but prefer to present most of a page at the same time -
with TCP head of line blocking subsequent pipelined objects, multiple
objects would be delayed if only few sessions are used. With only one
object per session, but 20+ in parallel, only the objects of directly
impacted flows are delayed).

I think of IW10 more in terms of minimizing latency for small objects
(regardless of application protocol). After all, with IW10 fast recovery
does have a chance to get triggered, whereas with IW3, only Early
Retransmit and RTO are available but no fast retransmit.

My original concerns about IW10 were based on my (obviously unfounded)
assumption, that IW10 will only show better latency for small objects
when the server is also making use of advanced features (sack, early
retransmit, lost retransmit detection, F-RTO, Eifel...).

But the most recent dataset shows, that this goal - minimizing latency -
can be achived even with standard NewReno, and higher packet loss ratios
(which are expected).

It appears to me, that we are now discussing if the undoubtedly
increased packet loss rate (~ +24%) is more troublesome (to the network)
than the faster delivery of objects (to the clients) (~ -13%). And, as
mentioned above, will the increased aggressiveness of flows (because
fast recovery may be leading to more load early on) hurt the network
beyond that...

And as Mark pointed out, this effects are really only transient, for
bulk transfers the behavior is unchanged. And, senders who are adjacent
to links with very small queues would tend to hurt themselves the most -
potentially causing a change to IW3 locally if it improves their sending
performance.

Does that make sense, or am I off track?

Thanks,

Richard Scheffenegger

> -----Original Message-----
> From: Marco Mellia [mailto:mellia@tlc.polito.it] 
> Sent: Freitag, 12. November 2010 16:14
> To: L.Wood@surrey.ac.uk
> Cc: tcpm@ietf.org; tmrg-interest@icsi.berkeley.edu
> Subject: Re: [tcpm] Increasing the Initial Window - Notes
> 
> 
> Definitively agree with you.
> 
> Eventually, the question is if the IEFT should suggest a 
> switch to a more aggressive setting when non standard 
> solutions are common practice and can possibly interfere 
> (badly) with this change.
> 
> I mean, let's say now all linux/windows/bsd servers start 
> switching to IW 10, so that 99% of server traffic now arrives 
> with a definitively increase burstines. What are the global 
> implication of this switch coupled with today common practice 
> of already tweaking the TCP limits when IW is more 
> conservative? Say today those tweaks are used by 10% of 
> traffic, so they do not interfere too much. What happens 
> tomorrow when all traffic becomes more aggressive?
> 
> My intuition is that the increase the default IW to 10 would 
> benefit in some cases where there is lot of bandwidth, but in 
> some cases it would definitively hurt, especially when 
> thinking about mobile terminals, which btw seems to me being 
> very important in these days.
> 
> This is just my 2c.
> 
> Ciao
> Marco
> 
> Il 12/11/2010 15:05, L.Wood@surrey.ac.uk ha scritto:
> > The point is that implementations don't toe the lne on 
> number of connections per host (6 connections per hostname? 
> Not persistent flows, but actual TCP connections?), and we're 
> not going to move the line to suit them.
> >
> > The same goes for initial window, I think.
> >
> > Lloyd Wood
> > http://tinyurl.com/lloydwood-ccsr
> > ________________________________________
> > From: tcpm-bounces@ietf.org [tcpm-bounces@ietf.org] On 
> Behalf Of Marco 
> > Mellia [mellia@tlc.polito.it]
> > Sent: 12 November 2010 14:03
> > Cc: tcpm@ietf.org; tmrg-interest@icsi.berkeley.edu
> > Subject: Re: [tcpm] Increasing the Initial Window - Notes
> >
> > well...
> > check this then
> > http://www.browserscope.org/
> > in the networking statistics...
> >
> >
> > Il 12/11/2010 10:48, L.Wood@surrey.ac.uk ha scritto:
> >    
> >>> If today a brower opens 10 flows with a server using IW=1 
> you gets 
> >>> 10
> >>>        
> >> packets equivalent IW
> >> And the browser is ignoring RFC 2068 section 8.1:
> >>      Clients that use persistent connections SHOULD limit 
> the number of
> >>      simultaneous connections that they maintain to a 
> given server. A
> >>      single-user client SHOULD maintain AT MOST 2 
> connections with any
> >>      server or proxy.
> >>      
> > --
> >
> > Ciao,                    /\/\/\rco
> >
> > +-----------------------------------+
> > | Marco Mellia - Assistant Professor|
> > | Skypeid: mgmellia                 |
> > | Tel: +39-011-090-4173             |
> > | Cel: +39-331-6714789              |   /"\  .. . . . . . . 
> . . . . . .
> > | Politecnico di Torino             |   \ /  . ASCII Ribbon 
> Campaign  .
> > | Corso Duca degli Abruzzi 24       |    X   .- NO HTML/RTF 
> in e-mail .
> > | Torino - 10129 - Italy            |   / \  .- NO Word 
> docs in e-mail.
> > | http://www.telematica.polito.it   |        .. . . . . . . 
> . . . . . .
> > +-----------------------------------+
> > The box said "Requires Windows 95 or Better." So I installed Linux.
> >
> > _______________________________________________
> > tcpm mailing list
> > tcpm@ietf.org
> > https://www.ietf.org/mailman/listinfo/tcpm
> > _______________________________________________
> > tcpm mailing list
> > tcpm@ietf.org
> > https://www.ietf.org/mailman/listinfo/tcpm
> >
> >    
> 
> -- 
> 
> Ciao,                    /\/\/\rco
> 
> +-----------------------------------+
> | Marco Mellia - Assistant Professor|
> | Skypeid: mgmellia                 |
> | Tel: +39-011-090-4173             |
> | Cel: +39-331-6714789              |   /"\  .. . . . . . . . 
> . . . . .
> | Politecnico di Torino             |   \ /  . ASCII Ribbon 
> Campaign  .
> | Corso Duca degli Abruzzi 24       |    X   .- NO HTML/RTF 
> in e-mail .
> | Torino - 10129 - Italy            |   / \  .- NO Word docs 
> in e-mail.
> | http://www.telematica.polito.it   |        .. . . . . . . . 
> . . . . .
> +-----------------------------------+
> The box said "Requires Windows 95 or Better." So I installed Linux.
> 
> _______________________________________________
> tcpm mailing list
> tcpm@ietf.org
> https://www.ietf.org/mailman/listinfo/tcpm
> 