What is Internet fairness?
In discussions about P2P (peer-to-peer) traffic, there are widely divergent views on what people's inherent rights are when it comes to sharing Internet capacity.
In designing TCP/IP and the Internet over 40 years ago, no rule was made as to how capacity was distributed at each choke point, but the result was "equal capacity (rate) per flow." This most likely reflected our concept of voice networks. Also, in the early years there really was typically just one flow in each direction per person (used for file transfers, voice calls, typing streams, etc.).
Now, however, computers can divide any single flow, such as a file transfer, into many flows with ease. So if we stick with "equal capacity per flow," the only way for each user to get his fair share is to use more and more flows per application.
Outconsuming your neighbor to get what you need will only lead to an arms race that will cause performance problems for a large part of the population.
For example, because of the inherent properties of P2P, we now see a small fraction (under 5 percent) of users consuming 80 to 95 percent of all network capacity with P2P applications. The rest of us are slowed to a crawl, since our one flow gets the same capacity as each of the many flows allotted to each P2P application.
And P2P applications are almost insatiable. If more capacity is added, P2P takes an even higher fraction of the bandwidth. Currently, there is enough video out there to give P2P applications a virtually unlimited appetite.
Economically, P2P is not egalitarian. Where people are charged a flat rate for Internet access, those not using P2P may be getting far less capacity, and far poorer performance, for the same price as P2P users. This is not an ideal situation for the users, or for service providers or organizations with networks that are compromised by P2P.
Thus, it is time to change the network design to one that inherently supports the concept of "equal capacity for equal payment." The current flow equality is a joint result of the design of TCP and the network. Changing TCP is unworkable and unenforceable. Changing the network turns out to be quite economical and straightforward.
The change only needs to be implemented at the network edge, where each subscriber's traffic can be seen. Network equipment is always controlling the rate of every TCP flow -- today, by making all flows equal in rate. Alternatively, when network equipment measures the traffic per subscriber, this information can be used to control each subscriber's flow rates separately, such that in any short period they all receive equitable capacity for what they pay.
[Ed. note: Dr. Lawrence Roberts is the founder and CEO of Anagran Inc. , a flow-based routing company.]
With network equipment that has been designed to manage overload through flow rate control rather than random packet loss, managing subscriber traffic and honoring SLAs based on subscribed usage rates becomes straightforward.
P2P traffic is just the first of a wave of applications that get more capacity by using multiple flows. We need to change the old concept of "equal capacity per flow" to "equal capacity for equal pay" -- or the arms race for more and more flows will hurt us all.
— Dr. Lawrence G. Roberts, CEO, Anagran Inc.