Core-Stateless Fair Queueing: Achieving Approximately Fair Bandwidth Allocations in High Speed Networks

This paper is the design and implementation of a stateless fair queueing algorithm. The cute trick they used is that they have gateway routers tacking state onto the packets, allowing for stateful decisions without state.

This paper was much clearer than the FQ paper. I found some interesting tidbits in this, particularly in regards to the end-to-end argument. As we move these flow-oriented schemes into the network, I begin to question how much of TCP is being reimplemented in the network. For instance, they had a misbehaving UDP connection that hosed the TCP connections for one experiment. The stateless FQ solved this, giving each equal bandwidth. If you think about it, a good portion of TCP is enforcing that exact behavior. So, why are all of the hosts using TCP? If stateless FQ exists, they can likely use UDP and get the same throughput and congestion avoidance behaviors.

I know there's more to TCP than just that, it just caught my attention.

I found the tagging of packets to be really interesting. That's how Click does a lot of it's work, though they call it "painting" the packet. I question the security. If this is a great function, why not have the ultimate gateways, the hosts, tag their packets? Well, that's presumably a security concern. What would it take for the clients to implement this? Would it simply be a sliding window?

It's late and this isn't well reasoned, we'll see about class tomorrow.

This paper was great and should totally be left in.