(no subject)

In this paper, the authors show that when they redesign the internet
from scratch, they can significantly increase performance. This is
largely unsurprising.

More specifically, they propose a replacement for TCP, called XCP,
that involves routers transmitting congestion information in message
headers to the end-points. While this may have applications in brand-
new research networks, the odds of such a system ever being put into
use in the internet are very slim, since it requires the replacement
of every router in the network. It also would not have worked for
the design of the original internet, which required interoperability
with existing networks.

I worry as well about complexity at the router level with XCP.
Whereas today's routers route based upon the IP protocol, and do not
require that the router perform any analysis of the TCP or UDP packet
inside, this paper introduces the concept of an XCP router, analysing
both the IP and XCP headers.

Since XCP is presented as a replacement for TCP, I am unsure how it
would handle traffic such as VoIP which relies upon UDP.

In general, this paper annoyed me.
Received on Mon Oct 02 2006 - 22:08:06 EDT