Review - TCP Nice

Review of TCP Nice: A Mechanism for Background Transfers
by Ivan Hernández

The paper presents TCP Nice a self-tuning transport protocol that
provides background transfers with two goals in mind; first, cause no
interference to foreground flows and, second, consume 100% of
available spare bandwidth.

TCP Nice is built on top of TCP Vegas. TCP Nice adds Vegas three new
elements. First, Nice uses a more sensitive network detector, in order
to do this Nice monitors the flows and signals congestion if the queue
size exceeds a fraction of the estimated queue size; if this excess
happens, then the congestion window is reduced by half, otherwise Nice
will fall back on Vegas rules, and if a packet is lost, then will fall
back on Reno rules. Second, Nice uses a multiplicative reduction in
response to increasing round trip time. This two modifications are
aimed to avoid packet loss, which will damage the performance of
foreground flows. And the third element, Nice can reduce the
congestion window below one; in this case, the packets ACKs will be
used to monitor the network waiting for congestion to dissipate.

The authors supports Nice with an analysis of the network protocol in
which the three design features show their role in reducing the
interference to foreground flows. Then, they test Nice in a simulator
with very impressive results; in several of the results Nice just
performs almost as good as a router prioritization. Nice performs good
in the microbenchmark and in the two case study applications.

I found very interesting TCP Nice. It really provides a good tool for
prefetching applications. A strength of Nice is that in order to use
it, it just has to be implemented in the sender and an application
just has to specify with a system call option the kind of TCP
connection.
Received on Wed Oct 04 2006 - 22:47:17 EDT