Paper Review : TCP Vegas :
End to End Congestion Avoidance on a Global Internet
Reviewer : Seh Leng Lim
This paper describes an implementation of TCP, based
on modifications to the Reno implementation of TCP, that is refered to as TCP
Vegas.
The main contribution of the paper is its proposal
for improving TCP, which includes a new timeout mechanism, a novel approach to
congestion avoidance that tries to control the number of extra buffers the
connection occupies in the network, and a modified slow-start mechanism. TCP
Vegas has very promising experimental results from both simulation and the
Internet, whereby TCP Vegas achieves between 37 and 71% better throughput with
one-fifth to one-half losses, as compared to TCP Reno. TCP Vegas is just as
fair as TCP Reno, it does not suffer from stability problems and does not
adversely affect latency.
The key main ideas expounded are:
(a) A new
retransmission mechanism which retransmits relevant segment when the difference
between the current time and the time recorded for sending is greater than the
timeout value
(b) A congestion avoidance
mechanism which limits the number of buffers per connection
(c) A modified slow-start
mechanism which allows exponential growth every other RTT, thereby enabling
detection and avoidance of congestion during slow-start
I think that the paper has a significant
contribution (rating of 4) to the study of congestion avoidance mechanisms in
the computer networks. The paper does have convincing experimental results to
show that TCP Vegas does improve performance over TCP Reno on the Internet. At
the same time, the authors have also attempted to perform simulations to study
issues such as how Vegas connections affect Reno connections, and what happens
when all connections are running over Vegas.So far, the simulation results have
been promising.
The authors mentioned two alternative approaches
such as having an intelligent buffer manager in the router to guarantee
real-time connections as well as selective ACKs to decrease the number of
unnecessarily retransmitted packets. However, the authors have brushed them
both away as having no influence on TCP Vegas. For the first, the authors
opined that real-time connections can request (and pay ) for a minimally
acceptable bandwidth, and then use TCP Vegas to achieve end-to-end congestion
control. For the second, the authors declare that selective acks have very
little effect on TCP Vegas’ retransmissions as there were only 6 KB per MB of
unnecessarily retransmitted packets.
Researchers and builders who build TCP/IP stacks,
routers or even internet applications will have a better appreciations of TCP
Vegas and how it improves network connectivity.