Re: [76attendees] A Net Neutrality comment

[Fred Baker <[e-mail]>]

On Nov 9, 2009, at 9:21 AM, Richard Barnes wrote:

> Thanks for this little study.  Note that this doesn't necessarily  
> argue for *application* throttling, as much as for *user* throttling.

violent agreement, and specifically I would argue *against*  
application throttling. It also is an argument for proper network  
configuration, which may also be relevant here.

I invite attention to ftp://ftpeng.cisco.com/fred/RTT/index.html, and  
more specifically ftp://ftpeng.cisco.com/fred/RTT/Pages/5.html and its  
related study. I was in a hotel in Dublin last January and did a  
similar study. In this case, I measured the behavior of a DSL network  
that I think was configured with effectively infinite buffer and no  
allowance for loss. The spikes are interesting. I don't have hard data  
(the tcpdump trace) to make this assertion, but I suspect I wound up  
with as many as nine copies of the same datagram in queue, driving  
very high congestion when it was unwarranted. Had the network been  
configured with a reasonable buffer depth - 30 ms, 100 ms, whatever -  
and loss after that, goodput would have been improved. The problem  
was, I surmise, that TCP was driving to the cliff, and THERE WAS NO  
CLIFF.