In my mind, we're really arguing for _very_ intelligent borders that
manage and route traffic to provide enhanced service. Sure, we may need
some additional mechanisms from the interior network to make this work
really well, but everything in our approach seems to be about pushing
decisions to the edges so we can take advantage of our knowledge about
traffic flows.
Finally, I suggest that you change the line:
"we demonstrate that this approach offers the potential for an order "
to something vaguer, less controversial, and something easier to
demonstrate, namely:
"we demonstrate the potential for an order..."
> -----Original Message-----
> From: tom@emigrant [SMTP:tom@emigrant]
> Sent: Thursday, June 25, 1998 4:27 PM
> To: syn@cs
> Subject: comments
>
>
> For Hot Interconnects, I need to send this out tomorrow.
> comments welcome.
>
> tom
> --------
>
> Title
>
> Detour: A Case for A Virtual Internet
>
> Authors
>
> Thomas Anderson, Amit Aggarwal, David Becker, Neal Cardwell, Andrew
> Collins,
> Eric Hoffman, Stefan Savage, John Snell, Amin Vahdat, Geoff Voelker,
> and John Zahorjan
>
> Department of Computer Science and Engineering
> University of Washington
>
> Abstract
>
> Despite its obvious success, robustness, and scalability, the
> Internet suffers from a number of end-to-end performance and
> availability problems, including long and unpredictable latencies
> for short connections, high drop rates, low throughput for long
> transfers
> even when there is ample link bandwidth, poor route selection, and
> periods of
> unavailability despite the existence of redundant paths. These
> problems
> are not easy to fix in the current Internet; for example, total
> throughput
> can get significantly worse as we increase buffering, increase link
> bandwidth, add links, or add load. The need to provide backward
> compatibility makes it nearly impossible to directly address
> these issues by modifying the Internet protocols.
>
> Our approach is to exploit the inefficiencies of the Internet
> by deploying a "virtual Internet" of intelligent routers
> spread at key access and interchange points around the network.
> Routers in the virtual Internet tunnel across the physical Internet
> to communicate with one another, enabling the incremental deployment
> of new routing, congestion control, buffer management, link
> scheduling, and transport protocols. To explore these ideas,
> we are building a prototype virtual Internet, called Detour;
> we demonstrate that this approach offers the potential for an order
> of magnitude improvement in latency, bandwidth, packet loss rates,
> and availability relative to today's physical Internet.