There has been a lot of discussion about "network neutrality" as being a core property of internet service. This discussion includes a number of inter-related concepts, which not every advocate agrees to. In particular, the wikipedia article on the subject identifies:
"* Non-discrimination means that all traffic over the network (typically or exclusively digital packets or bits) is treated the same by the network, including the traffic originating with the network operator. This principle of 'bit parity' means that all bits are treated as 'just bits', and no bit traffic is prioritized over other bits, and none is hampered or disabled.
* Interconnection means that network operators have both a duty of interconnection and a right of interconnection to any other network operator. Networks must be constructed so that there are a reasonable number of accessible interconnect points; that traffic is carried to and from rival networks at reasonable rates; and that the network is built with sufficient excess capacity to accomodate the reasonably foreseeable traffic that may be presented at the head-ends or peering points. Without a right of interconnection, there is no network.
* Access means that any end user can connect to any other end-user. End users may be people, but the term could also mean devices (modems, routers, switches) or even other networks. Access means that a piece of content, say, an email message, has a right to enter the network, and if properly addressed, be received by the other end user, even if said user is on another network. In other words, traffic can begin at any point on the network and be delivered to any other point."
All of these have merits. One could argue that the ideas of access and inter-connection are in a sense constitutional (like Freedom of Speech). However, the idea of non-discrimination can learn from transportation.
"Free" (i.e. untolled) roads in the United States accept any vehicle on a first come-first serve basis, and do not charge any vehicle more for use of the network capacity at a given time (aside from the difference in gas utilization, and some extra taxes from trucks, which are not place or time specific). The consequence is congestion, over-use at peak times at particular places. Users are not paying for the congestion imposed on others. This would be merely wasteful if all users were identical. However users are not identical, some travel is more urgent than other, yet the urgent travel (excepting emergency vehicles) does not get priority.
The high occupancy vehicle lane was the first serious attempt at differentiating travel by value in post-Interstate America. First deployed on Shirley Highway in the Virginia suburbs of Washington, HOV lanes gave priority to high occupancy vehicles (those with multiple passengers) on the theory that cars with 2 or more people had a higher value of time than cars with only 1 person.
However HOV lanes were not terribly efficient, most such lanes had excess capacity. The notion arose in the 1970s (Ward Elliott) to charge non-HOVs for use of that capacity. These High Occupancy Toll lanes are generally parallel to "free" facilities, and allow users to buy their way out of congestion. Users with a high economic value for their trips can now pay to get to their destination faster (just as we can send packages overnight by Federal Express instead of taking a longer time with the post office).
The parallel free roads are no worse for the existance of formerly HOV/now HOT lanes if the alternative were no extra capacity at all. But if that capacity were taken away from the free lanes, some people would pay a premium to jump the queue, while those who don't pay take longer. I discuss this in Chapter 11 of Financing Transportation Networks.
This disruption may be heavily opposed when the packets are people, it depends very much on the assumed base case. However bits don't complain when delayed or routed the long way. So more urgent applications (real-time communication, e.g.) should be able to pay a premium to get a faster (higher quality) service than non-real-time applications. Of course when there are scarce resources, privelaging some traffic (i.e. discriminating for) will discriminate against other traffic. This may be inequitable to some, but economically efficient. We need to balance those two goals.