RE: Shaw, really a free speedup?

ACKs are certainly affected with asymmetric networks but tweaking
filtering rules/queuing or adding something like RED aren't going to
give you a 40% download boost unless the upload is really punished.
I've played with dummynet and altq with boosts around 5-10% -- and
that's with a lot of playing around with queue size, faq reading, and
midnight incantations. More memory available to routers or smarter
handling can help maintain 1000's of connections, but I can't see Shaw
spending money on infrastructure for the greatest benefit of torrents.

One way to settle this.... is there anyway to download a ton of data via
UDP???

Chris Twa

Saskaweb IT Solutions
ph: 306-229-5742
e: support@saskaweb.com

-----Original Message-----
From: owner-linux@slg.org [mailto:owner-linux@slg.org] On Behalf Of Dave
Hall
Sent: 2008-12-16 09:51
To: linux@slg.org
Subject: Re: Shaw, really a free speedup?

On Tue, Dec 16, 2008 at 08:23:07AM -0600, Les Klassen Hamm wrote:
> ADMIN@WHATEVERCOMPUTES.COM wrote:
> >I had Soho installed last Thursday and the tech that did the install
> >was telling me that what they did was add a router in every local
area network.
> >By doing that they would be causing half the traffic or on each
> >router or "twice the speed". He about laughed as well so I'm thinking

> >it's a marketing approach to speed....
>
> I'd be ready to agree whole-heartedly, but the download speed of large

> files has clearly changed - ie my knoppix download. If that's done by
> trickery, it's worthwhile trickery, because it makes a real difference

> in that one area. I don't suspect my torrents are much different
though.

Here's my hypothesis based on what I know about TCP/IP and statistical
multiplexing of asymmetric data connections:

Possibility 1) Shaw simply changed the configured customer "download"
bandwidth limit somewhere in the network. Given their arbitrary claims
which would have to pass some sort of advertising standard, this would
be the most likely. An upgrade on the "upload" capacity on the cable
network may have been included and probably would be necessary to
sustain improved TCP downloads during peak demand.

Possibility 2) Shaw has implemented some sort of "deep packet
inspection"
or traffic shaping to better manage uploads. Peer to peer file sharing
is really detrimental to TCP downloads on an asymmetric network with
limited upload bandwidth. Basically it causes upstream congestion which
slows down TCP acknowledgements, a feedback mechanism TCP uses to ensure
a reliable data connection and manage speed.

It would be possible to devise and experiment playing with TCP
congestion control algorightms to determine if Shaw simply increased
bandwidth ceilings or if they are tinkering with upload traffic. I'm
not on Shaw so unfortunately I can't test it.

If anyone is interested, traditional congestion control algorithms are
explained in RFC 2581: http://www.ietf.org/rfc/rfc2581.txt

I've read that Microsoft came up with some sort of hybrid technique in
Vista. I honestly haven't had time to read about it. Here is a paper
that compares standard TCP to a couple of advanced algorithms:
http://www.slac.stanford.edu/cgi-wrap/getdoc/slac-tn-06-005.pdf

It might be intersting to see how much bandwidth you can coax out of
Shaw by merely playing with TCP parameters. I suspect the 1200kBps,
just under 10Mbps would be the theoretical maximum throughput. Here's
an excellent place to start to tune TCP on Linux:
http://www-didc.lbl.gov/TCP-tuning/linux.html

-- 
Dave