Summary: TCPVegas

The paper presents TCPVegas - an implementation of TCP that achieves better
throughput and less packet losses as compared to TCPReno.

Three key ideas presented are:

1) New retransmission scheme is designed for more timely decision about
retransmission of a dropped segment. The goal of the retransmit mechanism is
to detect lost packets even though there is no second or third duplicate
ACK. Vegas treats certain ACKs as a hint to check if a timeout should occur.

2) Congestion Avoidance Mechanism: The new congestion scheme described below
allows TCP to anticipate congestion, and adjust it's sending rate
accordingly. The goal of Vegas is to maintain right amount of extra data in
the network. TCPVegas computes difference(Diff) bettween actual throughput
(windowsize/ActualRTT) and base throughput (windowsize/BaseRTT) and makes
decisions every other RTT:
Diff < A, W = W +1, Diff > B, W = W - 1
Default values for A and B are 1 and 3 respectively. A and B indication of
bottleneck queue length.

3) Change to Slow-start algorithm:
Vegas presents a scheme that avoids the initial packet losses in TCPReno
while trying to find the available bandwidth during the initial use of
slow-start. To detect and avoid congestion during slow-start, vegas allows
exponential growth only every other RTT. (In Reno's case, new packet is sent
whenever ACK is received. So, it may end up sending more than one packet
within one RTT). Additionally, available bandwidth estimate is used to set
the threshold window which makes Vegas less likely to overshoot the
available bandwidth.

Throughout the paper, simulations and measurements done over various
internet topologies are used to support the arguements. The later sections
provide the evidence that Vegas is as fair as Reno and it does not suffer
from latency though there is a slight latency(almost negligible) because of
RTT caliculations.

The graphs presented in the paper are not intuitive and would require
detailed explanation. Though paper has lot of good ideas, but it lacks good
presentation of the ideas.

-- 
~Kiran