Summary: TCPNice

In today's world, there is an increasing need for developing efficient
protocols for background transfers especially for data backup and data
replication services. TCPNice develops a novel approach to congestion
control for Internet background transfers that interfers a little with the
foreground transfers.

Most of the previous work deal with the fair allocation of throughput for
all flows but these protocols impact foreground flows significantly as the
number of background flows increases. The primary goal of the TCPNice is to
ensure that there is no interference for foreground flows because of
background flows. The subordinate goal is to utilize most of the available
bandwidth. Since these goals are conflicting, TCPNice trades off spare
bandwidth utilization for no interference with foreground flows.

TCPNice makes three modifications to TCP vegas to achieve its goals.
1) More sensitive Congestion Detector: Signalling congestion when traffic
exceeds a fraction of maximum queue capacity
2) Multiplicative reduction in response to increasing round trip times:
Reduce the window by 2 to increase the agressiveness of reduction for
background flows.
3) By allowing the window sizes to reduce below one, TCP Nice retains the
non-interference property. In this phase, packets act as network probes
waiting for congestion to dissipate.

The paper evaluates TCP nice using extensive simulations and measurements
done over the actual Internet. Especially, I like the idea of case study
applications for evaluating the performance of the TCPNice.

The key strength of the paper is that it provides a end-to-end strategy to
make use of available bandwidth without interfering with foreground flows
thus making application design simpler. Additionally, these advantages
become more relevant as number of end hosts using DSL/Modem increase in the
internet. However, the paper provides preliminary results for deploying
TCPNice in hetrogenous networks. Though the protocol is self-tuning,
threshold values might have to tweaked in hetrogenous networks. Also,
TCPNice doesnt' provide any protection for background flows if there are
misbehaving hosts trying to flood the network. Finally, it doesn't explore
the difficulties involved in deploying both TCPVegas and TCPNice at the same
time.

-- 
~Kiran