[ NNSquad ] Re: Interview with Bell Canada's Mirko Bibic

[Barry Gold <bgold@matrix-consultants.com>]

Kyle Rosenthal wrote:
> "Internet Congestion a Reality"
>
> Interview with Bell Canada's Mirko Bibic includes his perspective on the 
> recent activity with the CRTC.
>
> http://www.cbc.ca/technology/story/2008/05/30/tech-qandabibic.html

Interesting.  Bibic is right about one thing: you can have congestion 
building up in the links and not see it at the endpoints, up to a 
certain point.  This is pretty elemental queuing theory: you can run a 
network at 90% of capacity and see virtually no retransmission and 
contstant latency.  Somewhere between 90% and 99% you see both measures 
start to climb pretty steeply, until at some point the network becomes 
nearly unusable -- too many packets get dropped, resulting in 
retransmission, which only increases the number of packets entering the 
network.  Sometimes called "chattering," the network equivalent of 
"thrashing" in a Virtual Memory system.

You can see the same thing on a freeway.  If it's up to 80% of capacity, 
 you'll see traffic flowing along at 10mph over the speed limit.  At 
90%, you're moving at the speed limit, with occasional slowdowns for no 
visible reason.  At 95%, it's stop-and-stop.

The problem is, we outsiders can't see what is happening on Bell 
Canada's network.  Bibic may be telling the truth, it's possible that 
Bell's networks are only a few percent below the point where trouble 
starts being visible at the endpoints.  Or he may be lying -- or the 
people who give him information may be giving him only part of the 
picture, which works out to much the same thing in terms of the quality 
of information flowing from him.

One question that was raised in a comment: why is Bell Canada 
overselling their network.  That makes sense for a retail ISP -- you 
give customers the ability to download stuff fast (e.g., 3Mbit/sec), but 
not a guarantee that they will see the maximum speed all the time. 
Viewing something on YouTube at 1AM is a lot different from viewing the 
same clip at 11AM.

But AFAIK the backbone carriers (like BC) sell _guaranteed_ bandwidth to 
their customers.  That means that if you buy 100Mb/sec download and 
20Mb/sec upload, you should be able to receive 100Mb/sec and send 
20Mb/sec total, assuming the endpoints can produce/consume the data that 
fast.  BC should not be selling capacity that they don't have, and if 
90% is the maximum utilization that can be supported without 
"chattering," then they should sell only 90% of their total capacity.

But then, I don't know exactly how BC's "Gateway Access Service" service 
works.  Maybe they sell, say, 50Gb/month of download and 16Gb of upload 
-- meaning you get that much for a fixed price, and if you exceed it you 
pay for the extra usage -- but the instaneous rate isn't guaranteed.  In 
that case, yes, they could have a problem at peak times, and some sort 
of "reasonable" traffic management would be a good thing and should be 
permitted.  So it really depends on how the pricing structure for GAS is 
defined.