Very interesting idea. Maybe one solution would be to encourage convergence of application/content providers and infrastructure companies (Time Warner). Wldn't it be funny if BGP becomes Bill Gates Protocol;)
This reminds me of a Wired article arnd 1996 http://www.wired.com/wired/archive/4.10/atm_pr.htm l
Overprovisioning may be a simpler solution but the key factor is the burstiness of traffic. Most of today's traffic is P2P with high variability. And then there is the "Groves Giveth and Gates Taketh Away" phenomenon. Any extra bandwidth u throw into wld be consumed by ever demanding applications. 2 yrs ago ppl were downloading mp3s and now they r into DivX movies. I think much of the confusion on this QoS vs Overprovisioning debate lies in the temporality of congestion. Internet experiences congestion usually for millisecond durations when say all the traffic flows to a slashdotted site. This does not mean that the avg demand is higher than the available capacity. Such short term congestion cannot be combated by overprovisioning. At present we rely on TCP to do the dirty work but no one has much of a clue if UDP flows come into the picture. Luckily over 95% of net traffic is TCP so we r kinda saved for the time being!. One nice thing with pricing is it gives the users an incentive to send the most important data when the network is congested. It also acts as a feedback to the network to decide whether and how to avoid congestion. If congestion persists for the long term it is an indication of supply-demand mismatch which can be alleviated by overprovisioning. The revenues generated by congestion pricing can be used for this purpose. So QoS and overprovisioning are in a way complementary. Usage pricing has however associated problems of system complexity, accounting and revenue division as well as lack of standards. And like Odlyzko always says ppl r always risk averse and hate change.
Slashdot Mirror
Very interesting idea. Maybe one solution would be to encourage convergence of application/content providers and infrastructure companies (Time Warner). Wldn't it be funny if BGP becomes Bill Gates Protocol ;)
This reminds me of a Wired article arnd 1996 http://www.wired.com/wired/archive/4.10/atm_pr.htm l
Overprovisioning may be a simpler solution but the key factor is the burstiness of traffic. Most of today's traffic is P2P with high variability. And then there is the "Groves Giveth and Gates Taketh Away" phenomenon. Any extra bandwidth u throw into wld be consumed by ever demanding applications. 2 yrs ago ppl were downloading mp3s and now they r into DivX movies. I think much of the confusion on this QoS vs Overprovisioning debate lies in the temporality of congestion. Internet experiences congestion usually for millisecond durations when say all the traffic flows to a slashdotted site. This does not mean that the avg demand is higher than the available capacity. Such short term congestion cannot be combated by overprovisioning. At present we rely on TCP to do the dirty work but no one has much of a clue if UDP flows come into the picture. Luckily over 95% of net traffic is TCP so we r kinda saved for the time being!. One nice thing with pricing is it gives the users an incentive to send the most important data when the network is congested. It also acts as a feedback to the network to decide whether and how to avoid congestion. If congestion persists for the long term it is an indication of supply-demand mismatch which can be alleviated by overprovisioning. The revenues generated by congestion pricing can be used for this purpose. So QoS and overprovisioning are in a way complementary. Usage pricing has however associated problems of system complexity, accounting and revenue division as well as lack of standards. And like Odlyzko always says ppl r always risk averse and hate change.