IP Convergence: Beyond VoIP, Beyond Cost Savings

We usually think of QoS as a way of making sure that our applications work the way they are intended.  That is the whole point, right?  Case in point, in my my last post QoS and your PC, we discussed how to implement QoS to provide a better user experience for desktop video conferencing.

But it doesn't always work out that way.

Some years ago now, a company called Caspian Networks created a carrier-class routing platform that analyzed flows in real-time, and it could restrict throughput on that flow based on configuration information.  The impact of this capability is rather interesting...rather than limiting the throughput of a type of application on a general level (a la the Diffserv model), say limiting P2P traffic to 15% of the overall backbone capacity, it can limit the throughput of each application individually, regardless of whether there is congestion in the network or not!

Just imagine the havoc this could cause from a net neutrality perspective.  Are you an RBOC or cable company worried about Skype?  Just limit the throughput on any Skype flow to the point where it'll sound horrible, but not so bad that it won't work at all.  In fact, you can do the same for any application that competes with something you sell yourself.

I first learned of Caspian Networks many years ago through an "economics of the Internet" mailing list hosted by Gordon Cook, and there was a gentleman participating by the name of Dr. Larry Roberts who believed that the economics of the Internet could be radically improved through restricting traffic that was deemed to detract value from the network rather than add to it.  As one of the founders of Caspian, he took this belief and created a product around it.  Caspian folded last year, but Larry started a new company in 2004 called Anagran, and it does pretty much the same thing that Caspian did.  Looking at Anagran's web site, they're clearly focusing on the improved quality that they can give to applications, and yet at the lower right hand corner of the web site, it still mentions "controls P2P and optimizes capacity."

I was reminded of Dr. Roberts and Anagran recently upon reading the following article about Comcast and some technology they have implemented that slows down P2P applications such as BitTorrent, eDonkey and Gnutella, which you can read about in the Associated Press article Comcast blocks Some Internet Traffic.  I don't know if Comcast is using Anagran, but I would guess that they have used something very similar.

QoS is like a gun.  It's not the thing that is evil, it's how it's used.  It does, however, highlight some of the problems with the economic model of the Internet (which was the main topic of the conversation on the previously mentioned mailing list).  As prices on bandwidth have decreased over the years and the basic job of transferring bits has become a commodity, the industry analysts and business strategists have told the carrier to "move up the stack" into applications in order to reach profit margin salvation, and in response carriers have invested what must amount to hundreds of billions of dollars in development only to deploy an application that some else now makes available for free or for profit right on their commodity pipe.  While I believe that the only solution for the carrier to build a better application rather than try to break the competitions, it does pose a risk.  If carriers cannot get the margins they are looking through by selling applications, there will be little choice but to "un-commoditize" basic connectivity and raise prices.  Whether you believe Metcalfe's law or Andrew Odlyzko's more conservative version, maybe the applications are nothing more than fodder for connecting people and are the driver for selling bigger and more profitable pipes just like the latest Microsoft product drives computer upgrades?

Either way, I think the realization of these truths has been a contributor to the stabilization in bandwidth pricing over the past year, and might suggest that this stabilization will at least continue for some time.

Hat tip to Norm for the Comcast article reference.

by Dave Siegel