Re: Broadband initiatives - impact to your network?

["Kevin Oberman" <oberman () es net>]

> Date: Mon, 28 Jun 2010 16:46:37 -0700
> From: "George Bonser" <gbonser () seven com>
>
> > -----Original Message-----
> > From: Jonathan Feldman 
> > Sent: Monday, June 28, 2010 4:14 PM
> > To: Randy Bush
> > Cc: nanog () nanog org
> > Subject: Re: Broadband initiatives - impact to your network?
> >
> > I've never claimed to be particularly bright, but I do like to
> > challenge assumptions.
>
> It isn't only the amount of bandwidth available but also in many cases
> the protocols used to transmit the data.  It takes smarter than the
> average bear to figure out how to get data across a fat pipe over a long
> distance at a high rate.  TCP protocols are limited by the number of
> packets allowed to be "in flight" according to how the stack is
> configured.  One might need to go to unorthodox or rather new methods to
> use all the available bandwidth.
>
> There are many cases of someone being stymied as to why they can't even
> get anywhere near 10 megabits of throughput on a GigE path from Los
> Angeles to London using FTP, for example.  In many cases the
> responsibility of getting data from point A to point B is handled by
> people who don't bring their network operators into the discussion where
> problems like this can be pointed out to them.  Often the first time the
> enterprise network group hears about it is when someone complains that
> the "fast pipe" to $continent is "slow" and therefore must be broken and
> that is generally followed by the demand that it be fixed immediately if
> that demand is not included in the first email. 
>
> That is when conversations bearing sounds like mpscp and uftp begin and
> then someone says "aw, screw it, just send them a disk".

If you really want to improve on the performance of data transfers over
long distances (e.g. across an ocean), take a look at
http://fasterdata.es.net. The Department of Energy and ESnet provides
this information primarily for researchers needing to over large volumes
of data over many thousands of kilometers.

While some of the information will be beyond the capabilities of the
average network user and either end can cause the performance problems,
the information can explain a bit about why the problems exists and does
provide some simple changes that can greatly enhance transfer speed.
-- 
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: oberman () es net                       Phone: +1 510 486-8634
Key fingerprint:059B 2DDF 031C 9BA3 14A4  EADA 927D EBB3 987B 3751