TCP Vegas: End-to-end Congestion Avoidance on a Global Internet

Motivation: TCP Vegas is an implementation of TCP that achieves better
throughput on the internet and much fewer losses than Reno.

Key Points:

1) The three techniques that Vegas uses to increase throughput and
decrease losses and subsequent transmissions:
a) More timely retransmission - Calculate RTT by comparing send time
with ACK receive time. Use this more accurate RTT estimate for deciding
to retransmit by comparing this RTT estimate with timeout value.

b) Congestion Avoidance - Linearly adjust TCP's congestion window
upwards or downwards based on comparison between the measured throughput
rate with an expected throughput rate.

c) Modified Slow start - The slow-start strategy is modified by reducing
the doubling of the congestion window size to every two RTT.

2) Vegas does add a little overhead of managing the queue for early
segment loss detection and running the congestion avoidance code.

3) Vegas was designed to compete for throughput approximately fairly
with Reno.

4) Vegas backs off when the number of packets enqueued from its flow
increases but it does not necessarily backoff when the number of packets
enqueued by other flow increases.
5) Since Vegas requires changes to the TCP sender only, individuals can
easily deploy it.

6) Although the problems with Vegas have been mentioned in the paper and
the need of Vegas* has been well motivated, the modification has not
been clearly explained. "Vegas* schedules an event at a certain point in
the future" did not sound concrete enough.

 Vegas remains more efficient than Reno for all the workload studied in
the paper. It can be easily deployed. Although, co-existing Reno and
Vegas senders need to be more fair so that unexpected results are not
produced during the heterogeneous deployment.
Received on Wed Oct 04 2006 - 22:36:16 EDT