Quality of Service (QoS) has become a bad name in Internet. Two reasons explain this situation. First, after fifteen years of research there is still no QoS in the Internet. Second, some people argue that the Internet does not need any QoS.
We argue in Research: Internet Pricing that there is no QoS mostly because there is no mechanisms to provide the economic incentives. We also argue there that the ISPs need QoS to make money with the Internet.
Independently of economics, we studied QoS in the context of multiplexing and of TCP.
The key idea of packet switching is statistical multiplexing. Statistical multiplexing exploits the fact that different connections are active at slightly different times. Consequently, the maximum rate of the superposition of many connections is significantly less than the sum of the maximum rates of the individual connections. The transmission rate needed to transport a large number of connections is larger than the sum of the average rates of the connections, but less than the sum of the maximum rates. The notion of effective bandwidth is a precise mathematical way to define the rate needed to multiplex connections. We have obtained some key results on effective bandwidth.
In the Internet, about 80% of the traffic corresponds to TCP connections. These connections adjust their rate based on the congestion in the network. We have designed algorithms to adjust the rate of connections so that they share the network resources fairly.
This paper with Jeonghoon Mo explains the improved TCP algorithms. This other paper with Tilo Hamann describes experiments with higher-order window adjustment algorithms. The effective bandwidth ideas were decribed in this paper with Kesidis and Chang.