Review: TCP Nice

Paper: TCP Nice: A Mechanism for Background Transfers

Name: Robert Danek.
Course: CS2209, Fall '06.

    Allowing the ability for applications to do background transfers can
potentially improve an application's performance in a number of
respects, including service latency and scalability. Web prefetching is
provided as an example of such an application.

    The problem is that one needs to balance the usage of network
resources for background transfers with the requirements of foreground
requests. Applications running foreground requests expect those requests
to be handled in a timely manner, whereas background transfers are more
"nice-to-have". To this end, the paper proposes a modification to the
TCP protocol, called TCP Nice, that allows background transfers to occur
without interfering with demand transfers, while at the same time trying
to consume as much available spare bandwidth as it can.

    TCP Nice is an extension of TCP Vegas. It works by being more
sensitive to detecting incipient congestion than either TCP Vegas or TCP
Reno. This means that when other foreground flows are competing for
bandwidth and there's a risk of packet loss due to congestion, TCP Nice
will back off more aggressively than Vegas or Reno. This feature is
supported by multiplicative reduction in the size of of the congestion
window when round trip times increase, and allowing for a minimum
congestion window size less than one.
       
    This was a good paper. The authors recognized that experimental
evidence was insufficient to prove the claims that they were making
about Nice's non-interference properties. As such, they studied their
mechanism from an analytical point of view, in addition to running
experiments.
Received on Wed Oct 04 2006 - 17:41:38 EDT